Minor addition

for more information, see https://pre-commit.ci

.cache/calibration/aloha_default/left_follower.json

@@ -1,68 +0,0 @@
-{
-    "homing_offset": [
-        2048,
-        3072,
-        3072,
-        -1024,
-        -1024,
-        2048,
-        -2048,
-        2048,
-        -2048
-    ],
-    "drive_mode": [
-        1,
-        1,
-        1,
-        0,
-        0,
-        1,
-        0,
-        1,
-        0
-    ],
-    "start_pos": [
-        2015,
-        3058,
-        3061,
-        1071,
-        1071,
-        2035,
-        2152,
-        2029,
-        2499
-    ],
-    "end_pos": [
-        -1008,
-        -1963,
-        -1966,
-        2141,
-        2143,
-        -971,
-        3043,
-        -1077,
-        3144
-    ],
-    "calib_mode": [
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "LINEAR"
-    ],
-    "motor_names": [
-        "waist",
-        "shoulder",
-        "shoulder_shadow",
-        "elbow",
-        "elbow_shadow",
-        "forearm_roll",
-        "wrist_angle",
-        "wrist_rotate",
-        "gripper"
-    ]
-}

.cache/calibration/aloha_default/left_leader.json

@@ -1,68 +0,0 @@
-{
-    "homing_offset": [
-        2048,
-        3072,
-        3072,
-        -1024,
-        -1024,
-        2048,
-        -2048,
-        2048,
-        -1024
-    ],
-    "drive_mode": [
-        1,
-        1,
-        1,
-        0,
-        0,
-        1,
-        0,
-        1,
-        0
-    ],
-    "start_pos": [
-        2035,
-        3024,
-        3019,
-        979,
-        981,
-        1982,
-        2166,
-        2124,
-        1968
-    ],
-    "end_pos": [
-        -990,
-        -2017,
-        -2015,
-        2078,
-        2076,
-        -1030,
-        3117,
-        -1016,
-        2556
-    ],
-    "calib_mode": [
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "LINEAR"
-    ],
-    "motor_names": [
-        "waist",
-        "shoulder",
-        "shoulder_shadow",
-        "elbow",
-        "elbow_shadow",
-        "forearm_roll",
-        "wrist_angle",
-        "wrist_rotate",
-        "gripper"
-    ]
-}

.cache/calibration/aloha_default/right_follower.json

@@ -1,68 +0,0 @@
-{
-    "homing_offset": [
-        2048,
-        3072,
-        3072,
-        -1024,
-        -1024,
-        2048,
-        -2048,
-        2048,
-        -2048
-    ],
-    "drive_mode": [
-        1,
-        1,
-        1,
-        0,
-        0,
-        1,
-        0,
-        1,
-        0
-    ],
-    "start_pos": [
-        2056,
-        2895,
-        2896,
-        1191,
-        1190,
-        2018,
-        2051,
-        2056,
-        2509
-    ],
-    "end_pos": [
-        -1040,
-        -2004,
-        -2006,
-        2126,
-        2127,
-        -1010,
-        3050,
-        -1117,
-        3143
-    ],
-    "calib_mode": [
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "LINEAR"
-    ],
-    "motor_names": [
-        "waist",
-        "shoulder",
-        "shoulder_shadow",
-        "elbow",
-        "elbow_shadow",
-        "forearm_roll",
-        "wrist_angle",
-        "wrist_rotate",
-        "gripper"
-    ]
-}

.cache/calibration/aloha_default/right_leader.json

@@ -1,68 +0,0 @@
-{
-    "homing_offset": [
-        2048,
-        3072,
-        3072,
-        -1024,
-        -1024,
-        2048,
-        -2048,
-        2048,
-        -2048
-    ],
-    "drive_mode": [
-        1,
-        1,
-        1,
-        0,
-        0,
-        1,
-        0,
-        1,
-        0
-    ],
-    "start_pos": [
-        2068,
-        3034,
-        3030,
-        1038,
-        1041,
-        1991,
-        1948,
-        2090,
-        1985
-    ],
-    "end_pos": [
-        -1025,
-        -2014,
-        -2015,
-        2058,
-        2060,
-        -955,
-        3091,
-        -940,
-        2576
-    ],
-    "calib_mode": [
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "DEGREE",
-        "LINEAR"
-    ],
-    "motor_names": [
-        "waist",
-        "shoulder",
-        "shoulder_shadow",
-        "elbow",
-        "elbow_shadow",
-        "forearm_roll",
-        "wrist_angle",
-        "wrist_rotate",
-        "gripper"
-    ]
-}

.gitattributes

@@ -11,10 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 *.memmap filter=lfs diff=lfs merge=lfs -text
 *.stl filter=lfs diff=lfs merge=lfs -text
 *.safetensors filter=lfs diff=lfs merge=lfs -text
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.arrow filter=lfs diff=lfs merge=lfs -text
 *.json !text !filter !merge !diff
+tests/artifacts/cameras/*.png filter=lfs diff=lfs merge=lfs -text
+*.bag filter=lfs diff=lfs merge=lfs -text

.github/workflows/build_documentation.yml

@@ -0,0 +1,23 @@
+name: Build documentation
+
+on:
+  workflow_dispatch:
+  push:
+    paths:
+      - "docs/**"
+    branches:
+    - main
+    - doc-builder*
+    - v*-release
+
+
+jobs:
+  build:  # zizmor: ignore[excessive-permissions] We follow the same pattern as in Transformers
+    uses: huggingface/doc-builder/.github/workflows/build_main_documentation.yml@main
+    with:
+      commit_sha: ${{ github.sha }}
+      package: lerobot
+      additional_args: --not_python_module
+    secrets:
+      token: ${{ secrets.HUGGINGFACE_PUSH }}
+      hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}

.github/workflows/build_pr_documentation.yml

@@ -0,0 +1,19 @@
+name: Build PR Documentation
+
+on:
+  pull_request:
+    paths:
+      - "docs/**"
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  build:  # zizmor: ignore[excessive-permissions] We follow the same pattern as in Transformers
+    uses: huggingface/doc-builder/.github/workflows/build_pr_documentation.yml@main
+    with:
+      commit_sha: ${{ github.event.pull_request.head.sha }}
+      pr_number: ${{ github.event.number }}
+      package: lerobot
+      additional_args: --not_python_module

.github/workflows/test-docker-build.yml

@@ -41,7 +41,7 @@ jobs:
 
       - name: Get changed files
         id: changed-files
-        uses: tj-actions/changed-files@v44
+        uses: tj-actions/changed-files@3f54ebb830831fc121d3263c1857cfbdc310cdb9 #v42
         with:
           files: docker/**
           json: "true"

.github/workflows/test.yml

@@ -126,7 +126,7 @@ jobs:
       # portaudio19-dev is needed to install pyaudio
         run: |
           sudo apt-get update && \
-          sudo apt-get install -y libegl1-mesa-dev portaudio19-dev
+          sudo apt-get install -y libegl1-mesa-dev ffmpeg portaudio19-dev
 
       - name: Install uv and python
         uses: astral-sh/setup-uv@v5

.github/workflows/upload_pr_documentation.yml

@@ -0,0 +1,16 @@
+name: Upload PR Documentation
+
+on: # zizmor: ignore[dangerous-triggers] We follow the same pattern as in Transformers
+  workflow_run:
+    workflows: [ "Build PR Documentation" ]
+    types:
+    - completed
+
+jobs:
+  build:  # zizmor: ignore[excessive-permissions] We follow the same pattern as in Transformers
+    uses: huggingface/doc-builder/.github/workflows/upload_pr_documentation.yml@main
+    with:
+      package_name: lerobot
+    secrets:
+      hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}
+      comment_bot_token: ${{ secrets.COMMENT_BOT_TOKEN }}

.gitignore

@@ -12,6 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# Dev scripts
+.dev
+
 # Logging
 logs
 tmp
@@ -91,10 +94,8 @@ coverage.xml
 .hypothesis/
 .pytest_cache/
 
-# Ignore .cache except calibration
+# Ignore .cache
 .cache/*
-!.cache/calibration/
-!.cache/calibration/**
 
 # Translations
 *.mo

.pre-commit-config.yaml

@@ -36,8 +36,8 @@ repos:
       - id: end-of-file-fixer
       - id: trailing-whitespace
 
-  - repo: https://github.com/crate-ci/typos
-    rev: v1.30.2
+  - repo: https://github.com/adhtruong/mirrors-typos
+    rev: v1.31.1
     hooks:
       - id: typos
         args: [--force-exclude]
@@ -48,7 +48,7 @@ repos:
     -   id: pyupgrade
 
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.9.10
+    rev: v0.11.5
     hooks:
       - id: ruff
         args: [--fix]
@@ -57,12 +57,12 @@ repos:
 
   ##### Security #####
   - repo: https://github.com/gitleaks/gitleaks
-    rev: v8.24.0
+    rev: v8.24.3
     hooks:
       - id: gitleaks
 
   - repo: https://github.com/woodruffw/zizmor-pre-commit
-    rev: v1.4.1
+    rev: v1.5.2
     hooks:
       - id: zizmor
 

CONTRIBUTING.md

@@ -269,9 +269,6 @@ Follow these steps to start contributing:
    the PR as a draft PR. These are useful to avoid duplicated work, and to differentiate
    it from PRs ready to be merged;
 4. Make sure existing tests pass;
-<!-- 5. Add high-coverage tests. No quality testing = no merge.
-
-See an example of a good PR here: https://github.com/huggingface/lerobot/pull/ -->
 
 ### Tests
 

README.md

@@ -23,21 +23,35 @@
 </div>
 
 <h2 align="center">
-    <p><a href="https://github.com/huggingface/lerobot/blob/main/examples/10_use_so100.md">
-        Build Your Own SO-100 Robot!</a></p>
+    <p><a href="https://github.com/huggingface/lerobot/blob/main/examples/12_use_so101.md">
+        Build Your Own SO-101 Robot!</a></p>
 </h2>
 
 <div align="center">
-  <img src="media/so100/leader_follower.webp?raw=true" alt="SO-100 leader and follower arms" title="SO-100 leader and follower arms" width="50%">
+  <div style="display: flex; gap: 1rem; justify-content: center; align-items: center;" >
+    <img
+      src="media/so101/so101.webp?raw=true"
+      alt="SO-101 follower arm"
+      title="SO-101 follower arm"
+      style="width: 40%;"
+    />
+    <img
+      src="media/so101/so101-leader.webp?raw=true"
+      alt="SO-101 leader arm"
+      title="SO-101 leader arm"
+      style="width: 40%;"
+    />
+  </div>
 
-  <p><strong>Meet the SO-100 – Just $110 per arm!</strong></p>
+
+  <p><strong>Meet the updated SO100, the SO-101 – Just €114 per arm!</strong></p>
   <p>Train it in minutes with a few simple moves on your laptop.</p>
   <p>Then sit back and watch your creation act autonomously! 🤯</p>
 
-  <p><a href="https://github.com/huggingface/lerobot/blob/main/examples/10_use_so100.md">
-      Get the full SO-100 tutorial here.</a></p>
+  <p><a href="https://github.com/huggingface/lerobot/blob/main/examples/12_use_so101.md">
+      See the full SO-101 tutorial here.</a></p>
 
-  <p>Want to take it to the next level? Make your SO-100 mobile by building LeKiwi!</p>
+  <p>Want to take it to the next level? Make your SO-101 mobile by building LeKiwi!</p>
   <p>Check out the <a href="https://github.com/huggingface/lerobot/blob/main/examples/11_use_lekiwi.md">LeKiwi tutorial</a> and bring your robot to life on wheels.</p>
 
   <img src="media/lekiwi/kiwi.webp?raw=true" alt="LeKiwi mobile robot" title="LeKiwi mobile robot" width="50%">
@@ -51,7 +65,6 @@
 
 ---
 
-
 🤗 LeRobot aims to provide models, datasets, and tools for real-world robotics in PyTorch. The goal is to lower the barrier to entry to robotics so that everyone can contribute and benefit from sharing datasets and pretrained models.
 
 🤗 LeRobot contains state-of-the-art approaches that have been shown to transfer to the real-world with a focus on imitation learning and reinforcement learning.
@@ -98,14 +111,25 @@ conda create -y -n lerobot python=3.10
 conda activate lerobot
 ```
 
+When using `miniconda`, install `ffmpeg` in your environment:
+```bash
+conda install ffmpeg -c conda-forge
+```
+
+> **NOTE:** This usually installs `ffmpeg 7.X` for your platform compiled with the `libsvtav1` encoder. If `libsvtav1` is not supported (check supported encoders with `ffmpeg -encoders`), you can:
+>  - _[On any platform]_ Explicitly install `ffmpeg 7.X` using:
+>  ```bash
+>  conda install ffmpeg=7.1.1 -c conda-forge
+>  ```
+>  - _[On Linux only]_ Install [ffmpeg build dependencies](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#GettheDependencies) and [compile ffmpeg from source with libsvtav1](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#libsvtav1), and make sure you use the corresponding ffmpeg binary to your install with `which ffmpeg`.
+
 Install 🤗 LeRobot:
 ```bash
 pip install -e .
 ```
 
-> **NOTE:** Depending on your platform, If you encounter any build errors during this step
-you may need to install `cmake` and `build-essential` for building some of our dependencies.
-On linux: `sudo apt-get install cmake build-essential`
+> **NOTE:** If you encounter build errors, you may need to install additional dependencies (`cmake`, `build-essential`, and `ffmpeg libs`). On Linux, run:
+`sudo apt-get install cmake build-essential python3-dev pkg-config libavformat-dev libavcodec-dev libavdevice-dev libavutil-dev libswscale-dev libswresample-dev libavfilter-dev pkg-config`. For other systems, see: [Compiling PyAV](https://pyav.org/docs/develop/overview/installation.html#bring-your-own-ffmpeg)
 
 For simulations, 🤗 LeRobot comes with gymnasium environments that can be installed as extras:
 - [aloha](https://github.com/huggingface/gym-aloha)
@@ -197,7 +221,7 @@ dataset attributes:
   │  ├ episode_index (int64): index of the episode for this sample
   │  ├ frame_index (int64): index of the frame for this sample in the episode ; starts at 0 for each episode
   │  ├ timestamp (float32): timestamp in the episode
-  │  ├ next.done (bool): indicates the end of en episode ; True for the last frame in each episode
+  │  ├ next.done (bool): indicates the end of an episode ; True for the last frame in each episode
   │  └ index (int64): general index in the whole dataset
   ├ episode_data_index: contains 2 tensors with the start and end indices of each episode
   │  ├ from (1D int64 tensor): first frame index for each episode — shape (num episodes,) starts with 0
@@ -246,7 +270,7 @@ See `python lerobot/scripts/eval.py --help` for more instructions.
 
 ### Train your own policy
 
-Check out [example 3](./examples/3_train_policy.py) that illustrate how to train a model using our core library in python, and [example 4](./examples/4_train_policy_with_script.md) that shows how to use our training script from command line.
+Check out [example 3](./examples/3_train_policy.py) that illustrates how to train a model using our core library in python, and [example 4](./examples/4_train_policy_with_script.md) that shows how to use our training script from command line.
 
 To use wandb for logging training and evaluation curves, make sure you've run `wandb login` as a one-time setup step. Then, when running the training command above, enable WandB in the configuration by adding `--wandb.enable=true`.
 
@@ -297,7 +321,7 @@ Once you have trained a policy you may upload it to the Hugging Face hub using a
 You first need to find the checkpoint folder located inside your experiment directory (e.g. `outputs/train/2024-05-05/20-21-12_aloha_act_default/checkpoints/002500`). Within that there is a `pretrained_model` directory which should contain:
 - `config.json`: A serialized version of the policy configuration (following the policy's dataclass config).
 - `model.safetensors`: A set of `torch.nn.Module` parameters, saved in [Hugging Face Safetensors](https://huggingface.co/docs/safetensors/index) format.
-- `train_config.json`: A consolidated configuration containing all parameter userd for training. The policy configuration should match `config.json` exactly. Thisis useful for anyone who wants to evaluate your policy or for reproducibility.
+- `train_config.json`: A consolidated configuration containing all parameters used for training. The policy configuration should match `config.json` exactly. This is useful for anyone who wants to evaluate your policy or for reproducibility.
 
 To upload these to the hub, run the following:
 ```bash

benchmarks/video/README.md

@@ -51,7 +51,7 @@ For a comprehensive list and documentation of these parameters, see the ffmpeg d
 ### Decoding parameters
 **Decoder**
 We tested two video decoding backends from torchvision:
-- `pyav` (default)
+- `pyav`
 - `video_reader` (requires to build torchvision from source)
 
 **Requested timestamps**

benchmarks/video/capture_camera_feed.py

@@ -17,12 +17,21 @@
 
 import argparse
 import datetime as dt
+import os
+import time
 from pathlib import Path
 
 import cv2
+import rerun as rr
+
+# see https://rerun.io/docs/howto/visualization/limit-ram
+RERUN_MEMORY_LIMIT = os.getenv("LEROBOT_RERUN_MEMORY_LIMIT", "5%")
 
 
-def display_and_save_video_stream(output_dir: Path, fps: int, width: int, height: int):
+def display_and_save_video_stream(output_dir: Path, fps: int, width: int, height: int, duration: int):
+    rr.init("lerobot_capture_camera_feed")
+    rr.spawn(memory_limit=RERUN_MEMORY_LIMIT)
+
     now = dt.datetime.now()
     capture_dir = output_dir / f"{now:%Y-%m-%d}" / f"{now:%H-%M-%S}"
     if not capture_dir.exists():
@@ -39,24 +48,21 @@ def display_and_save_video_stream(output_dir: Path, fps: int, width: int, height
     cap.set(cv2.CAP_PROP_FRAME_HEIGHT, height)
 
     frame_index = 0
-    while True:
+    start_time = time.time()
+    while time.time() - start_time < duration:
         ret, frame = cap.read()
 
         if not ret:
             print("Error: Could not read frame.")
             break
-
-        cv2.imshow("Video Stream", frame)
+        rr.log("video/stream", rr.Image(frame.numpy()), static=True)
         cv2.imwrite(str(capture_dir / f"frame_{frame_index:06d}.png"), frame)
         frame_index += 1
 
-        # Break the loop on 'q' key press
-        if cv2.waitKey(1) & 0xFF == ord("q"):
-            break
-
-    # Release the capture and destroy all windows
+    # Release the capture
     cap.release()
-    cv2.destroyAllWindows()
+
+    # TODO(Steven): Add a graceful shutdown via a close() method for the Viewer context, though not currently supported in the Rerun API.
 
 
 if __name__ == "__main__":
@@ -86,5 +92,11 @@ if __name__ == "__main__":
         default=720,
         help="Height of the captured images.",
     )
+    parser.add_argument(
+        "--duration",
+        type=int,
+        default=20,
+        help="Duration in seconds for which the video stream should be captured.",
+    )
     args = parser.parse_args()
     display_and_save_video_stream(**vars(args))

benchmarks/video/run_video_benchmark.py

@@ -67,7 +67,7 @@ def parse_int_or_none(value) -> int | None:
 def check_datasets_formats(repo_ids: list) -> None:
     for repo_id in repo_ids:
         dataset = LeRobotDataset(repo_id)
-        if dataset.video:
+        if len(dataset.meta.video_keys) > 0:
             raise ValueError(
                 f"Use only image dataset for running this benchmark. Video dataset provided: {repo_id}"
             )
@@ -416,7 +416,7 @@ if __name__ == "__main__":
         "--vcodec",
         type=str,
         nargs="*",
-        default=["libx264", "libx265", "libsvtav1"],
+        default=["libx264", "hevc", "libsvtav1"],
         help="Video codecs to be tested",
     )
     parser.add_argument(
@@ -446,7 +446,7 @@ if __name__ == "__main__":
     #     nargs="*",
     #     default=[0, 1],
     #     help="Use the fastdecode tuning option. 0 disables it. "
-    #         "For libx264 and libx265, only 1 is possible. "
+    #         "For libx264 and libx265/hevc, only 1 is possible. "
     #         "For libsvtav1, 1, 2 or 3 are possible values with a higher number meaning a faster decoding optimization",
     # )
     parser.add_argument(

docker/lerobot-gpu-dev/Dockerfile

@@ -14,7 +14,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     tcpdump sysstat screen tmux \
     libglib2.0-0 libgl1-mesa-glx libegl1-mesa \
     speech-dispatcher portaudio19-dev libgeos-dev \
-    python${PYTHON_VERSION} python${PYTHON_VERSION}-venv \
+    python${PYTHON_VERSION} python${PYTHON_VERSION}-venv python${PYTHON_VERSION}-dev \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 
 # Install ffmpeg build dependencies. See:

docs/README.md

@@ -0,0 +1,137 @@
+<!---
+Copyright 2020 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+-->
+
+# Generating the documentation
+
+To generate the documentation, you first have to build it. Several packages are necessary to build the doc,
+you can install them with the following command, at the root of the code repository:
+
+```bash
+pip install -e ".[docs]"
+```
+
+You will also need `nodejs`. Please refer to their [installation page](https://nodejs.org/en/download)
+
+---
+**NOTE**
+
+You only need to generate the documentation to inspect it locally (if you're planning changes and want to
+check how they look before committing for instance). You don't have to `git commit` the built documentation.
+
+---
+
+## Building the documentation
+
+Once you have setup the `doc-builder` and additional packages, you can generate the documentation by
+typing the following command:
+
+```bash
+doc-builder build lerobot docs/source/ --build_dir ~/tmp/test-build
+```
+
+You can adapt the `--build_dir` to set any temporary folder that you prefer. This command will create it and generate
+the MDX files that will be rendered as the documentation on the main website. You can inspect them in your favorite
+Markdown editor.
+
+## Previewing the documentation
+
+To preview the docs, first install the `watchdog` module with:
+
+```bash
+pip install watchdog
+```
+
+Then run the following command:
+
+```bash
+doc-builder preview lerobot docs/source/
+```
+
+The docs will be viewable at [http://localhost:3000](http://localhost:3000). You can also preview the docs once you have opened a PR. You will see a bot add a comment to a link where the documentation with your changes lives.
+
+---
+**NOTE**
+
+The `preview` command only works with existing doc files. When you add a completely new file, you need to update `_toctree.yml` & restart `preview` command (`ctrl-c` to stop it & call `doc-builder preview ...` again).
+
+---
+
+## Adding a new element to the navigation bar
+
+Accepted files are Markdown (.md).
+
+Create a file with its extension and put it in the source directory. You can then link it to the toc-tree by putting
+the filename without the extension in the [`_toctree.yml`](https://github.com/huggingface/lerobot/blob/main/docs/source/_toctree.yml) file.
+
+## Renaming section headers and moving sections
+
+It helps to keep the old links working when renaming the section header and/or moving sections from one document to another. This is because the old links are likely to be used in Issues, Forums, and Social media and it'd make for a much more superior user experience if users reading those months later could still easily navigate to the originally intended information.
+
+Therefore, we simply keep a little map of moved sections at the end of the document where the original section was. The key is to preserve the original anchor.
+
+So if you renamed a section from: "Section A" to "Section B", then you can add at the end of the file:
+
+```
+Sections that were moved:
+
+[ <a href="#section-b">Section A</a><a id="section-a"></a> ]
+```
+and of course, if you moved it to another file, then:
+
+```
+Sections that were moved:
+
+[ <a href="../new-file#section-b">Section A</a><a id="section-a"></a> ]
+```
+
+Use the relative style to link to the new file so that the versioned docs continue to work.
+
+For an example of a rich moved sections set please see the very end of [the transformers Trainer doc](https://github.com/huggingface/transformers/blob/main/docs/source/en/main_classes/trainer.md).
+
+### Adding a new tutorial
+
+Adding a new tutorial or section is done in two steps:
+
+- Add a new file under `./source`. This file can either be ReStructuredText (.rst) or Markdown (.md).
+- Link that file in `./source/_toctree.yml` on the correct toc-tree.
+
+Make sure to put your new file under the proper section. If you have a doubt, feel free to ask in a Github Issue or PR.
+
+### Writing source documentation
+
+Values that should be put in `code` should either be surrounded by backticks: \`like so\`. Note that argument names
+and objects like True, None or any strings should usually be put in `code`.
+
+#### Writing a multi-line code block
+
+Multi-line code blocks can be useful for displaying examples. They are done between two lines of three backticks as usual in Markdown:
+
+
+````
+```
+# first line of code
+# second line
+# etc
+```
+````
+
+#### Adding an image
+
+Due to the rapidly growing repository, it is important to make sure that no files that would significantly weigh down the repository are added. This includes images, videos, and other non-text files. We prefer to leverage a hf.co hosted `dataset` like
+the ones hosted on [`hf-internal-testing`](https://huggingface.co/hf-internal-testing) in which to place these files and reference
+them by URL. We recommend putting them in the following dataset: [huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images).
+If an external contribution, feel free to add the images to your PR and ask a Hugging Face member to migrate your images
+to this dataset.

docs/source/_toctree.yml

@@ -0,0 +1,26 @@
+- sections:
+  - local: index
+    title: LeRobot
+  - local: installation
+    title: Installation
+  title: Get started
+- sections:
+  - local: getting_started_real_world_robot
+    title: Getting Started with Real-World Robots
+  - local: cameras
+    title: Cameras
+  title: "Tutorials"
+- sections:
+  - local: so101
+    title: SO-101
+  - local: so100
+    title: SO-100
+  - local: koch
+    title: Koch v1.1
+  - local: lekiwi
+    title: LeKiwi
+  title: "Robots"
+- sections:
+  - local: contributing
+    title: Contribute to LeRobot
+  title: "Contribute"

docs/source/cameras.mdx

@@ -0,0 +1,190 @@
+# Cameras
+
+Here we describe how to set up and use a camera with LeRobot. We support different ways of capturing videos in LeRobot, such as using a phone camera, an integrated laptop camera, an external webcam, or an Intel realsense camera.
+## Set up Cameras
+
+There are three ways to connect and use a camera with LeRobot:
+1. Use [Camera Class](./setup_cameras?use+phone=Mac#use-opencvcamera), which allows you to use any camera: usb, realsense, laptop webcam
+2. Use [iPhone camera](./setup_cameras?use+phone=Mac#use-your-phone) with MacOS
+3. Use [Phone camera](./setup_cameras?use+phone=Linux#use-your-phone) on Linux
+
+### Use Camera Class
+
+In LeRobot, you can efficiently record frames from most cameras using either the OpenCVCamera class or the RealSenseCamera class. For more details on compatibility for the OpenCVCamera class, see [Video I/O with OpenCV Overview](https://docs.opencv.org/4.x/d0/da7/videoio_overview.html).
+
+To instantiate a camera, you need a camera index. When you only have one camera like a webcam of a laptop, the camera index is usually `0` but it might differ, and the camera index might change if you reboot your computer or re-plug your camera. This behavior depends on your operating system.
+
+To find the camera indices, run the following script:
+```bash
+python lerobot/find_cameras.py list-cameras
+```
+
+The output will look something like this if you have two cameras connected:
+```
+--- Detected Cameras ---
+Camera #0:
+  Name: OpenCV Camera @ 0
+  Type: OpenCV
+  Id: 0
+  Backend api: AVFOUNDATION
+  Default stream profile:
+    Format: 16.0
+    Width: 1920
+    Height: 1080
+    Fps: 15.0
+--------------------
+Camera #1:
+  Name: OpenCV Camera @ 1
+  Type: OpenCV
+  Id: 1
+  Backend api: AVFOUNDATION
+  Default stream profile:
+    Format: 16.0
+    Width: 1920
+    Height: 1080
+    Fps: 1.0
+--------------------
+```
+
+> [!WARNING]
+> On , you could get this error: `Error finding RealSense cameras: failed to set power state`, this can be solved by running the same command with `sudo` permissions.
+
+### Use your phone
+<hfoptions id="use phone">
+<hfoption id="Mac">
+
+To use your iPhone as a camera on macOS, enable the Continuity Camera feature:
+- Ensure your Mac is running macOS 13 or later, and your iPhone is on iOS 16 or later.
+- Sign in both devices with the same Apple ID.
+- Connect your devices with a USB cable or turn on Wi-Fi and Bluetooth for a wireless connection.
+
+For more details, visit [Apple support](https://support.apple.com/en-gb/guide/mac-help/mchl77879b8a/mac).
+
+Your iPhone should be detected automatically when running the camera setup script in the next section.
+
+</hfoption>
+<hfoption id="Linux">
+
+If you want to use your phone as a camera on Linux, follow these steps to set up a virtual camera
+
+1. *Install `v4l2loopback-dkms` and `v4l-utils`*. Those packages are required to create virtual camera devices (`v4l2loopback`) and verify their settings with the `v4l2-ctl` utility from `v4l-utils`. Install them using:
+```python
+sudo apt install v4l2loopback-dkms v4l-utils
+```
+2. *Install [DroidCam](https://droidcam.app) on your phone*. This app is available for both iOS and Android.
+3. *Install [OBS Studio](https://obsproject.com)*. This software will help you manage the camera feed. Install it using [Flatpak](https://flatpak.org):
+```python
+flatpak install flathub com.obsproject.Studio
+```
+4. *Install the DroidCam OBS plugin*. This plugin integrates DroidCam with OBS Studio. Install it with:
+```python
+flatpak install flathub com.obsproject.Studio.Plugin.DroidCam
+```
+5. *Start OBS Studio*. Launch with:
+```python
+flatpak run com.obsproject.Studio
+```
+6. *Add your phone as a source*. Follow the instructions [here](https://droidcam.app/obs/usage). Be sure to set the resolution to `640x480`.
+7. *Adjust resolution settings*. In OBS Studio, go to `File > Settings > Video`. Change the `Base(Canvas) Resolution` and the `Output(Scaled) Resolution` to `640x480` by manually typing it in.
+8. *Start virtual camera*. In OBS Studio, follow the instructions [here](https://obsproject.com/kb/virtual-camera-guide).
+9. *Verify the virtual camera setup*. Use `v4l2-ctl` to list the devices:
+```python
+v4l2-ctl --list-devices
+```
+You should see an entry like:
+```
+VirtualCam (platform:v4l2loopback-000):
+/dev/video1
+```
+10. *Check the camera resolution*. Use `v4l2-ctl` to ensure that the virtual camera output resolution is `640x480`. Change `/dev/video1` to the port of your virtual camera from the output of `v4l2-ctl --list-devices`.
+```python
+v4l2-ctl -d /dev/video1 --get-fmt-video
+```
+You should see an entry like:
+```
+>>> Format Video Capture:
+>>>	Width/Height      : 640/480
+>>>	Pixel Format      : 'YUYV' (YUYV 4:2:2)
+```
+
+Troubleshooting: If the resolution is not correct you will have to delete the Virtual Camera port and try again as it cannot be changed.
+
+If everything is set up correctly, you can proceed with the rest of the tutorial.
+
+</hfoption>
+</hfoptions>
+
+## Use Cameras
+
+Below are two examples, demonstrating how to work with the API.
+
+- **Asynchronous frame capture** using an OpenCV-based camera
+- **Color and depth capture** using an Intel RealSense camera
+
+### Asynchronous OpenCV Camera
+
+This snippet shows how to:
+
+* Construct an `OpenCVCameraConfig` with your desired FPS, resolution, color mode, and rotation.
+* Instantiate and connect an `OpenCVCamera`, performing a warm-up read.
+* Read frames asynchronously in a loop via `async_read(timeout_ms)`.
+
+```python
+from lerobot.common.cameras.opencv.configuration_opencv import OpenCVCameraConfig
+from lerobot.common.cameras.opencv.camera_opencv import OpenCVCamera
+from lerobot.common.cameras.configs import ColorMode, Cv2Rotation
+
+config = OpenCVCameraConfig(
+    index_or_path=0,
+    fps=15,
+    width=1920,
+    height=1080,
+    color_mode=ColorMode.RGB,
+    rotation=Cv2Rotation.NO_ROTATION
+)
+
+camera = OpenCVCamera(config)
+camera.connect(do_warmup_read=True)
+
+try:
+    for i in range(10):
+        frame = camera.async_read(timeout_ms=1000)
+        print(f"Async frame {i} shape:", frame.shape)
+finally:
+    camera.disconnect()
+```
+
+### Intel RealSense Camera (Color + Depth)
+
+This snippet shows how to:
+
+* Create a `RealSenseCameraConfig` specifying your camera’s serial number and enabling depth.
+* Instantiate and connect a `RealSenseCamera` with warm-up.
+* Capture a color frame via `read()` and a depth map via `read_depth()`.
+
+```python
+from lerobot.common.cameras.intel.configuration_realsense import RealSenseCameraConfig
+from lerobot.common.cameras.intel.camera_realsense import RealSenseCamera
+from lerobot.common.cameras.configs import ColorMode, Cv2Rotation
+
+config = RealSenseCameraConfig(
+    serial_number="233522074606",
+    fps=15,
+    width=640,
+    height=480,
+    color_mode=ColorMode.RGB,
+    use_depth=True,
+    rotation=Cv2Rotation.NO_ROTATION
+)
+
+camera = RealSenseCamera(config)
+camera.connect(do_warmup_read=True)
+
+try:
+    color_frame = camera.read()
+    depth_map = camera.read_depth()
+    print("Color frame shape:", color_frame.shape)
+    print("Depth map shape:", depth_map.shape)
+finally:
+    camera.disconnect()
+```

docs/source/contributing.md

@@ -0,0 +1 @@
+../../CONTRIBUTING.md

docs/source/getting_started_real_world_robot.mdx

@@ -0,0 +1,384 @@
+# Getting Started with Real-World Robots
+
+This tutorial will explain how to train a neural network to control a real robot autonomously.
+
+**You'll learn:**
+1. How to record and visualize your dataset.
+2. How to train a policy using your data and prepare it for evaluation.
+3. How to evaluate your policy and visualize the results.
+
+By following these steps, you'll be able to replicate tasks, such as picking up a Lego block and placing it in a bin with a high success rate, as shown in [this video](https://x.com/RemiCadene/status/1814680760592572934).
+
+This tutorial isn’t tied to a specific robot: we walk you through the commands and API snippets you can adapt for any supported platform.
+
+During data collection, you’ll use a “teloperation” device, such as a leader arm or keyboard to teleoperate the robot and record its motion trajectories.
+
+Once you’ve gathered enough trajectories, you’ll train a neural network to imitate these trajectories and deploy the trained model so your robot can perform the task autonomously.
+
+If you run into any issues at any point, jump into our [Discord community](https://discord.com/invite/s3KuuzsPFb) for support.
+
+## Set up and Calibrate
+
+If you haven't yet set up and calibrated your robot and teleop device, please do so by following the robot-specific tutorial.
+
+## Teleoperate
+
+In this example, we’ll demonstrate how to teleoperate the SO101 robot. For each command, we also provide a corresponding API example.
+
+<hfoptions id="teleoperate_so101">
+<hfoption id="Command">
+```bash
+python -m lerobot.teleoperate \
+    --robot.type=so101_follower \
+    --robot.port=/dev/tty.usbmodem58760431541 \
+    --robot.cameras="{}" \
+    --robot.id=my_red_robot_arm \
+    --teleop.type=so101_leader \
+    --teleop.port=/dev/tty.usbmodem58760431551 \
+    --teleop.id=my_blue_leader_arm
+```
+</hfoption>
+<hfoption id="API example">
+```python
+from lerobot.common.teleoperators.so101_leader import SO101LeaderConfig, SO101Leader
+from lerobot.common.robots.so101_follower import SO101FollowerConfig, SO101Follower
+
+robot_config = SO101FollowerConfig(
+    port="/dev/tty.usbmodem58760431541",
+    id="my_red_robot_arm",
+)
+
+teleop_config = SO101LeaderConfig(
+    port="/dev/tty.usbmodem58760431551",
+    id="my_blue_leader_arm",
+)
+
+robot = SO101Follower(robot_config)
+teleop_device = SO101Leader(teleop_config)
+robot.connect()
+teleop_device.connect()
+
+while True:
+    action = teleop_device.get_action()
+    robot.send_action(action)
+```
+</hfoption>
+</hfoptions>
+
+The teleoperate command will automatically:
+1. Identify any missing calibrations and initiate the calibration procedure.
+2. Connect the robot and teleop device and start teleoperation.
+
+## Cameras
+
+To add cameras to your setup, follow this [Guide](./cameras#setup-cameras).
+
+## Teleoperate with cameras
+
+With `rerun`, you can teleoperate again while simultaneously visualizing the camera feeds and joint positions. In this example, we’re using the Koch arm.
+
+<hfoptions id="teleoperate_koch_camera">
+<hfoption id="Command">
+```bash
+python -m lerobot.teleoperate \
+    --robot.type=koch_follower \
+    --robot.port=/dev/tty.usbmodem58760431541 \
+    --robot.id=my_koch_robot \
+    --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 1920, height: 1080, fps: 30}}" \
+    --teleop.type=koch_leader \
+    --teleop.port=/dev/tty.usbmodem58760431551 \
+    --teleop.id=my_koch_teleop \
+    --display_data=true
+```
+</hfoption>
+<hfoption id="API example">
+```python
+from lerobot.common.cameras.opencv.configuration_opencv import OpenCVCameraConfig
+from lerobot.common.teleoperators.koch_leader import KochLeaderConfig, KochLeader
+from lerobot.common.robots.koch_follower import KochFollowerConfig, KochFollower
+
+camera_config = {
+    "front": OpenCVCameraConfig(index_or_path=0, width=1920, height=1080, fps=30)
+}
+
+robot_config = KochFollowerConfig(
+    port="/dev/tty.usbmodem585A0076841",
+    id="my_red_robot_arm",
+    cameras=camera_config
+)
+
+teleop_config = KochLeaderConfig(
+    port="/dev/tty.usbmodem58760431551",
+    id="my_blue_leader_arm",
+)
+
+robot = KochFollower(robot_config)
+teleop_device = KochLeader(teleop_config)
+robot.connect()
+teleop_device.connect()
+
+while True:
+    observation = robot.get_observation()
+    action = teleop_device.get_action()
+    robot.send_action(action)
+```
+</hfoption>
+</hfoptions>
+
+## Teleoperate LeKiwi
+
+> [!TIP]
+> If you're using a Mac, you might need to give Terminal permission to access your keyboard for teleoperation. Go to System Preferences > Security & Privacy > Input Monitoring and check the box for Terminal.
+
+TODO(pepijn): modify these commands with new API, and explain you can also map the arms to a keyboard with the new API now
+
+To teleoperate SSH into your Raspberry Pi, and run `conda activate lerobot` and this script:
+```bash
+python -m lerobot.common.robots.lekiwi.lekiwi_host
+```
+
+Then on your laptop, also run `conda activate lerobot` and this script:
+```bash
+python -m lerobot.teleoperate \
+    --robot.type=lekiwi \
+    --robot.port=/dev/tty.usbmodem58760431541 \
+    --robot.cameras="{}" \
+    --robot.id=my_lekiwi \
+    --teleop.type=so101_leader \
+    --teleop.port=/dev/tty.usbmodem58760431551 \
+    --teleop.id=my_blue_leader_arm
+```
+
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`. For the `--control.type=remote_robot` you will also need to set `--control.viewer_ip` and `--control.viewer_port`
+
+You should see on your laptop something like this: ```[INFO] Connected to remote robot at tcp://172.17.133.91:5555 and video stream at tcp://172.17.133.91:5556.``` Now you can move the leader arm and use the keyboard (w,a,s,d) to drive forward, left, backwards, right. And use (z,x) to turn left or turn right. You can use (r,f) to increase and decrease the speed of the mobile robot. There are three speed modes, see the table below:
+
+| Speed Mode | Linear Speed (m/s) | Rotation Speed (deg/s) |
+| ---------- | ------------------ | ---------------------- |
+| Fast       | 0.4                | 90                     |
+| Medium     | 0.25               | 60                     |
+| Slow       | 0.1                | 30                     |
+
+
+| Key | Action         |
+| --- | -------------- |
+| W   | Move forward   |
+| A   | Move left      |
+| S   | Move backward  |
+| D   | Move right     |
+| Z   | Turn left      |
+| X   | Turn right     |
+| R   | Increase speed |
+| F   | Decrease speed |
+
+> [!TIP]
+>  If you use a different keyboard, you can change the keys for each command in the [`LeKiwiConfig`](../lerobot/common/robot_devices/robots/configs.py).
+
+### Wired version
+If you have the **wired** LeKiwi version, please run all commands on your laptop.
+
+## Record a dataset
+
+Once you're familiar with teleoperation, you can record your first dataset.
+
+We use the Hugging Face hub features for uploading your dataset. If you haven't previously used the Hub, make sure you can login via the cli using a write-access token, this token can be generated from the [Hugging Face settings](https://huggingface.co/settings/tokens).
+
+Add your token to the CLI by running this command:
+```bash
+huggingface-cli login --token ${HUGGINGFACE_TOKEN} --add-to-git-credential
+```
+
+Then store your Hugging Face repository name in a variable:
+```bash
+HF_USER=$(huggingface-cli whoami | head -n 1)
+echo $HF_USER
+```
+
+Now you can record a dataset. To record 2 episodes and upload your dataset to the hub, execute this command tailored to the SO101.
+```bash
+python -m lerobot.record \
+    --robot.type=so101_follower \
+    --robot.port=/dev/tty.usbmodem585A0076841 \
+    --robot.id=my_red_robot_arm \
+    --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 1920, height: 1080, fps: 30}}" \
+    --teleop.type=so101_leader \
+    --teleop.port=/dev/tty.usbmodem58760431551 \
+    --teleop.id=my_blue_leader_arm \
+    --display_data=true \
+    --dataset.repo_id=aliberts/record-test \
+    --dataset.num_episodes=2 \
+    --dataset.single_task="Grab the black cube"
+```
+
+You will see a lot of lines appearing like this one:
+```
+INFO 2024-08-10 15:02:58 ol_robot.py:219 dt:33.34 (30.0hz) dtRlead: 5.06 (197.5hz) dtWfoll: 0.25 (3963.7hz) dtRfoll: 6.22 (160.7hz) dtRlaptop: 32.57 (30.7hz) dtRphone: 33.84 (29.5hz)
+```
+
+| Field | Meaning |
+|:---|:---|
+| `2024-08-10 15:02:58` | Timestamp when `print` was called. |
+| `ol_robot.py:219` | Source file and line number of the `print` call (`lerobot/scripts/control_robot.py` at line `219`). |
+| `dt: 33.34 (30.0 Hz)` | Delta time (ms) between teleop steps (target: 30.0 Hz, `--fps 30`). Yellow if step is too slow. |
+| `dtRlead: 5.06 (197.5 Hz)` | Delta time (ms) for reading present position from the **leader arm**. |
+| `dtWfoll: 0.25 (3963.7 Hz)` | Delta time (ms) for writing goal position to the **follower arm** (asynchronous). |
+| `dtRfoll: 6.22 (160.7 Hz)` | Delta time (ms) for reading present position from the **follower arm**. |
+| `dtRlaptop: 32.57 (30.7 Hz)` | Delta time (ms) for capturing an image from the **laptop camera** (async thread). |
+| `dtRphone: 33.84 (29.5 Hz)` | Delta time (ms) for capturing an image from the **phone camera** (async thread). |
+
+
+#### Dataset upload
+Locally, your dataset is stored in this folder: `~/.cache/huggingface/lerobot/{repo-id}` (e.g. `data/cadene/so101_test`). At the end of data recording, your dataset will be uploaded on your Hugging Face page (e.g. https://huggingface.co/datasets/cadene/so101_test) that you can obtain by running:
+```bash
+echo https://huggingface.co/datasets/${HF_USER}/so101_test
+```
+Your dataset will be automatically tagged with `LeRobot` for the community to find it easily, and you can also add custom tags (in this case `tutorial` for example).
+
+You can look for other LeRobot datasets on the hub by searching for `LeRobot` [tags](https://huggingface.co/datasets?other=LeRobot).
+
+#### Record function
+
+The `record` function provides a suite of tools for capturing and managing data during robot operation:
+
+##### 1. Data Storage
+- Data is stored using the `LeRobotDataset` format and is stored on disk during recording.
+- By default, the dataset is pushed to your Hugging Face page after recording.
+  - To disable uploading, use `--dataset.push_to_hub=False`.
+
+##### 2. Checkpointing and Resuming
+- Checkpoints are automatically created during recording.
+- If an issue occurs, you can resume by re-running the same command with `--control.resume=true`.
+- To start recording from scratch, **manually delete** the dataset directory.
+
+##### 3. Recording Parameters
+Set the flow of data recording using command-line arguments:
+- `--warmup_time_s=10`
+  Number of seconds before starting data collection (default: **10 seconds**).
+  Allows devices to warm up and synchronize.
+- `--dataset.episode_time_s=60`
+  Duration of each data recording episode (default: **60 seconds**).
+- `--dataset.reset_time_s=60`
+  Duration for resetting the environment after each episode (default: **60 seconds**).
+- `--dataset.num_episodes=50`
+  Total number of episodes to record (default: **50**).
+
+##### 4. Keyboard Controls During Recording
+Control the data recording flow using keyboard shortcuts:
+- Press **Right Arrow (`→`)**: Early stop the current episode or reset time and move to the next.
+- Press **Left Arrow (`←`)**: Cancel the current episode and re-record it.
+- Press **Escape (`ESC`)**: Immediately stop the session, encode videos, and upload the dataset.
+
+#### Tips for gathering data
+
+Once you're comfortable with data recording, you can create a larger dataset for training. A good starting task is grasping an object at different locations and placing it in a bin. We suggest recording at least 50 episodes, with 10 episodes per location. Keep the cameras fixed and maintain consistent grasping behavior throughout the recordings. Also make sure the object you are manipulating is visible on the camera's. A good rule of thumb is you should be able to do the task yourself by only looking at the camera images.
+
+In the following sections, you’ll train your neural network. After achieving reliable grasping performance, you can start introducing more variations during data collection, such as additional grasp locations, different grasping techniques, and altering camera positions.
+
+Avoid adding too much variation too quickly, as it may hinder your results.
+
+If you want to dive deeper into this important topic, you can check out the [blog post](https://huggingface.co/blog/lerobot-datasets#what-makes-a-good-dataset) we wrote on what makes a good dataset.
+
+
+#### Troubleshooting:
+- On Linux, if the left and right arrow keys and escape key don't have any effect during data recording, make sure you've set the `$DISPLAY` environment variable. See [pynput limitations](https://pynput.readthedocs.io/en/latest/limitations.html#linux).
+
+## Visualize a dataset
+
+If you uploaded your dataset to the hub with `--control.push_to_hub=true`, you can [visualize your dataset online](https://huggingface.co/spaces/lerobot/visualize_dataset) by copy pasting your repo id given by:
+```bash
+echo ${HF_USER}/so101_test
+```
+
+If you didn't upload with `--control.push_to_hub=false`, you can visualize it locally with (via a window in the browser `http://127.0.0.1:9090` with the visualization tool):
+```bash
+python lerobot/scripts/visualize_dataset_html.py \
+  --repo-id ${HF_USER}/so101_test \
+  --local-files-only 1
+```
+
+This will launch a local web server that looks like this:
+<div style="text-align:center;">
+  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/lerobot/visualize_dataset_html.webp?raw=true" alt="Koch v1.1 leader and follower arms" title="Koch v1.1 leader and follower arms" width="100%"></img>
+</div>
+
+## Replay an episode
+
+A useful feature is the `replay` function, which allows you to replay any episode that you've recorded or episodes from any dataset out there. This function helps you test the repeatability of your robot's actions and assess transferability across robots of the same model.
+
+You can replay the first episode on your robot with:
+```bash
+python -m lerobot.replay \
+    --robot.type=so101_follower \
+    --robot.port=/dev/tty.usbmodem58760431541 \
+    --robot.id=black \
+    --dataset.repo_id=aliberts/record-test \
+    --dataset.episode=2
+```
+
+Your robot should replicate movements similar to those you recorded. For example, check out [this video](https://x.com/RemiCadene/status/1793654950905680090) where we use `replay` on a Aloha robot from [Trossen Robotics](https://www.trossenrobotics.com).
+
+## Train a policy
+
+To train a policy to control your robot, use the [`python lerobot/scripts/train.py`](../lerobot/scripts/train.py) script. A few arguments are required. Here is an example command:
+```bash
+python lerobot/scripts/train.py \
+  --dataset.repo_id=${HF_USER}/so101_test \
+  --policy.type=act \
+  --output_dir=outputs/train/act_so101_test \
+  --job_name=act_so101_test \
+  --policy.device=cuda \
+  --wandb.enable=true
+```
+
+Let's explain the command:
+1. We provided the dataset as argument with `--dataset.repo_id=${HF_USER}/so101_test`.
+2. We provided the policy with `policy.type=act`. This loads configurations from [`configuration_act.py`](../lerobot/common/policies/act/configuration_act.py). Importantly, this policy will automatically adapt to the number of motor states, motor actions and cameras of your robot (e.g. `laptop` and `phone`) which have been saved in your dataset.
+4. We provided `policy.device=cuda` since we are training on a Nvidia GPU, but you could use `policy.device=mps` to train on Apple silicon.
+5. We provided `wandb.enable=true` to use [Weights and Biases](https://docs.wandb.ai/quickstart) for visualizing training plots. This is optional but if you use it, make sure you are logged in by running `wandb login`.
+
+Training should take several hours. You will find checkpoints in `outputs/train/act_so101_test/checkpoints`.
+
+To resume training from a checkpoint, below is an example command to resume from `last` checkpoint of the `act_so101_test` policy:
+```bash
+python lerobot/scripts/train.py \
+  --config_path=outputs/train/act_so101_test/checkpoints/last/pretrained_model/train_config.json \
+  --resume=true
+```
+
+#### Upload policy checkpoints
+
+Once training is done, upload the latest checkpoint with:
+```bash
+huggingface-cli upload ${HF_USER}/act_so101_test \
+  outputs/train/act_so101_test/checkpoints/last/pretrained_model
+```
+
+You can also upload intermediate checkpoints with:
+```bash
+CKPT=010000
+huggingface-cli upload ${HF_USER}/act_so101_test${CKPT} \
+  outputs/train/act_so101_test/checkpoints/${CKPT}/pretrained_model
+```
+
+## Evaluate your policy
+
+TODO(pepijn): modify this command further
+
+You can use the `record` script from [`lerobot/record.py`](https://github.com/huggingface/lerobot/blob/main/lerobot/record.py) but with a policy checkpoint as input. For instance, run this command to record 10 evaluation episodes:
+```bash
+python -m lerobot.record \
+    --robot.type=so101_follower \
+    --robot.port=/dev/tty.usbmodem585A0076841 \
+    --robot.id=my_red_robot_arm \
+    --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 1920, height: 1080, fps: 30}}" \
+    --teleop.type=so101_leader \
+    --teleop.port=/dev/tty.usbmodem58760431551 \
+    --teleop.id=my_blue_leader_arm \
+    --display_data=true \
+    --dataset.repo_id=aliberts/eval_act \
+    --policy.checkpoint_path=outputs/train/act_so101_test/checkpoints/last/pretrained_model \
+```
+
+As you can see, it's almost the same command as previously used to record your training dataset. Two things changed:
+1. There is an additional `--control.policy.path` argument which indicates the path to your policy checkpoint with  (e.g. `outputs/train/eval_act_so101_test/checkpoints/last/pretrained_model`). You can also use the model repository if you uploaded a model checkpoint to the hub (e.g. `${HF_USER}/act_so101_test`).
+2. The name of dataset begins by `eval` to reflect that you are running inference (e.g. `${HF_USER}/eval_act_so101_test`).

docs/source/index.mdx

@@ -0,0 +1,19 @@
+<div class="flex justify-center">
+  <a target="_blank" href="https://huggingface.co/lerobot">
+      <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/lerobot/lerobot-logo-thumbnail.png" style="width: 100%"></img>
+  </a>
+</div>
+
+# LeRobot
+
+**State-of-the-art machine learning for real-world robotics**
+
+🤗 LeRobot aims to provide models, datasets, and tools for real-world robotics in PyTorch. The goal is to lower the barrier for entry to robotics so that everyone can contribute and benefit from sharing datasets and pretrained models.
+
+🤗 LeRobot contains state-of-the-art approaches that have been shown to transfer to the real-world with a focus on imitation learning and reinforcement learning.
+
+🤗 LeRobot already provides a set of pretrained models, datasets with human collected demonstrations, and simulated environments so that everyone can get started.
+
+🤗 LeRobot hosts pretrained models and datasets on the LeRobot HuggingFace page.
+
+Join the LeRobot community on [Discord](https://discord.gg/s3KuuzsPFb)

docs/source/installation.mdx

@@ -0,0 +1,100 @@
+# Installation
+
+## Install LeRobot
+
+Download our source code:
+```bash
+git clone https://github.com/huggingface/lerobot.git
+cd lerobot
+```
+
+Create a virtual environment with Python 3.10, using [`Miniconda`](https://docs.anaconda.com/miniconda/install/#quick-command-line-install)
+```bash
+conda create -y -n lerobot python=3.10
+```
+
+Now restart the shell by running:
+<hfoptions id="shell_restart">
+<hfoption id="Windows">
+
+```bash
+source ~/.bashrc
+```
+</hfoption>
+<hfoption id="Mac">
+
+```bash
+source ~/.bash_profile
+```
+</hfoption>
+<hfoption id="zshell">
+
+```bash
+source ~/.zshrc
+```
+</hfoption>
+</hfoptions>
+
+Then activate your conda environment, you have to do this each time you open a shell to use lerobot:
+```bash
+conda activate lerobot
+```
+
+When using `miniconda`, install `ffmpeg` in your environment:
+```bash
+conda install ffmpeg -c conda-forge
+```
+
+> [!TIP]
+> This usually installs `ffmpeg 7.X` for your platform compiled with the `libsvtav1` encoder. If `libsvtav1` is not supported (check supported encoders with `ffmpeg -encoders`), you can:
+>  - _[On any platform]_ Explicitly install `ffmpeg 7.X` using:
+>  ```bash
+>  conda install ffmpeg=7.1.1 -c conda-forge
+>  ```
+>  - _[On Linux only]_ Install [ffmpeg build dependencies](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#GettheDependencies) and [compile ffmpeg from source with libsvtav1](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#libsvtav1), and make sure you use the corresponding ffmpeg binary to your install with `which ffmpeg`.
+
+Install 🤗 LeRobot:
+```bash
+cd lerobot && pip install
+```
+
+## Troubleshooting
+If you encounter build errors, you may need to install additional dependencies: `cmake`, `build-essential`, and `ffmpeg libs`.
+To install these for linux run:
+```bash
+sudo apt-get install cmake build-essential python-dev pkg-config libavformat-dev libavcodec-dev libavdevice-dev libavutil-dev libswscale-dev libswresample-dev libavfilter-dev pkg-config
+```
+For other systems, see: [Compiling PyAV](https://pyav.org/docs/develop/overview/installation.html#bring-your-own-ffmpeg)
+
+## Sim
+For simulations, 🤗 LeRobot comes with gymnasium environments that can be installed as extras:
+- [aloha](https://github.com/huggingface/gym-aloha)
+- [xarm](https://github.com/huggingface/gym-xarm)
+- [pusht](https://github.com/huggingface/gym-pusht)
+
+For instance, to install 🤗 LeRobot with aloha and pusht, use:
+```bash
+pip install -e ".[aloha, pusht]"
+```
+
+## Motor SDK
+For Koch v1.1 install the Dynamixel SDK, for SO100/SO101/Moss install the Feetech SDK.
+
+<hfoptions id="install motors">
+<hfoption id="Feetech">
+```bash
+pip install -e ".[feetech]"
+```
+</hfoption>
+<hfoption id="Dynamixel">
+```bash
+pip install -e ".[dynamixel]"
+```
+</hfoption>
+</hfoptions>
+
+## W&B
+To use [Weights and Biases](https://docs.wandb.ai/quickstart) for experiment tracking, log in with
+```bash
+wandb login
+```

docs/source/koch.md

@@ -0,0 +1 @@
+../../lerobot/common/robots/koch_follower/koch.md

docs/source/lekiwi.md

@@ -0,0 +1 @@
+../../lerobot/common/robots/lekiwi/lekiwi.md

docs/source/so100.md

@@ -0,0 +1 @@
+../../lerobot/common/robots/so100_follower/so100.md

docs/source/so101.md

@@ -0,0 +1 @@
+../../lerobot/common/robots/so101_follower/so101.md

examples/1_load_lerobot_dataset.py

@@ -119,7 +119,7 @@ print(dataset.features[camera_key]["shape"])
 delta_timestamps = {
     # loads 4 images: 1 second before current frame, 500 ms before, 200 ms before, and current frame
     camera_key: [-1, -0.5, -0.20, 0],
-    # loads 8 state vectors: 1.5 seconds before, 1 second before, ... 200 ms, 100 ms, and current frame
+    # loads 6 state vectors: 1.5 seconds before, 1 second before, ... 200 ms, 100 ms, and current frame
     "observation.state": [-1.5, -1, -0.5, -0.20, -0.10, 0],
     # loads 64 action vectors: current frame, 1 frame in the future, 2 frames, ... 63 frames in the future
     "action": [t / dataset.fps for t in range(64)],
@@ -143,6 +143,6 @@ dataloader = torch.utils.data.DataLoader(
 
 for batch in dataloader:
     print(f"{batch[camera_key].shape=}")  # (32, 4, c, h, w)
-    print(f"{batch['observation.state'].shape=}")  # (32, 5, c)
+    print(f"{batch['observation.state'].shape=}")  # (32, 6, c)
     print(f"{batch['action'].shape=}")  # (32, 64, c)
     break

examples/2_evaluate_pretrained_policy.py

@@ -13,12 +13,12 @@
 # limitations under the License.
 
 """
-This scripts demonstrates how to evaluate a pretrained policy from the HuggingFace Hub or from your local
+This script demonstrates how to evaluate a pretrained policy from the HuggingFace Hub or from your local
 training outputs directory. In the latter case, you might want to run examples/3_train_policy.py first.
 
 It requires the installation of the 'gym_pusht' simulation environment. Install it by running:
 ```bash
-pip install -e ".[pusht]"`
+pip install -e ".[pusht]"
 ```
 """
 
@@ -119,7 +119,7 @@ while not done:
     rewards.append(reward)
     frames.append(env.render())
 
-    # The rollout is considered done when the success state is reach (i.e. terminated is True),
+    # The rollout is considered done when the success state is reached (i.e. terminated is True),
     # or the maximum number of iterations is reached (i.e. truncated is True)
     done = terminated | truncated | done
     step += 1

examples/3_train_policy.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""This scripts demonstrates how to train Diffusion Policy on the PushT environment.
+"""This script demonstrates how to train Diffusion Policy on the PushT environment.
 
 Once you have trained a model with this script, you can try to evaluate it on
 examples/2_evaluate_pretrained_policy.py

examples/4_train_policy_with_script.md

@@ -1,10 +1,10 @@
 This tutorial will explain the training script, how to use it, and particularly how to configure everything needed for the training run.
-> **Note:** The following assume you're running these commands on a machine equipped with a cuda GPU. If you don't have one (or if you're using a Mac), you can add `--policy.device=cpu` (`--policy.device=mps` respectively). However, be advised that the code executes much slower on cpu.
+> **Note:** The following assumes you're running these commands on a machine equipped with a cuda GPU. If you don't have one (or if you're using a Mac), you can add `--policy.device=cpu` (`--policy.device=mps` respectively). However, be advised that the code executes much slower on cpu.
 
 
 ## The training script
 
-LeRobot offers a training script at [`lerobot/scripts/train.py`](../../lerobot/scripts/train.py). At a high level it does the following:
+LeRobot offers a training script at [`lerobot/scripts/train.py`](../lerobot/scripts/train.py). At a high level it does the following:
 
 - Initialize/load a configuration for the following steps using.
 - Instantiates a dataset.
@@ -21,9 +21,9 @@ In the training script, the main function `train` expects a `TrainPipelineConfig
 def train(cfg: TrainPipelineConfig):
 ```
 
-You can inspect the `TrainPipelineConfig` defined in [`lerobot/configs/train.py`](../../lerobot/configs/train.py) (which is heavily commented and meant to be a reference to understand any option)
+You can inspect the `TrainPipelineConfig` defined in [`lerobot/configs/train.py`](../lerobot/configs/train.py) (which is heavily commented and meant to be a reference to understand any option)
 
-When running the script, inputs for the command line are parsed thanks to the `@parser.wrap()` decorator and an instance of this class is automatically generated. Under the hood, this is done with [Draccus](https://github.com/dlwh/draccus) which is a tool dedicated for this purpose. If you're familiar with Hydra, Draccus can similarly load configurations from config files (.json, .yaml) and also override their values through command line inputs. Unlike Hydra, these configurations are pre-defined in the code through dataclasses rather than being defined entirely in config files. This allows for more rigorous serialization/deserialization, typing, and to manipulate configuration as objects directly in the code and not as dictionaries or namespaces (which enables nice features in an IDE such as autocomplete, jump-to-def, etc.)
+When running the script, inputs for the command line are parsed thanks to the `@parser.wrap()` decorator and an instance of this class is automatically generated. Under the hood, this is done with [Draccus](https://github.com/dlwh/draccus) which is a tool dedicated to this purpose. If you're familiar with Hydra, Draccus can similarly load configurations from config files (.json, .yaml) and also override their values through command line inputs. Unlike Hydra, these configurations are pre-defined in the code through dataclasses rather than being defined entirely in config files. This allows for more rigorous serialization/deserialization, typing, and to manipulate configuration as objects directly in the code and not as dictionaries or namespaces (which enables nice features in an IDE such as autocomplete, jump-to-def, etc.)
 
 Let's have a look at a simplified example. Amongst other attributes, the training config has the following attributes:
 ```python
@@ -43,14 +43,14 @@ class DatasetConfig:
 ```
 
 This creates a hierarchical relationship where, for example assuming we have a `cfg` instance of `TrainPipelineConfig`, we can access the `repo_id` value with `cfg.dataset.repo_id`.
-From the command line, we can specify this value with using a very similar syntax `--dataset.repo_id=repo/id`.
+From the command line, we can specify this value by using a very similar syntax `--dataset.repo_id=repo/id`.
 
 By default, every field takes its default value specified in the dataclass. If a field doesn't have a default value, it needs to be specified either from the command line or from a config file – which path is also given in the command line (more in this below). In the example above, the `dataset` field doesn't have a default value which means it must be specified.
 
 
 ## Specifying values from the CLI
 
-Let's say that we want to train [Diffusion Policy](../../lerobot/common/policies/diffusion) on the [pusht](https://huggingface.co/datasets/lerobot/pusht) dataset, using the [gym_pusht](https://github.com/huggingface/gym-pusht) environment for evaluation. The command to do so would look like this:
+Let's say that we want to train [Diffusion Policy](../lerobot/common/policies/diffusion) on the [pusht](https://huggingface.co/datasets/lerobot/pusht) dataset, using the [gym_pusht](https://github.com/huggingface/gym-pusht) environment for evaluation. The command to do so would look like this:
 ```bash
 python lerobot/scripts/train.py \
     --dataset.repo_id=lerobot/pusht \
@@ -60,10 +60,10 @@ python lerobot/scripts/train.py \
 
 Let's break this down:
 - To specify the dataset, we just need to specify its `repo_id` on the hub which is the only required argument in the `DatasetConfig`. The rest of the fields have default values and in this case we are fine with those so we can just add the option `--dataset.repo_id=lerobot/pusht`.
-- To specify the policy, we can just select diffusion policy using `--policy` appended with `.type`. Here, `.type` is a special argument which allows us to select config classes inheriting from `draccus.ChoiceRegistry` and that have been decorated with the `register_subclass()` method. To have a better explanation of this feature, have a look at this [Draccus demo](https://github.com/dlwh/draccus?tab=readme-ov-file#more-flexible-configuration-with-choice-types). In our code, we use this mechanism mainly to select policies, environments, robots, and some other components like optimizers. The policies available to select are located in [lerobot/common/policies](../../lerobot/common/policies)
-- Similarly, we select the environment with `--env.type=pusht`. The different environment configs are available in [`lerobot/common/envs/configs.py`](../../lerobot/common/envs/configs.py)
+- To specify the policy, we can just select diffusion policy using `--policy` appended with `.type`. Here, `.type` is a special argument which allows us to select config classes inheriting from `draccus.ChoiceRegistry` and that have been decorated with the `register_subclass()` method. To have a better explanation of this feature, have a look at this [Draccus demo](https://github.com/dlwh/draccus?tab=readme-ov-file#more-flexible-configuration-with-choice-types). In our code, we use this mechanism mainly to select policies, environments, robots, and some other components like optimizers. The policies available to select are located in [lerobot/common/policies](../lerobot/common/policies)
+- Similarly, we select the environment with `--env.type=pusht`. The different environment configs are available in [`lerobot/common/envs/configs.py`](../lerobot/common/envs/configs.py)
 
-Let's see another example. Let's say you've been training [ACT](../../lerobot/common/policies/act) on [lerobot/aloha_sim_insertion_human](https://huggingface.co/datasets/lerobot/aloha_sim_insertion_human) using the [gym-aloha](https://github.com/huggingface/gym-aloha) environment for evaluation with:
+Let's see another example. Let's say you've been training [ACT](../lerobot/common/policies/act) on [lerobot/aloha_sim_insertion_human](https://huggingface.co/datasets/lerobot/aloha_sim_insertion_human) using the [gym-aloha](https://github.com/huggingface/gym-aloha) environment for evaluation with:
 ```bash
 python lerobot/scripts/train.py \
     --policy.type=act \
@@ -74,7 +74,7 @@ python lerobot/scripts/train.py \
 > Notice we added `--output_dir` to explicitly tell where to write outputs from this run (checkpoints, training state, configs etc.). This is not mandatory and if you don't specify it, a default directory will be created from the current date and time, env.type and policy.type. This will typically look like `outputs/train/2025-01-24/16-10-05_aloha_act`.
 
 We now want to train a different policy for aloha on another task. We'll change the dataset and use [lerobot/aloha_sim_transfer_cube_human](https://huggingface.co/datasets/lerobot/aloha_sim_transfer_cube_human) instead. Of course, we also need to change the task of the environment as well to match this other task.
-Looking at the [`AlohaEnv`](../../lerobot/common/envs/configs.py) config, the task is `"AlohaInsertion-v0"` by default, which corresponds to the task we trained on in the command above. The [gym-aloha](https://github.com/huggingface/gym-aloha?tab=readme-ov-file#description) environment also has the `AlohaTransferCube-v0` task which corresponds to this other task we want to train on. Putting this together, we can train this new policy on this different task using:
+Looking at the [`AlohaEnv`](../lerobot/common/envs/configs.py) config, the task is `"AlohaInsertion-v0"` by default, which corresponds to the task we trained on in the command above. The [gym-aloha](https://github.com/huggingface/gym-aloha?tab=readme-ov-file#description) environment also has the `AlohaTransferCube-v0` task which corresponds to this other task we want to train on. Putting this together, we can train this new policy on this different task using:
 ```bash
 python lerobot/scripts/train.py \
     --policy.type=act \
@@ -135,7 +135,7 @@ will start a training run with the same configuration used for training [lerobot
 
 ## Resume training
 
-Being able to resume a training run is important in case it crashed or aborted for any reason. We'll demonstrate how to that here.
+Being able to resume a training run is important in case it crashed or aborted for any reason. We'll demonstrate how to do that here.
 
 Let's reuse the command from the previous run and add a few more options:
 ```bash

examples/5_control_robot.md

@@ -46,13 +46,6 @@ Using `uv`:
 uv sync --extra "dynamixel"
 ```
 
-/!\ For Linux only, ffmpeg and opencv requires conda install for now. Run this exact sequence of commands:
-```bash
-conda install -c conda-forge ffmpeg
-pip uninstall opencv-python
-conda install -c conda-forge "opencv>=4.10.0"
-```
-
 You are now ready to plug the 5V power supply to the motor bus of the leader arm (the smaller one) since all its motors only require 5V.
 
 Then plug the 12V power supply to the motor bus of the follower arm. It has two motors that need 12V, and the rest will be powered with 5V through the voltage convertor.
@@ -62,6 +55,9 @@ Finally, connect both arms to your computer via USB. Note that the USB doesn't p
 Now you are ready to configure your motors for the first time, as detailed in the sections below. In the upcoming sections, you'll learn about our classes and functions by running some python code in an interactive session, or by copy-pasting it in a python file.
 
 If you have already configured your motors the first time, you can streamline the process by directly running the teleoperate script (which is detailed further in the tutorial):
+
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`.
+
 ```bash
 python lerobot/scripts/control_robot.py \
   --robot.type=koch \
@@ -87,7 +83,7 @@ python lerobot/scripts/configure_motor.py \
   --brand dynamixel \
   --model xl330-m288 \
   --baudrate 1000000 \
-  --ID 1
+  --id 1
 ```
 
 Then unplug your first motor and plug the second motor and set its ID to 2.
@@ -97,7 +93,7 @@ python lerobot/scripts/configure_motor.py \
   --brand dynamixel \
   --model xl330-m288 \
   --baudrate 1000000 \
-  --ID 2
+  --id 2
 ```
 
 Redo the process for all your motors until ID 6.
@@ -292,6 +288,11 @@ Steps:
    - Scan for devices. All 12 motors should appear.
    - Select the motors one by one and move the arm. Check that the graphical indicator near the top right shows the movement.
 
+** There is a common issue with the Dynamixel XL430-W250 motors where the motors become undiscoverable after upgrading their firmware from Mac and Windows Dynamixel Wizard2 applications.  When this occurs, it is required to do a firmware recovery (Select `DYNAMIXEL Firmware Recovery` and follow the prompts).   There are two known workarounds to conduct this firmware reset:
+  1) Install the Dynamixel Wizard on a linux machine and complete the firmware recovery
+  2) Use the Dynamixel U2D2 in order to perform the reset with Windows or Mac.  This U2D2 can be purchased [here](https://www.robotis.us/u2d2/).
+  For either solution, open DYNAMIXEL Wizard 2.0 and select the appropriate port. You will likely be unable to see the motor in the GUI at this time. Select `Firmware Recovery`, carefully choose the correct model, and wait for the process to complete. Finally, re-scan to confirm the firmware recovery was successful.
+
 **Read and Write with DynamixelMotorsBus**
 
 To get familiar with how `DynamixelMotorsBus` communicates with the motors, you can start by reading data from them. Copy past this code in the same interactive python session:
@@ -376,7 +377,7 @@ robot = ManipulatorRobot(robot_config)
 
 The `KochRobotConfig` is used to set the associated settings and calibration process. For instance, we activate the torque of the gripper of the leader Koch v1.1 arm and position it at a 40 degree angle to use it as a trigger.
 
-For the [Aloha bimanual robot](https://aloha-2.github.io), we would use `AlohaRobotConfig` to set different settings such as a secondary ID for shadow joints (shoulder, elbow). Specific to Aloha, LeRobot comes with default calibration files stored in in `.cache/calibration/aloha_default`. Assuming the motors have been properly assembled, no manual calibration step is expected for Aloha.
+For the [Aloha bimanual robot](https://aloha-2.github.io), we would use `AlohaRobotConfig` to set different settings such as a secondary ID for shadow joints (shoulder, elbow). Specific to Aloha, LeRobot comes with default calibration files stored in `.cache/calibration/aloha_default`. Assuming the motors have been properly assembled, no manual calibration step is expected for Aloha.
 
 **Calibrate and Connect the ManipulatorRobot**
 
@@ -398,7 +399,7 @@ And here are the corresponding positions for the leader arm:
 
 You can watch a [video tutorial of the calibration procedure](https://youtu.be/8drnU9uRY24) for more details.
 
-During calibration, we count the number of full 360-degree rotations your motors have made since they were first used. That's why we ask yo to move to this arbitrary "zero" position. We don't actually "set" the zero position, so you don't need to be accurate. After calculating these "offsets" to shift the motor values around 0, we need to assess the rotation direction of each motor, which might differ. That's why we ask you to rotate all motors to roughly 90 degrees, to measure if the values changed negatively or positively.
+During calibration, we count the number of full 360-degree rotations your motors have made since they were first used. That's why we ask you to move to this arbitrary "zero" position. We don't actually "set" the zero position, so you don't need to be accurate. After calculating these "offsets" to shift the motor values around 0, we need to assess the rotation direction of each motor, which might differ. That's why we ask you to rotate all motors to roughly 90 degrees, to measure if the values changed negatively or positively.
 
 Finally, the rest position ensures that the follower and leader arms are roughly aligned after calibration, preventing sudden movements that could damage the motors when starting teleoperation.
 
@@ -621,7 +622,7 @@ camera_01_frame_000047.png
 
 Note: Some cameras may take a few seconds to warm up, and the first frame might be black or green.
 
-Finally, run this code to instantiate and connectyour camera:
+Finally, run this code to instantiate and connect your camera:
 ```python
 from lerobot.common.robot_devices.cameras.configs import OpenCVCameraConfig
 from lerobot.common.robot_devices.cameras.opencv import OpenCVCamera
@@ -829,16 +830,6 @@ It contains:
 - `dtRphone:33.84 (29.5hz)` which is the delta time of capturing an image from the phone camera in the thread running asynchronously.
 
 Troubleshooting:
-- On Linux, if you encounter a hanging issue when using cameras, uninstall opencv and re-install it with conda:
-```bash
-pip uninstall opencv-python
-conda install -c conda-forge opencv=4.10.0
-```
-- On Linux, if you encounter any issue during video encoding with `ffmpeg: unknown encoder libsvtav1`, you can:
-  - install with conda-forge by running `conda install -c conda-forge ffmpeg` (it should be compiled with `libsvtav1`),
-  - or, install [Homebrew](https://brew.sh) and run `brew install ffmpeg` (it should be compiled with `libsvtav1`),
-  - or, install [ffmpeg build dependencies](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#GettheDependencies) and [compile ffmpeg from source with libsvtav1](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#libsvtav1),
-  - and, make sure you use the corresponding ffmpeg binary to your install with `which ffmpeg`.
 - On Linux, if the left and right arrow keys and escape key don't have any effect during data recording, make sure you've set the `$DISPLAY` environment variable. See [pynput limitations](https://pynput.readthedocs.io/en/latest/limitations.html#linux).
 
 At the end of data recording, your dataset will be uploaded on your Hugging Face page (e.g. https://huggingface.co/datasets/cadene/koch_test) that you can obtain by running:

examples/advanced/2_calculate_validation_loss.py

@@ -66,7 +66,7 @@ def main():
     print(f"Number of episodes in full dataset: {total_episodes}")
     print(f"Number of episodes in training dataset (90% subset): {len(train_episodes)}")
     print(f"Number of episodes in validation dataset (10% subset): {len(val_episodes)}")
-    # - Load train an val datasets
+    # - Load train and val datasets
     train_dataset = LeRobotDataset(
         "lerobot/pusht", episodes=train_episodes, delta_timestamps=delta_timestamps
     )

examples/port_datasets/pusht_zarr.py

@@ -1,243 +0,0 @@
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import shutil
-from pathlib import Path
-
-import numpy as np
-from huggingface_hub import HfApi
-
-from lerobot.common.constants import HF_LEROBOT_HOME
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION, LeRobotDataset
-from lerobot.common.datasets.push_dataset_to_hub._download_raw import download_raw
-
-PUSHT_TASK = "Push the T-shaped blue block onto the T-shaped green target surface."
-PUSHT_FEATURES = {
-    "observation.state": {
-        "dtype": "float32",
-        "shape": (2,),
-        "names": {
-            "axes": ["x", "y"],
-        },
-    },
-    "action": {
-        "dtype": "float32",
-        "shape": (2,),
-        "names": {
-            "axes": ["x", "y"],
-        },
-    },
-    "next.reward": {
-        "dtype": "float32",
-        "shape": (1,),
-        "names": None,
-    },
-    "next.success": {
-        "dtype": "bool",
-        "shape": (1,),
-        "names": None,
-    },
-    "observation.environment_state": {
-        "dtype": "float32",
-        "shape": (16,),
-        "names": [
-            "keypoints",
-        ],
-    },
-    "observation.image": {
-        "dtype": None,
-        "shape": (3, 96, 96),
-        "names": [
-            "channels",
-            "height",
-            "width",
-        ],
-    },
-}
-
-
-def build_features(mode: str) -> dict:
-    features = PUSHT_FEATURES
-    if mode == "keypoints":
-        features.pop("observation.image")
-    else:
-        features.pop("observation.environment_state")
-        features["observation.image"]["dtype"] = mode
-
-    return features
-
-
-def load_raw_dataset(zarr_path: Path):
-    try:
-        from lerobot.common.datasets.push_dataset_to_hub._diffusion_policy_replay_buffer import (
-            ReplayBuffer as DiffusionPolicyReplayBuffer,
-        )
-    except ModuleNotFoundError as e:
-        print("`gym_pusht` is not installed. Please install it with `pip install 'lerobot[gym_pusht]'`")
-        raise e
-
-    zarr_data = DiffusionPolicyReplayBuffer.copy_from_path(zarr_path)
-    return zarr_data
-
-
-def calculate_coverage(zarr_data):
-    try:
-        import pymunk
-        from gym_pusht.envs.pusht import PushTEnv, pymunk_to_shapely
-    except ModuleNotFoundError as e:
-        print("`gym_pusht` is not installed. Please install it with `pip install 'lerobot[gym_pusht]'`")
-        raise e
-
-    block_pos = zarr_data["state"][:, 2:4]
-    block_angle = zarr_data["state"][:, 4]
-
-    num_frames = len(block_pos)
-
-    coverage = np.zeros((num_frames,), dtype=np.float32)
-    # 8 keypoints with 2 coords each
-    keypoints = np.zeros((num_frames, 16), dtype=np.float32)
-
-    # Set x, y, theta (in radians)
-    goal_pos_angle = np.array([256, 256, np.pi / 4])
-    goal_body = PushTEnv.get_goal_pose_body(goal_pos_angle)
-
-    for i in range(num_frames):
-        space = pymunk.Space()
-        space.gravity = 0, 0
-        space.damping = 0
-
-        # Add walls.
-        walls = [
-            PushTEnv.add_segment(space, (5, 506), (5, 5), 2),
-            PushTEnv.add_segment(space, (5, 5), (506, 5), 2),
-            PushTEnv.add_segment(space, (506, 5), (506, 506), 2),
-            PushTEnv.add_segment(space, (5, 506), (506, 506), 2),
-        ]
-        space.add(*walls)
-
-        block_body, block_shapes = PushTEnv.add_tee(space, block_pos[i].tolist(), block_angle[i].item())
-        goal_geom = pymunk_to_shapely(goal_body, block_body.shapes)
-        block_geom = pymunk_to_shapely(block_body, block_body.shapes)
-        intersection_area = goal_geom.intersection(block_geom).area
-        goal_area = goal_geom.area
-        coverage[i] = intersection_area / goal_area
-        keypoints[i] = PushTEnv.get_keypoints(block_shapes).flatten()
-
-    return coverage, keypoints
-
-
-def calculate_success(coverage: float, success_threshold: float):
-    return coverage > success_threshold
-
-
-def calculate_reward(coverage: float, success_threshold: float):
-    return np.clip(coverage / success_threshold, 0, 1)
-
-
-def main(raw_dir: Path, repo_id: str, mode: str = "video", push_to_hub: bool = True):
-    if mode not in ["video", "image", "keypoints"]:
-        raise ValueError(mode)
-
-    if (HF_LEROBOT_HOME / repo_id).exists():
-        shutil.rmtree(HF_LEROBOT_HOME / repo_id)
-
-    if not raw_dir.exists():
-        download_raw(raw_dir, repo_id="lerobot-raw/pusht_raw")
-
-    zarr_data = load_raw_dataset(zarr_path=raw_dir / "pusht_cchi_v7_replay.zarr")
-
-    env_state = zarr_data["state"][:]
-    agent_pos = env_state[:, :2]
-
-    action = zarr_data["action"][:]
-    image = zarr_data["img"]  # (b, h, w, c)
-
-    if image.dtype == np.float32 and image.max() == np.float32(255):
-        # HACK: images are loaded as float32 but they actually encode uint8 data
-        image = image.astype(np.uint8)
-
-    episode_data_index = {
-        "from": np.concatenate(([0], zarr_data.meta["episode_ends"][:-1])),
-        "to": zarr_data.meta["episode_ends"],
-    }
-
-    # Calculate success and reward based on the overlapping area
-    # of the T-object and the T-area.
-    coverage, keypoints = calculate_coverage(zarr_data)
-    success = calculate_success(coverage, success_threshold=0.95)
-    reward = calculate_reward(coverage, success_threshold=0.95)
-
-    features = build_features(mode)
-    dataset = LeRobotDataset.create(
-        repo_id=repo_id,
-        fps=10,
-        robot_type="2d pointer",
-        features=features,
-        image_writer_threads=4,
-    )
-    episodes = range(len(episode_data_index["from"]))
-    for ep_idx in episodes:
-        from_idx = episode_data_index["from"][ep_idx]
-        to_idx = episode_data_index["to"][ep_idx]
-        num_frames = to_idx - from_idx
-
-        for frame_idx in range(num_frames):
-            i = from_idx + frame_idx
-            idx = i + (frame_idx < num_frames - 1)
-            frame = {
-                "action": action[i],
-                # Shift reward and success by +1 until the last item of the episode
-                "next.reward": reward[idx : idx + 1],
-                "next.success": success[idx : idx + 1],
-                "task": PUSHT_TASK,
-            }
-
-            frame["observation.state"] = agent_pos[i]
-
-            if mode == "keypoints":
-                frame["observation.environment_state"] = keypoints[i]
-            else:
-                frame["observation.image"] = image[i]
-
-            dataset.add_frame(frame)
-
-        dataset.save_episode()
-
-    if push_to_hub:
-        dataset.push_to_hub()
-        hub_api = HfApi()
-        hub_api.create_tag(repo_id, tag=CODEBASE_VERSION, repo_type="dataset")
-
-
-if __name__ == "__main__":
-    # To try this script, modify the repo id with your own HuggingFace user (e.g cadene/pusht)
-    repo_id = "lerobot/pusht"
-
-    modes = ["video", "image", "keypoints"]
-    # Uncomment if you want to try with a specific mode
-    # modes = ["video"]
-    # modes = ["image"]
-    # modes = ["keypoints"]
-
-    raw_dir = Path("data/lerobot-raw/pusht_raw")
-    for mode in modes:
-        if mode in ["image", "keypoints"]:
-            repo_id += f"_{mode}"
-
-        # download and load raw dataset, create LeRobotDataset, populate it, push to hub
-        main(raw_dir, repo_id=repo_id, mode=mode)
-
-        # Uncomment if you want to load the local dataset and explore it
-        # dataset = LeRobotDataset(repo_id=repo_id)
-        # breakpoint()

examples/robots/lekiwi_client_app.py

@@ -0,0 +1,98 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+import time
+
+from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
+from lerobot.common.robots.lekiwi.config_lekiwi import LeKiwiClientConfig
+from lerobot.common.robots.lekiwi.lekiwi_client import OBS_STATE, LeKiwiClient
+from lerobot.common.teleoperators.keyboard import KeyboardTeleop, KeyboardTeleopConfig
+from lerobot.common.teleoperators.so100 import SO100Leader, SO100LeaderConfig
+
+NB_CYCLES_CLIENT_CONNECTION = 250
+
+
+def main():
+    logging.info("Configuring Teleop Devices")
+    leader_arm_config = SO100LeaderConfig(port="/dev/tty.usbmodem58760434171")
+    leader_arm = SO100Leader(leader_arm_config)
+
+    keyboard_config = KeyboardTeleopConfig()
+    keyboard = KeyboardTeleop(keyboard_config)
+
+    logging.info("Configuring LeKiwi Client")
+    robot_config = LeKiwiClientConfig(remote_ip="192.0.2.42", id="lekiwi")
+    robot = LeKiwiClient(robot_config)
+
+    logging.info("Creating LeRobot Dataset")
+
+    # The observations that we get are expected to be in body frame (x,y,theta)
+    obs_dict = {f"{OBS_STATE}." + key: value for key, value in robot.state_feature.items()}
+    # The actions that we send are expected to be in wheel frame (motor encoders)
+    act_dict = {"action." + key: value for key, value in robot.action_feature.items()}
+
+    features_dict = {
+        **act_dict,
+        **obs_dict,
+        **robot.camera_features,
+    }
+    dataset = LeRobotDataset.create(
+        repo_id="user/lekiwi" + str(int(time.time())),
+        fps=10,
+        features=features_dict,
+    )
+
+    logging.info("Connecting Teleop Devices")
+    leader_arm.connect()
+    keyboard.connect()
+
+    logging.info("Connecting remote LeKiwi")
+    robot.connect()
+
+    if not robot.is_connected or not leader_arm.is_connected or not keyboard.is_connected:
+        logging.error("Failed to connect to all devices")
+        return
+
+    logging.info("Starting LeKiwi teleoperation")
+    i = 0
+    while i < NB_CYCLES_CLIENT_CONNECTION:
+        arm_action = leader_arm.get_action()
+        base_action = keyboard.get_action()
+        action = {**arm_action, **base_action} if len(base_action) > 0 else arm_action
+
+        action_sent = robot.send_action(action)
+        observation = robot.get_observation()
+
+        frame = {**action_sent, **observation}
+        frame.update({"task": "Dummy Example Task Dataset"})
+
+        logging.info("Saved a frame into the dataset")
+        dataset.add_frame(frame)
+        i += 1
+
+    logging.info("Disconnecting Teleop Devices and LeKiwi Client")
+    robot.disconnect()
+    leader_arm.disconnect()
+    keyboard.disconnect()
+
+    logging.info("Uploading dataset to the hub")
+    dataset.save_episode()
+    dataset.push_to_hub()
+
+    logging.info("Finished LeKiwi cleanly")
+
+
+if __name__ == "__main__":
+    main()

lerobot/__init__.py

@@ -181,7 +181,7 @@ available_robots = [
     "koch_bimanual",
     "aloha",
     "so100",
-    "moss",
+    "so101",
 ]
 
 # lists all available cameras from `lerobot/common/robot_devices/cameras`

lerobot/calibrate.py

@@ -0,0 +1,84 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Helper to recalibrate your device (robot or teleoperator).
+
+Example:
+
+```shell
+python -m lerobot.calibrate \
+    --teleop.type=so100_leader \
+    --teleop.port=/dev/tty.usbmodem58760431551 \
+    --teleop.id=blue
+```
+"""
+
+import logging
+from dataclasses import asdict, dataclass
+from pprint import pformat
+
+import draccus
+
+from lerobot.common.cameras.opencv.configuration_opencv import OpenCVCameraConfig  # noqa: F401
+from lerobot.common.cameras.realsense.configuration_realsense import RealSenseCameraConfig  # noqa: F401
+from lerobot.common.robots import (  # noqa: F401
+    Robot,
+    RobotConfig,
+    koch_follower,
+    lekiwi,
+    make_robot_from_config,
+    so100_follower,
+    so101_follower,
+)
+from lerobot.common.teleoperators import (  # noqa: F401
+    Teleoperator,
+    TeleoperatorConfig,
+    koch_leader,
+    make_teleoperator_from_config,
+    so100_leader,
+    so101_leader,
+)
+from lerobot.common.utils.utils import init_logging
+
+
+@dataclass
+class CalibrateConfig:
+    teleop: TeleoperatorConfig | None = None
+    robot: RobotConfig | None = None
+
+    def __post_init__(self):
+        if bool(self.teleop) == bool(self.robot):
+            raise ValueError("Choose either a teleop or a robot.")
+
+        self.device = self.robot if self.robot else self.teleop
+
+
+@draccus.wrap()
+def calibrate(cfg: CalibrateConfig):
+    init_logging()
+    logging.info(pformat(asdict(cfg)))
+
+    if isinstance(cfg.device, RobotConfig):
+        device = make_robot_from_config(cfg.device)
+    elif isinstance(cfg.device, TeleoperatorConfig):
+        device = make_teleoperator_from_config(cfg.device)
+
+    device.connect(calibrate=False)
+    device.calibrate()
+    device.disconnect()
+
+
+if __name__ == "__main__":
+    calibrate()

lerobot/common/cameras/__init__.py

@@ -1,4 +1,17 @@
-from .camera import Camera
-from .configs import CameraConfig
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
 
-__all__ = ["Camera", "CameraConfig"]
+from .camera import Camera
+from .configs import CameraConfig, ColorMode, Cv2Rotation
+from .utils import make_cameras_from_configs

lerobot/common/cameras/camera.py

@@ -1,25 +1,122 @@
+#!/usr/bin/env python
+
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import abc
+from typing import Any, Dict, List
 
 import numpy as np
 
+from .configs import CameraConfig, ColorMode
+
 
 class Camera(abc.ABC):
+    """Base class for camera implementations.
+
+    Defines a standard interface for camera operations across different backends.
+    Subclasses must implement all abstract methods.
+
+    Manages basic camera properties (FPS, resolution) and core operations:
+    - Connection/disconnection
+    - Frame capture (sync/async)
+
+    Attributes:
+        fps (int | None): Configured frames per second
+        width (int | None): Frame width in pixels
+        height (int | None): Frame height in pixels
+        warmup_time (int | None): Time reading frames before returning from connect (in seconds)
+
+    Example:
+        class MyCamera(Camera):
+            def __init__(self, config): ...
+            @property
+            def is_connected(self) -> bool: ...
+            def connect(self, warmup=True): ...
+            # Plus other required methods
+    """
+
+    def __init__(self, config: CameraConfig):
+        """Initialize the camera with the given configuration.
+
+        Args:
+            config: Camera configuration containing FPS and resolution.
+        """
+        self.fps: int | None = config.fps
+        self.width: int | None = config.width
+        self.height: int | None = config.height
+        self.warmup_time: int | None = config.warmup_time
+
+    @property
     @abc.abstractmethod
-    def connect(self):
+    def is_connected(self) -> bool:
+        """Check if the camera is currently connected.
+
+        Returns:
+            bool: True if the camera is connected and ready to capture frames,
+                  False otherwise.
+        """
+        pass
+
+    @staticmethod
+    @abc.abstractmethod
+    def find_cameras() -> List[Dict[str, Any]]:
+        """Detects available cameras connected to the system.
+        Returns:
+            List[Dict[str, Any]]: A list of dictionaries,
+            where each dictionary contains information about a detected camera.
+        """
         pass
 
     @abc.abstractmethod
-    def read(self, temporary_color: str | None = None) -> np.ndarray:
+    def connect(self, warmup: bool = True) -> None:
+        """Establish connection to the camera.
+
+        Args:
+            warmup: If True (default), captures a warmup frame before returning. Useful
+                   for cameras that require time to adjust capture settings.
+                   If False, skips the warmup frame.
+        """
         pass
 
     @abc.abstractmethod
-    def async_read(self) -> np.ndarray:
+    def read(self, color_mode: ColorMode | None = None) -> np.ndarray:
+        """Capture and return a single frame from the camera.
+
+        Args:
+            color_mode: Desired color mode for the output frame. If None,
+                        uses the camera's default color mode.
+
+        Returns:
+            np.ndarray: Captured frame as a numpy array.
+        """
         pass
 
     @abc.abstractmethod
-    def disconnect(self):
+    def async_read(self, timeout_ms: float = ...) -> np.ndarray:
+        """Asynchronously capture and return a single frame from the camera.
+
+        Args:
+            timeout_ms: Maximum time to wait for a frame in milliseconds.
+                        Defaults to implementation-specific timeout.
+
+        Returns:
+            np.ndarray: Captured frame as a numpy array.
+        """
         pass
 
-    def __del__(self):
-        if getattr(self, "is_connected", False):
-            self.disconnect()
+    @abc.abstractmethod
+    def disconnect(self) -> None:
+        """Disconnect from the camera and release resources."""
+        pass

lerobot/common/cameras/configs.py

@@ -1,11 +1,45 @@
+#!/usr/bin/env python
+
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 import abc
 from dataclasses import dataclass
+from enum import Enum
 
 import draccus
 
 
-@dataclass
+class ColorMode(str, Enum):
+    RGB = "rgb"
+    BGR = "bgr"
+
+
+class Cv2Rotation(int, Enum):
+    NO_ROTATION = 0
+    ROTATE_90 = 90
+    ROTATE_180 = 180
+    ROTATE_270 = -90
+
+
+@dataclass(kw_only=True)
 class CameraConfig(draccus.ChoiceRegistry, abc.ABC):
+    fps: int | None = None
+    width: int | None = None
+    height: int | None = None
+    warmup_time: int | None = None
+
     @property
     def type(self) -> str:
         return self.get_choice_name(self.__class__)

lerobot/common/cameras/intel/__init__.py

@@ -1,4 +0,0 @@
-from .camera_realsense import RealSenseCamera
-from .configuration_realsense import RealSenseCameraConfig
-
-__all__ = ["RealSenseCamera", "RealSenseCameraConfig"]

lerobot/common/cameras/intel/camera_realsense.py

@@ -1,535 +0,0 @@
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-"""
-This file contains utilities for recording frames from Intel Realsense cameras.
-"""
-
-import argparse
-import concurrent.futures
-import logging
-import math
-import shutil
-import threading
-import time
-import traceback
-from collections import Counter
-from pathlib import Path
-from threading import Thread
-
-import numpy as np
-from PIL import Image
-
-from lerobot.common.errors import DeviceAlreadyConnectedError, DeviceNotConnectedError
-from lerobot.common.utils.robot_utils import (
-    busy_wait,
-)
-from lerobot.common.utils.utils import capture_timestamp_utc
-
-from ..camera import Camera
-from .configuration_realsense import RealSenseCameraConfig
-
-SERIAL_NUMBER_INDEX = 1
-
-
-def find_cameras(raise_when_empty=True, mock=False) -> list[dict]:
-    """
-    Find the names and the serial numbers of the Intel RealSense cameras
-    connected to the computer.
-    """
-    if mock:
-        import tests.cameras.mock_pyrealsense2 as rs
-    else:
-        import pyrealsense2 as rs
-
-    cameras = []
-    for device in rs.context().query_devices():
-        serial_number = int(device.get_info(rs.camera_info(SERIAL_NUMBER_INDEX)))
-        name = device.get_info(rs.camera_info.name)
-        cameras.append(
-            {
-                "serial_number": serial_number,
-                "name": name,
-            }
-        )
-
-    if raise_when_empty and len(cameras) == 0:
-        raise OSError(
-            "Not a single camera was detected. Try re-plugging, or re-installing `librealsense` and its python wrapper `pyrealsense2`, or updating the firmware."
-        )
-
-    return cameras
-
-
-def save_image(img_array, serial_number, frame_index, images_dir):
-    try:
-        img = Image.fromarray(img_array)
-        path = images_dir / f"camera_{serial_number}_frame_{frame_index:06d}.png"
-        path.parent.mkdir(parents=True, exist_ok=True)
-        img.save(str(path), quality=100)
-        logging.info(f"Saved image: {path}")
-    except Exception as e:
-        logging.error(f"Failed to save image for camera {serial_number} frame {frame_index}: {e}")
-
-
-def save_images_from_cameras(
-    images_dir: Path,
-    serial_numbers: list[int] | None = None,
-    fps=None,
-    width=None,
-    height=None,
-    record_time_s=2,
-    mock=False,
-):
-    """
-    Initializes all the cameras and saves images to the directory. Useful to visually identify the camera
-    associated to a given serial number.
-    """
-    if serial_numbers is None or len(serial_numbers) == 0:
-        camera_infos = find_cameras(mock=mock)
-        serial_numbers = [cam["serial_number"] for cam in camera_infos]
-
-    if mock:
-        import tests.cameras.mock_cv2 as cv2
-    else:
-        import cv2
-
-    print("Connecting cameras")
-    cameras = []
-    for cam_sn in serial_numbers:
-        print(f"{cam_sn=}")
-        config = RealSenseCameraConfig(serial_number=cam_sn, fps=fps, width=width, height=height, mock=mock)
-        camera = RealSenseCamera(config)
-        camera.connect()
-        print(
-            f"RealSenseCamera({camera.serial_number}, fps={camera.fps}, width={camera.capture_width}, height={camera.capture_height}, color_mode={camera.color_mode})"
-        )
-        cameras.append(camera)
-
-    images_dir = Path(images_dir)
-    if images_dir.exists():
-        shutil.rmtree(
-            images_dir,
-        )
-    images_dir.mkdir(parents=True, exist_ok=True)
-
-    print(f"Saving images to {images_dir}")
-    frame_index = 0
-    start_time = time.perf_counter()
-    try:
-        with concurrent.futures.ThreadPoolExecutor(max_workers=1) as executor:
-            while True:
-                now = time.perf_counter()
-
-                for camera in cameras:
-                    # If we use async_read when fps is None, the loop will go full speed, and we will end up
-                    # saving the same images from the cameras multiple times until the RAM/disk is full.
-                    image = camera.read() if fps is None else camera.async_read()
-                    if image is None:
-                        print("No Frame")
-
-                    bgr_converted_image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
-
-                    executor.submit(
-                        save_image,
-                        bgr_converted_image,
-                        camera.serial_number,
-                        frame_index,
-                        images_dir,
-                    )
-
-                if fps is not None:
-                    dt_s = time.perf_counter() - now
-                    busy_wait(1 / fps - dt_s)
-
-                if time.perf_counter() - start_time > record_time_s:
-                    break
-
-                print(f"Frame: {frame_index:04d}\tLatency (ms): {(time.perf_counter() - now) * 1000:.2f}")
-
-                frame_index += 1
-    finally:
-        print(f"Images have been saved to {images_dir}")
-        for camera in cameras:
-            camera.disconnect()
-
-
-class RealSenseCamera(Camera):
-    """
-    The RealSenseCamera class is similar to OpenCVCamera class but adds additional features for Intel Real Sense cameras:
-    - is instantiated with the serial number of the camera - won't randomly change as it can be the case of OpenCVCamera for Linux,
-    - can also be instantiated with the camera's name — if it's unique — using RealSenseCamera.init_from_name(),
-    - depth map can be returned.
-
-    To find the camera indices of your cameras, you can run our utility script that will save a few frames for each camera:
-    ```bash
-    python lerobot/common/robot_devices/cameras/intelrealsense.py --images-dir outputs/images_from_intelrealsense_cameras
-    ```
-
-    When an RealSenseCamera is instantiated, if no specific config is provided, the default fps, width, height and color_mode
-    of the given camera will be used.
-
-    Example of instantiating with a serial number:
-    ```python
-    from lerobot.common.robot_devices.cameras.configs import RealSenseCameraConfig
-
-    config = RealSenseCameraConfig(serial_number=128422271347)
-    camera = RealSenseCamera(config)
-    camera.connect()
-    color_image = camera.read()
-    # when done using the camera, consider disconnecting
-    camera.disconnect()
-    ```
-
-    Example of instantiating with a name if it's unique:
-    ```
-    config = RealSenseCameraConfig(name="Intel RealSense D405")
-    ```
-
-    Example of changing default fps, width, height and color_mode:
-    ```python
-    config = RealSenseCameraConfig(serial_number=128422271347, fps=30, width=1280, height=720)
-    config = RealSenseCameraConfig(serial_number=128422271347, fps=90, width=640, height=480)
-    config = RealSenseCameraConfig(serial_number=128422271347, fps=90, width=640, height=480, color_mode="bgr")
-    # Note: might error out upon `camera.connect()` if these settings are not compatible with the camera
-    ```
-
-    Example of returning depth:
-    ```python
-    config = RealSenseCameraConfig(serial_number=128422271347, use_depth=True)
-    camera = RealSenseCamera(config)
-    camera.connect()
-    color_image, depth_map = camera.read()
-    ```
-    """
-
-    def __init__(
-        self,
-        config: RealSenseCameraConfig,
-    ):
-        self.config = config
-        if config.name is not None:
-            self.serial_number = self.find_serial_number_from_name(config.name)
-        else:
-            self.serial_number = config.serial_number
-
-        # Store the raw (capture) resolution from the config.
-        self.capture_width = config.width
-        self.capture_height = config.height
-
-        # If rotated by ±90, swap width and height.
-        if config.rotation in [-90, 90]:
-            self.width = config.height
-            self.height = config.width
-        else:
-            self.width = config.width
-            self.height = config.height
-
-        self.fps = config.fps
-        self.channels = config.channels
-        self.color_mode = config.color_mode
-        self.use_depth = config.use_depth
-        self.force_hardware_reset = config.force_hardware_reset
-        self.mock = config.mock
-
-        self.camera = None
-        self.is_connected = False
-        self.thread = None
-        self.stop_event = None
-        self.color_image = None
-        self.depth_map = None
-        self.logs = {}
-
-        if self.mock:
-            import tests.cameras.mock_cv2 as cv2
-        else:
-            import cv2
-
-        self.rotation = None
-        if config.rotation == -90:
-            self.rotation = cv2.ROTATE_90_COUNTERCLOCKWISE
-        elif config.rotation == 90:
-            self.rotation = cv2.ROTATE_90_CLOCKWISE
-        elif config.rotation == 180:
-            self.rotation = cv2.ROTATE_180
-
-    def find_serial_number_from_name(self, name):
-        camera_infos = find_cameras()
-        camera_names = [cam["name"] for cam in camera_infos]
-        this_name_count = Counter(camera_names)[name]
-        if this_name_count > 1:
-            # TODO(aliberts): Test this with multiple identical cameras (Aloha)
-            raise ValueError(
-                f"Multiple {name} cameras have been detected. Please use their serial number to instantiate them."
-            )
-
-        name_to_serial_dict = {cam["name"]: cam["serial_number"] for cam in camera_infos}
-        cam_sn = name_to_serial_dict[name]
-
-        return cam_sn
-
-    def connect(self):
-        if self.is_connected:
-            raise DeviceAlreadyConnectedError(f"RealSenseCamera({self.serial_number}) is already connected.")
-
-        if self.mock:
-            import tests.cameras.mock_pyrealsense2 as rs
-        else:
-            import pyrealsense2 as rs
-
-        config = rs.config()
-        config.enable_device(str(self.serial_number))
-
-        if self.fps and self.capture_width and self.capture_height:
-            # TODO(rcadene): can we set rgb8 directly?
-            config.enable_stream(
-                rs.stream.color, self.capture_width, self.capture_height, rs.format.rgb8, self.fps
-            )
-        else:
-            config.enable_stream(rs.stream.color)
-
-        if self.use_depth:
-            if self.fps and self.capture_width and self.capture_height:
-                config.enable_stream(
-                    rs.stream.depth, self.capture_width, self.capture_height, rs.format.z16, self.fps
-                )
-            else:
-                config.enable_stream(rs.stream.depth)
-
-        self.camera = rs.pipeline()
-        try:
-            profile = self.camera.start(config)
-            is_camera_open = True
-        except RuntimeError:
-            is_camera_open = False
-            traceback.print_exc()
-
-        # If the camera doesn't work, display the camera indices corresponding to
-        # valid cameras.
-        if not is_camera_open:
-            # Verify that the provided `serial_number` is valid before printing the traceback
-            camera_infos = find_cameras()
-            serial_numbers = [cam["serial_number"] for cam in camera_infos]
-            if self.serial_number not in serial_numbers:
-                raise ValueError(
-                    f"`serial_number` is expected to be one of these available cameras {serial_numbers}, but {self.serial_number} is provided instead. "
-                    "To find the serial number you should use, run `python lerobot/common/robot_devices/cameras/intelrealsense.py`."
-                )
-
-            raise OSError(f"Can't access RealSenseCamera({self.serial_number}).")
-
-        color_stream = profile.get_stream(rs.stream.color)
-        color_profile = color_stream.as_video_stream_profile()
-        actual_fps = color_profile.fps()
-        actual_width = color_profile.width()
-        actual_height = color_profile.height()
-
-        # Using `math.isclose` since actual fps can be a float (e.g. 29.9 instead of 30)
-        if self.fps is not None and not math.isclose(self.fps, actual_fps, rel_tol=1e-3):
-            # Using `OSError` since it's a broad that encompasses issues related to device communication
-            raise OSError(
-                f"Can't set {self.fps=} for RealSenseCamera({self.serial_number}). Actual value is {actual_fps}."
-            )
-        if self.capture_width is not None and self.capture_width != actual_width:
-            raise OSError(
-                f"Can't set {self.capture_width=} for RealSenseCamera({self.serial_number}). Actual value is {actual_width}."
-            )
-        if self.capture_height is not None and self.capture_height != actual_height:
-            raise OSError(
-                f"Can't set {self.capture_height=} for RealSenseCamera({self.serial_number}). Actual value is {actual_height}."
-            )
-
-        self.fps = round(actual_fps)
-        self.capture_width = round(actual_width)
-        self.capture_height = round(actual_height)
-
-        self.is_connected = True
-
-    def read(self, temporary_color: str | None = None) -> np.ndarray | tuple[np.ndarray, np.ndarray]:
-        """Read a frame from the camera returned in the format height x width x channels (e.g. 480 x 640 x 3)
-        of type `np.uint8`, contrarily to the pytorch format which is float channel first.
-
-        When `use_depth=True`, returns a tuple `(color_image, depth_map)` with a depth map in the format
-        height x width (e.g. 480 x 640) of type np.uint16.
-
-        Note: Reading a frame is done every `camera.fps` times per second, and it is blocking.
-        If you are reading data from other sensors, we advise to use `camera.async_read()` which is non blocking version of `camera.read()`.
-        """
-        if not self.is_connected:
-            raise DeviceNotConnectedError(
-                f"RealSenseCamera({self.serial_number}) is not connected. Try running `camera.connect()` first."
-            )
-
-        if self.mock:
-            import tests.cameras.mock_cv2 as cv2
-        else:
-            import cv2
-
-        start_time = time.perf_counter()
-
-        frame = self.camera.wait_for_frames(timeout_ms=5000)
-
-        color_frame = frame.get_color_frame()
-
-        if not color_frame:
-            raise OSError(f"Can't capture color image from RealSenseCamera({self.serial_number}).")
-
-        color_image = np.asanyarray(color_frame.get_data())
-
-        requested_color_mode = self.color_mode if temporary_color is None else temporary_color
-        if requested_color_mode not in ["rgb", "bgr"]:
-            raise ValueError(
-                f"Expected color values are 'rgb' or 'bgr', but {requested_color_mode} is provided."
-            )
-
-        # IntelRealSense uses RGB format as default (red, green, blue).
-        if requested_color_mode == "bgr":
-            color_image = cv2.cvtColor(color_image, cv2.COLOR_RGB2BGR)
-
-        h, w, _ = color_image.shape
-        if h != self.capture_height or w != self.capture_width:
-            raise OSError(
-                f"Can't capture color image with expected height and width ({self.height} x {self.width}). ({h} x {w}) returned instead."
-            )
-
-        if self.rotation is not None:
-            color_image = cv2.rotate(color_image, self.rotation)
-
-        # log the number of seconds it took to read the image
-        self.logs["delta_timestamp_s"] = time.perf_counter() - start_time
-
-        # log the utc time at which the image was received
-        self.logs["timestamp_utc"] = capture_timestamp_utc()
-
-        if self.use_depth:
-            depth_frame = frame.get_depth_frame()
-            if not depth_frame:
-                raise OSError(f"Can't capture depth image from RealSenseCamera({self.serial_number}).")
-
-            depth_map = np.asanyarray(depth_frame.get_data())
-
-            h, w = depth_map.shape
-            if h != self.capture_height or w != self.capture_width:
-                raise OSError(
-                    f"Can't capture depth map with expected height and width ({self.height} x {self.width}). ({h} x {w}) returned instead."
-                )
-
-            if self.rotation is not None:
-                depth_map = cv2.rotate(depth_map, self.rotation)
-
-            return color_image, depth_map
-        else:
-            return color_image
-
-    def read_loop(self):
-        while not self.stop_event.is_set():
-            if self.use_depth:
-                self.color_image, self.depth_map = self.read()
-            else:
-                self.color_image = self.read()
-
-    def async_read(self):
-        """Access the latest color image"""
-        if not self.is_connected:
-            raise DeviceNotConnectedError(
-                f"RealSenseCamera({self.serial_number}) is not connected. Try running `camera.connect()` first."
-            )
-
-        if self.thread is None:
-            self.stop_event = threading.Event()
-            self.thread = Thread(target=self.read_loop, args=())
-            self.thread.daemon = True
-            self.thread.start()
-
-        num_tries = 0
-        while self.color_image is None:
-            # TODO(rcadene, aliberts): intelrealsense has diverged compared to opencv over here
-            num_tries += 1
-            time.sleep(1 / self.fps)
-            if num_tries > self.fps and (self.thread.ident is None or not self.thread.is_alive()):
-                raise Exception(
-                    "The thread responsible for `self.async_read()` took too much time to start. There might be an issue. Verify that `self.thread.start()` has been called."
-                )
-
-        if self.use_depth:
-            return self.color_image, self.depth_map
-        else:
-            return self.color_image
-
-    def disconnect(self):
-        if not self.is_connected:
-            raise DeviceNotConnectedError(
-                f"RealSenseCamera({self.serial_number}) is not connected. Try running `camera.connect()` first."
-            )
-
-        if self.thread is not None and self.thread.is_alive():
-            # wait for the thread to finish
-            self.stop_event.set()
-            self.thread.join()
-            self.thread = None
-            self.stop_event = None
-
-        self.camera.stop()
-        self.camera = None
-
-        self.is_connected = False
-
-    def __del__(self):
-        if getattr(self, "is_connected", False):
-            self.disconnect()
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(
-        description="Save a few frames using `RealSenseCamera` for all cameras connected to the computer, or a selected subset."
-    )
-    parser.add_argument(
-        "--serial-numbers",
-        type=int,
-        nargs="*",
-        default=None,
-        help="List of serial numbers used to instantiate the `RealSenseCamera`. If not provided, find and use all available camera indices.",
-    )
-    parser.add_argument(
-        "--fps",
-        type=int,
-        default=30,
-        help="Set the number of frames recorded per seconds for all cameras. If not provided, use the default fps of each camera.",
-    )
-    parser.add_argument(
-        "--width",
-        type=str,
-        default=640,
-        help="Set the width for all cameras. If not provided, use the default width of each camera.",
-    )
-    parser.add_argument(
-        "--height",
-        type=str,
-        default=480,
-        help="Set the height for all cameras. If not provided, use the default height of each camera.",
-    )
-    parser.add_argument(
-        "--images-dir",
-        type=Path,
-        default="outputs/images_from_intelrealsense_cameras",
-        help="Set directory to save a few frames for each camera.",
-    )
-    parser.add_argument(
-        "--record-time-s",
-        type=float,
-        default=2.0,
-        help="Set the number of seconds used to record the frames. By default, 2 seconds.",
-    )
-    args = parser.parse_args()
-    save_images_from_cameras(**vars(args))

lerobot/common/cameras/intel/configuration_realsense.py

@@ -1,71 +0,0 @@
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from dataclasses import dataclass
-
-from ..configs import CameraConfig
-
-
-@CameraConfig.register_subclass("intelrealsense")
-@dataclass
-class RealSenseCameraConfig(CameraConfig):
-    """
-    Example of tested options for Intel Real Sense D405:
-
-    ```python
-    RealSenseCameraConfig(128422271347, 30, 640, 480)
-    RealSenseCameraConfig(128422271347, 60, 640, 480)
-    RealSenseCameraConfig(128422271347, 90, 640, 480)
-    RealSenseCameraConfig(128422271347, 30, 1280, 720)
-    RealSenseCameraConfig(128422271347, 30, 640, 480, use_depth=True)
-    RealSenseCameraConfig(128422271347, 30, 640, 480, rotation=90)
-    ```
-    """
-
-    name: str | None = None
-    serial_number: int | None = None
-    fps: int | None = None
-    width: int | None = None
-    height: int | None = None
-    color_mode: str = "rgb"
-    channels: int | None = None
-    use_depth: bool = False
-    force_hardware_reset: bool = True
-    rotation: int | None = None
-    mock: bool = False
-
-    def __post_init__(self):
-        # bool is stronger than is None, since it works with empty strings
-        if bool(self.name) and bool(self.serial_number):
-            raise ValueError(
-                f"One of them must be set: name or serial_number, but {self.name=} and {self.serial_number=} provided."
-            )
-
-        if self.color_mode not in ["rgb", "bgr"]:
-            raise ValueError(
-                f"`color_mode` is expected to be 'rgb' or 'bgr', but {self.color_mode} is provided."
-            )
-
-        self.channels = 3
-
-        at_least_one_is_not_none = self.fps is not None or self.width is not None or self.height is not None
-        at_least_one_is_none = self.fps is None or self.width is None or self.height is None
-        if at_least_one_is_not_none and at_least_one_is_none:
-            raise ValueError(
-                "For `fps`, `width` and `height`, either all of them need to be set, or none of them, "
-                f"but {self.fps=}, {self.width=}, {self.height=} were provided."
-            )
-
-        if self.rotation not in [-90, None, 90, 180]:
-            raise ValueError(f"`rotation` must be in [-90, None, 90, 180] (got {self.rotation})")

lerobot/common/cameras/opencv/__init__.py

@@ -1,4 +1,16 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 from .camera_opencv import OpenCVCamera
 from .configuration_opencv import OpenCVCameraConfig
-
-__all__ = ["OpenCVCamera", "OpenCVCameraConfig"]

lerobot/common/cameras/opencv/camera_opencv.py

@@ -13,507 +13,501 @@
 # limitations under the License.
 
 """
-This file contains utilities for recording frames from cameras. For more info look at `OpenCVCamera` docstring.
+Provides the OpenCVCamera class for capturing frames from cameras using OpenCV.
 """
 
-import argparse
-import concurrent.futures
+import contextlib
+import logging
 import math
 import platform
-import shutil
-import threading
+import queue
 import time
 from pathlib import Path
-from threading import Thread
+from threading import Event, Thread
+from typing import Any, Dict, List
 
+import cv2
 import numpy as np
-from PIL import Image
 
 from lerobot.common.errors import DeviceAlreadyConnectedError, DeviceNotConnectedError
-from lerobot.common.utils.robot_utils import (
-    busy_wait,
-)
-from lerobot.common.utils.utils import capture_timestamp_utc
 
 from ..camera import Camera
-from .configuration_opencv import OpenCVCameraConfig
+from ..utils import get_cv2_backend, get_cv2_rotation
+from .configuration_opencv import ColorMode, OpenCVCameraConfig
 
-# The maximum opencv device index depends on your operating system. For instance,
+# NOTE(Steven): The maximum opencv device index depends on your operating system. For instance,
 # if you have 3 cameras, they should be associated to index 0, 1, and 2. This is the case
 # on MacOS. However, on Ubuntu, the indices are different like 6, 16, 23.
 # When you change the USB port or reboot the computer, the operating system might
 # treat the same cameras as new devices. Thus we select a higher bound to search indices.
 MAX_OPENCV_INDEX = 60
 
-
-def find_cameras(raise_when_empty=False, max_index_search_range=MAX_OPENCV_INDEX, mock=False) -> list[dict]:
-    cameras = []
-    if platform.system() == "Linux":
-        print("Linux detected. Finding available camera indices through scanning '/dev/video*' ports")
-        possible_ports = [str(port) for port in Path("/dev").glob("video*")]
-        ports = _find_cameras(possible_ports, mock=mock)
-        for port in ports:
-            cameras.append(
-                {
-                    "port": port,
-                    "index": int(port.removeprefix("/dev/video")),
-                }
-            )
-    else:
-        print(
-            "Mac or Windows detected. Finding available camera indices through "
-            f"scanning all indices from 0 to {MAX_OPENCV_INDEX}"
-        )
-        possible_indices = range(max_index_search_range)
-        indices = _find_cameras(possible_indices, mock=mock)
-        for index in indices:
-            cameras.append(
-                {
-                    "port": None,
-                    "index": index,
-                }
-            )
-
-    return cameras
-
-
-def _find_cameras(
-    possible_camera_ids: list[int | str], raise_when_empty=False, mock=False
-) -> list[int | str]:
-    if mock:
-        import tests.cameras.mock_cv2 as cv2
-    else:
-        import cv2
-
-    camera_ids = []
-    for camera_idx in possible_camera_ids:
-        camera = cv2.VideoCapture(camera_idx)
-        is_open = camera.isOpened()
-        camera.release()
-
-        if is_open:
-            print(f"Camera found at index {camera_idx}")
-            camera_ids.append(camera_idx)
-
-    if raise_when_empty and len(camera_ids) == 0:
-        raise OSError(
-            "Not a single camera was detected. Try re-plugging, or re-installing `opencv2`, "
-            "or your camera driver, or make sure your camera is compatible with opencv2."
-        )
-
-    return camera_ids
-
-
-def is_valid_unix_path(path: str) -> bool:
-    """Note: if 'path' points to a symlink, this will return True only if the target exists"""
-    p = Path(path)
-    return p.is_absolute() and p.exists()
-
-
-def get_camera_index_from_unix_port(port: Path) -> int:
-    return int(str(port.resolve()).removeprefix("/dev/video"))
-
-
-def save_image(img_array, camera_index, frame_index, images_dir):
-    img = Image.fromarray(img_array)
-    path = images_dir / f"camera_{camera_index:02d}_frame_{frame_index:06d}.png"
-    path.parent.mkdir(parents=True, exist_ok=True)
-    img.save(str(path), quality=100)
-
-
-def save_images_from_cameras(
-    images_dir: Path,
-    camera_ids: list | None = None,
-    fps=None,
-    width=None,
-    height=None,
-    record_time_s=2,
-    mock=False,
-):
-    """
-    Initializes all the cameras and saves images to the directory. Useful to visually identify the camera
-    associated to a given camera index.
-    """
-    if camera_ids is None or len(camera_ids) == 0:
-        camera_infos = find_cameras(mock=mock)
-        camera_ids = [cam["index"] for cam in camera_infos]
-
-    print("Connecting cameras")
-    cameras = []
-    for cam_idx in camera_ids:
-        config = OpenCVCameraConfig(camera_index=cam_idx, fps=fps, width=width, height=height, mock=mock)
-        camera = OpenCVCamera(config)
-        camera.connect()
-        print(
-            f"OpenCVCamera({camera.camera_index}, fps={camera.fps}, width={camera.capture_width}, "
-            f"height={camera.capture_height}, color_mode={camera.color_mode})"
-        )
-        cameras.append(camera)
-
-    images_dir = Path(images_dir)
-    if images_dir.exists():
-        shutil.rmtree(
-            images_dir,
-        )
-    images_dir.mkdir(parents=True, exist_ok=True)
-
-    print(f"Saving images to {images_dir}")
-    frame_index = 0
-    start_time = time.perf_counter()
-    with concurrent.futures.ThreadPoolExecutor(max_workers=1) as executor:
-        while True:
-            now = time.perf_counter()
-
-            for camera in cameras:
-                # If we use async_read when fps is None, the loop will go full speed, and we will endup
-                # saving the same images from the cameras multiple times until the RAM/disk is full.
-                image = camera.read() if fps is None else camera.async_read()
-
-                executor.submit(
-                    save_image,
-                    image,
-                    camera.camera_index,
-                    frame_index,
-                    images_dir,
-                )
-
-            if fps is not None:
-                dt_s = time.perf_counter() - now
-                busy_wait(1 / fps - dt_s)
-
-            print(f"Frame: {frame_index:04d}\tLatency (ms): {(time.perf_counter() - now) * 1000:.2f}")
-
-            if time.perf_counter() - start_time > record_time_s:
-                break
-
-            frame_index += 1
-
-    print(f"Images have been saved to {images_dir}")
+logger = logging.getLogger(__name__)
 
 
 class OpenCVCamera(Camera):
     """
-    The OpenCVCamera class allows to efficiently record images from cameras. It relies on opencv2 to communicate
-    with the cameras. Most cameras are compatible. For more info, see the [Video I/O with OpenCV Overview](https://docs.opencv.org/4.x/d0/da7/videoio_overview.html).
+    Manages camera interactions using OpenCV for efficient frame recording.
 
-    An OpenCVCamera instance requires a camera index (e.g. `OpenCVCamera(camera_index=0)`). When you only have one camera
-    like a webcam of a laptop, the camera index is expected to be 0, but it might also be very different, and the camera index
-    might change if you reboot your computer or re-plug your camera. This behavior depends on your operation system.
+    This class provides a high-level interface to connect to, configure, and read
+    frames from cameras compatible with OpenCV's VideoCapture. It supports both
+    synchronous and asynchronous frame reading.
 
-    To find the camera indices of your cameras, you can run our utility script that will be save a few frames for each camera:
+    An OpenCVCamera instance requires a camera index (e.g., 0) or a device path
+    (e.g., '/dev/video0' on Linux). Camera indices can be unstable across reboots
+    or port changes, especially on Linux. Use the provided utility script to find
+    available camera indices or paths:
     ```bash
-    python lerobot/common/robot_devices/cameras/opencv.py --images-dir outputs/images_from_opencv_cameras
+    python -m lerobot.find_cameras opencv
     ```
 
-    When an OpenCVCamera is instantiated, if no specific config is provided, the default fps, width, height and color_mode
-    of the given camera will be used.
+    The camera's default settings (FPS, resolution, color mode) are used unless
+    overridden in the configuration.
 
-    Example of usage:
-    ```python
-    from lerobot.common.robot_devices.cameras.configs import OpenCVCameraConfig
+    Example:
+        ```python
+        from lerobot.common.cameras.opencv import OpenCVCamera
+        from lerobot.common.cameras.configuration_opencv import OpenCVCameraConfig, ColorMode, Cv2Rotation
 
-    config = OpenCVCameraConfig(camera_index=0)
-    camera = OpenCVCamera(config)
-    camera.connect()
-    color_image = camera.read()
-    # when done using the camera, consider disconnecting
-    camera.disconnect()
-    ```
+        # Basic usage with camera index 0
+        config = OpenCVCameraConfig(index_or_path=0)
+        camera = OpenCVCamera(config)
+        try:
+            camera.connect()
+            print(f"Connected to {camera}")
+            color_image = camera.read() # Synchronous read
+            print(f"Read frame shape: {color_image.shape}")
+            async_image = camera.async_read() # Asynchronous read
+            print(f"Async read frame shape: {async_image.shape}")
+        except Exception as e:
+            print(f"An error occurred: {e}")
+        finally:
+            camera.disconnect()
+            print(f"Disconnected from {camera}")
 
-    Example of changing default fps, width, height and color_mode:
-    ```python
-    config = OpenCVCameraConfig(camera_index=0, fps=30, width=1280, height=720)
-    config = OpenCVCameraConfig(camera_index=0, fps=90, width=640, height=480)
-    config = OpenCVCameraConfig(camera_index=0, fps=90, width=640, height=480, color_mode="bgr")
-    # Note: might error out open `camera.connect()` if these settings are not compatible with the camera
-    ```
+        # Example with custom settings
+        custom_config = OpenCVCameraConfig(
+            index_or_path='/dev/video0', # Or use an index
+            fps=30,
+            width=1280,
+            height=720,
+            color_mode=ColorMode.RGB,
+            rotation=Cv2Rotation.ROTATE_90
+        )
+        custom_camera = OpenCVCamera(custom_config)
+        # ... connect, read, disconnect ...
+        ```
     """
 
     def __init__(self, config: OpenCVCameraConfig):
+        """
+        Initializes the OpenCVCamera instance.
+
+        Args:
+            config: The configuration settings for the camera.
+        """
+        super().__init__(config)
+
         self.config = config
-        self.camera_index = config.camera_index
-        self.port = None
-
-        # Linux uses ports for connecting to cameras
-        if platform.system() == "Linux":
-            if isinstance(self.camera_index, int):
-                self.port = Path(f"/dev/video{self.camera_index}")
-            elif isinstance(self.camera_index, str) and is_valid_unix_path(self.camera_index):
-                self.port = Path(self.camera_index)
-                # Retrieve the camera index from a potentially symlinked path
-                self.camera_index = get_camera_index_from_unix_port(self.port)
-            else:
-                raise ValueError(f"Please check the provided camera_index: {self.camera_index}")
-
-        # Store the raw (capture) resolution from the config.
-        self.capture_width = config.width
-        self.capture_height = config.height
-
-        # If rotated by ±90, swap width and height.
-        if config.rotation in [-90, 90]:
-            self.width = config.height
-            self.height = config.width
-        else:
-            self.width = config.width
-            self.height = config.height
+        self.index_or_path = config.index_or_path
 
         self.fps = config.fps
-        self.channels = config.channels
         self.color_mode = config.color_mode
-        self.mock = config.mock
+        self.warmup_time = config.warmup_time
 
-        self.camera = None
-        self.is_connected = False
-        self.thread = None
-        self.stop_event = None
-        self.color_image = None
-        self.logs = {}
+        self.videocapture: cv2.VideoCapture | None = None
 
-        if self.mock:
-            import tests.cameras.mock_cv2 as cv2
-        else:
-            import cv2
+        self.thread: Thread | None = None
+        self.stop_event: Event | None = None
+        self.frame_queue: queue.Queue = queue.Queue(maxsize=1)
 
-        self.rotation = None
-        if config.rotation == -90:
-            self.rotation = cv2.ROTATE_90_COUNTERCLOCKWISE
-        elif config.rotation == 90:
-            self.rotation = cv2.ROTATE_90_CLOCKWISE
-        elif config.rotation == 180:
-            self.rotation = cv2.ROTATE_180
+        self.rotation: int | None = get_cv2_rotation(config.rotation)
+        self.backend: int = get_cv2_backend()
 
-    def connect(self):
+        if self.height and self.width:
+            if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+                self.capture_width, self.capture_height = self.height, self.width
+            else:
+                self.capture_width, self.capture_height = self.width, self.height
+
+    def __str__(self) -> str:
+        return f"{self.__class__.__name__}({self.index_or_path})"
+
+    @property
+    def is_connected(self) -> bool:
+        """Checks if the camera is currently connected and opened."""
+        return isinstance(self.videocapture, cv2.VideoCapture) and self.videocapture.isOpened()
+
+    def connect(self, warmup: bool = True):
+        """
+        Connects to the OpenCV camera specified in the configuration.
+
+        Initializes the OpenCV VideoCapture object, sets desired camera properties
+        (FPS, width, height), and performs initial checks.
+
+        Raises:
+            DeviceAlreadyConnectedError: If the camera is already connected.
+            ValueError: If the specified camera index/path is not found or accessible.
+            ConnectionError: If the camera is found but fails to open.
+            RuntimeError: If the camera opens but fails to apply requested FPS/resolution settings.
+        """
         if self.is_connected:
-            raise DeviceAlreadyConnectedError(f"OpenCVCamera({self.camera_index}) is already connected.")
+            raise DeviceAlreadyConnectedError(f"{self} is already connected.")
 
-        if self.mock:
-            import tests.cameras.mock_cv2 as cv2
-        else:
-            import cv2
+        # Use 1 thread for OpenCV operations to avoid potential conflicts or
+        # blocking in multi-threaded applications, especially during data collection.
+        cv2.setNumThreads(1)
 
-            # Use 1 thread to avoid blocking the main thread. Especially useful during data collection
-            # when other threads are used to save the images.
-            cv2.setNumThreads(1)
+        self.videocapture = cv2.VideoCapture(self.index_or_path, self.backend)
 
-        backend = (
-            cv2.CAP_V4L2
-            if platform.system() == "Linux"
-            else cv2.CAP_DSHOW
-            if platform.system() == "Windows"
-            else cv2.CAP_AVFOUNDATION
-            if platform.system() == "Darwin"
-            else cv2.CAP_ANY
-        )
+        if not self.videocapture.isOpened():
+            self.videocapture.release()
+            self.videocapture = None
+            raise ConnectionError(
+                f"Failed to open OpenCV camera {self.index_or_path}."
+                f"Run 'python -m lerobot.find_cameras opencv' for details about the available cameras in your system."
+            )
 
-        camera_idx = f"/dev/video{self.camera_index}" if platform.system() == "Linux" else self.camera_index
-        # First create a temporary camera trying to access `camera_index`,
-        # and verify it is a valid camera by calling `isOpened`.
-        tmp_camera = cv2.VideoCapture(camera_idx, backend)
-        is_camera_open = tmp_camera.isOpened()
-        # Release camera to make it accessible for `find_camera_indices`
-        tmp_camera.release()
-        del tmp_camera
+        self._configure_capture_settings()
 
-        # If the camera doesn't work, display the camera indices corresponding to
-        # valid cameras.
-        if not is_camera_open:
-            # Verify that the provided `camera_index` is valid before printing the traceback
-            cameras_info = find_cameras()
-            available_cam_ids = [cam["index"] for cam in cameras_info]
-            if self.camera_index not in available_cam_ids:
+        if warmup:
+            if self.warmup_time is None:
                 raise ValueError(
-                    f"`camera_index` is expected to be one of these available cameras {available_cam_ids}, but {self.camera_index} is provided instead. "
-                    "To find the camera index you should use, run `python lerobot/common/robot_devices/cameras/opencv.py`."
+                    f"Warmup time is not set for {self}. Please set a warmup time in the configuration."
                 )
+            logger.debug(f"Reading a warm-up frames for {self} for {self.warmup_time} seconds...")
+            start_time = time.time()
+            while time.time() - start_time < self.warmup_time:
+                self.read()
+                time.sleep(0.1)
 
-            raise OSError(f"Can't access OpenCVCamera({camera_idx}).")
+        logger.debug(f"Camera {self.index_or_path} connected and configured successfully.")
 
-        # Secondly, create the camera that will be used downstream.
-        # Note: For some unknown reason, calling `isOpened` blocks the camera which then
-        # needs to be re-created.
-        self.camera = cv2.VideoCapture(camera_idx, backend)
+    def _configure_capture_settings(self) -> None:
+        """
+        Applies the specified FPS, width, and height settings to the connected camera.
 
-        if self.fps is not None:
-            self.camera.set(cv2.CAP_PROP_FPS, self.fps)
-        if self.capture_width is not None:
-            self.camera.set(cv2.CAP_PROP_FRAME_WIDTH, self.capture_width)
-        if self.capture_height is not None:
-            self.camera.set(cv2.CAP_PROP_FRAME_HEIGHT, self.capture_height)
+        This method attempts to set the camera properties via OpenCV. It checks if
+        the camera successfully applied the settings and raises an error if not.
 
-        actual_fps = self.camera.get(cv2.CAP_PROP_FPS)
-        actual_width = self.camera.get(cv2.CAP_PROP_FRAME_WIDTH)
-        actual_height = self.camera.get(cv2.CAP_PROP_FRAME_HEIGHT)
+        Args:
+            fps: The desired frames per second. If None, the setting is skipped.
+            width: The desired capture width. If None, the setting is skipped.
+            height: The desired capture height. If None, the setting is skipped.
 
-        # Using `math.isclose` since actual fps can be a float (e.g. 29.9 instead of 30)
-        if self.fps is not None and not math.isclose(self.fps, actual_fps, rel_tol=1e-3):
-            # Using `OSError` since it's a broad that encompasses issues related to device communication
-            raise OSError(
-                f"Can't set {self.fps=} for OpenCVCamera({self.camera_index}). Actual value is {actual_fps}."
-            )
-        if self.capture_width is not None and not math.isclose(
-            self.capture_width, actual_width, rel_tol=1e-3
-        ):
-            raise OSError(
-                f"Can't set {self.capture_width=} for OpenCVCamera({self.camera_index}). Actual value is {actual_width}."
-            )
-        if self.capture_height is not None and not math.isclose(
-            self.capture_height, actual_height, rel_tol=1e-3
-        ):
-            raise OSError(
-                f"Can't set {self.capture_height=} for OpenCVCamera({self.camera_index}). Actual value is {actual_height}."
-            )
-
-        self.fps = round(actual_fps)
-        self.capture_width = round(actual_width)
-        self.capture_height = round(actual_height)
-        self.is_connected = True
-
-    def read(self, temporary_color_mode: str | None = None) -> np.ndarray:
-        """Read a frame from the camera returned in the format (height, width, channels)
-        (e.g. 480 x 640 x 3), contrarily to the pytorch format which is channel first.
-
-        Note: Reading a frame is done every `camera.fps` times per second, and it is blocking.
-        If you are reading data from other sensors, we advise to use `camera.async_read()` which is non blocking version of `camera.read()`.
+        Raises:
+            RuntimeError: If the camera fails to set any of the specified properties
+                          to the requested value.
+            DeviceNotConnectedError: If the camera is not connected when attempting
+                                     to configure settings.
         """
         if not self.is_connected:
-            raise DeviceNotConnectedError(
-                f"OpenCVCamera({self.camera_index}) is not connected. Try running `camera.connect()` first."
+            raise DeviceNotConnectedError(f"Cannot configure settings for {self} as it is not connected.")
+
+        if self.fps is None:
+            self.fps = self.videocapture.get(cv2.CAP_PROP_FPS)
+            logger.info(f"FPS set to camera default: {self.fps}.")
+        else:
+            self._validate_fps()
+
+        default_width = int(round(self.videocapture.get(cv2.CAP_PROP_FRAME_WIDTH)))
+        default_height = int(round(self.videocapture.get(cv2.CAP_PROP_FRAME_HEIGHT)))
+
+        if self.width is None or self.height is None:
+            if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+                self.width, self.height = default_height, default_width
+                self.capture_width, self.capture_height = default_width, default_height
+            else:
+                self.width, self.height = default_width, default_height
+                self.capture_width, self.capture_height = default_width, default_height
+            logger.info(f"Capture width set to camera default: {self.width}.")
+            logger.info(f"Capture height set to camera default: {self.height}.")
+        else:
+            self._validate_width_and_height()
+
+    def _validate_fps(self) -> None:
+        """Validates and sets the camera's frames per second (FPS)."""
+
+        success = self.videocapture.set(cv2.CAP_PROP_FPS, float(self.fps))
+        actual_fps = self.videocapture.get(cv2.CAP_PROP_FPS)
+        # Use math.isclose for robust float comparison
+        if not success or not math.isclose(self.fps, actual_fps, rel_tol=1e-3):
+            raise RuntimeError(
+                f"Failed to set requested FPS {self.fps} for {self}. Actual value reported: {actual_fps} set success: {success})."
             )
+        logger.debug(f"FPS set to {actual_fps} for {self}.")
+
+    def _validate_width_and_height(self) -> None:
+        """Validates and sets the camera's frame capture width and height."""
+
+        success = self.videocapture.set(cv2.CAP_PROP_FRAME_WIDTH, float(self.capture_width))
+        actual_width = int(round(self.videocapture.get(cv2.CAP_PROP_FRAME_WIDTH)))
+        if not success or self.capture_width != actual_width:
+            raise RuntimeError(
+                f"Failed to set requested capture width {self.capture_width} for {self}. Actual value: {actual_width} (set success: {success})."
+            )
+        logger.debug(f"Capture width set to {actual_width} for {self}.")
+
+        success = self.videocapture.set(cv2.CAP_PROP_FRAME_HEIGHT, float(self.capture_height))
+        actual_height = int(round(self.videocapture.get(cv2.CAP_PROP_FRAME_HEIGHT)))
+        if not success or self.capture_height != actual_height:
+            raise RuntimeError(
+                f"Failed to set requested capture height {self.capture_height} for {self}. Actual value: {actual_height} (set success: {success})."
+            )
+        logger.debug(f"Capture height set to {actual_height} for {self}.")
+
+    @staticmethod
+    def find_cameras() -> List[Dict[str, Any]]:
+        """
+        Detects available OpenCV cameras connected to the system.
+
+        On Linux, it scans '/dev/video*' paths. On other systems (like macOS, Windows),
+        it checks indices from 0 up to `MAX_OPENCV_INDEX`.
+
+        Returns:
+            List[Dict[str, Any]]: A list of dictionaries,
+            where each dictionary contains 'type', 'id' (port index or path),
+            and the default profile properties (width, height, fps, format).
+        """
+        found_cameras_info = []
+
+        if platform.system() == "Linux":
+            possible_paths = sorted(Path("/dev").glob("video*"), key=lambda p: p.name)
+            targets_to_scan = [str(p) for p in possible_paths]
+        else:
+            targets_to_scan = list(range(MAX_OPENCV_INDEX))
+
+        logger.debug(f"Found potential paths: {targets_to_scan}")
+        for target in targets_to_scan:
+            camera = cv2.VideoCapture(target)
+            if camera.isOpened():
+                default_width = int(camera.get(cv2.CAP_PROP_FRAME_WIDTH))
+                default_height = int(camera.get(cv2.CAP_PROP_FRAME_HEIGHT))
+                default_fps = camera.get(cv2.CAP_PROP_FPS)
+                default_format = camera.get(cv2.CAP_PROP_FORMAT)
+                camera_info = {
+                    "name": f"OpenCV Camera @ {target}",
+                    "type": "OpenCV",
+                    "id": target,
+                    "backend_api": camera.getBackendName(),
+                    "default_stream_profile": {
+                        "format": default_format,
+                        "width": default_width,
+                        "height": default_height,
+                        "fps": default_fps,
+                    },
+                }
+
+                found_cameras_info.append(camera_info)
+                camera.release()
+
+        if not found_cameras_info:
+            logger.warning("No OpenCV devices detected.")
+
+        logger.info(f"Detected OpenCV cameras: {[cam['id'] for cam in found_cameras_info]}")
+        return found_cameras_info
+
+    def read(self, color_mode: ColorMode | None = None) -> np.ndarray:
+        """
+        Reads a single frame synchronously from the camera.
+
+        This is a blocking call. It waits for the next available frame from the
+        camera hardware via OpenCV.
+
+        Args:
+            color_mode (Optional[ColorMode]): If specified, overrides the default
+                color mode (`self.color_mode`) for this read operation (e.g.,
+                request RGB even if default is BGR).
+
+        Returns:
+            np.ndarray: The captured frame as a NumPy array in the format
+                       (height, width, channels), using the specified or default
+                       color mode and applying any configured rotation.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is not connected.
+            RuntimeError: If reading the frame from the camera fails or if the
+                          received frame dimensions don't match expectations before rotation.
+            ValueError: If an invalid `color_mode` is requested.
+        """
+        if not self.is_connected:
+            raise DeviceNotConnectedError(f"{self} is not connected.")
 
         start_time = time.perf_counter()
 
-        ret, color_image = self.camera.read()
+        # NOTE(Steven): Are we okay with this blocking an undefined amount of time?
+        ret, frame = self.videocapture.read()
 
-        if not ret:
-            raise OSError(f"Can't capture color image from camera {self.camera_index}.")
+        if not ret or frame is None:
+            raise RuntimeError(
+                f"Failed to capture frame from {self}. '.read()' returned status={ret} and frame is None."
+            )
 
-        requested_color_mode = self.color_mode if temporary_color_mode is None else temporary_color_mode
+        # Post-process the frame (color conversion, dimension check, rotation)
+        processed_frame = self._postprocess_image(frame, color_mode)
 
-        if requested_color_mode not in ["rgb", "bgr"]:
+        read_duration_ms = (time.perf_counter() - start_time) * 1e3
+        logger.debug(f"{self} synchronous read took: {read_duration_ms:.1f}ms")
+
+        return processed_frame
+
+    def _postprocess_image(self, image: np.ndarray, color_mode: ColorMode | None = None) -> np.ndarray:
+        """
+        Applies color conversion, dimension validation, and rotation to a raw frame.
+
+        Args:
+            image (np.ndarray): The raw image frame (expected BGR format from OpenCV).
+            color_mode (Optional[ColorMode]): The target color mode (RGB or BGR). If None,
+                                             uses the instance's default `self.color_mode`.
+
+        Returns:
+            np.ndarray: The processed image frame.
+
+        Raises:
+            ValueError: If the requested `color_mode` is invalid.
+            RuntimeError: If the raw frame dimensions do not match the configured
+                          `width` and `height`.
+        """
+        requested_color_mode = self.color_mode if color_mode is None else color_mode
+
+        if requested_color_mode not in (ColorMode.RGB, ColorMode.BGR):
             raise ValueError(
-                f"Expected color values are 'rgb' or 'bgr', but {requested_color_mode} is provided."
+                f"Invalid requested color mode '{requested_color_mode}'. Expected {ColorMode.RGB} or {ColorMode.BGR}."
             )
 
-        # OpenCV uses BGR format as default (blue, green, red) for all operations, including displaying images.
-        # However, Deep Learning framework such as LeRobot uses RGB format as default to train neural networks,
-        # so we convert the image color from BGR to RGB.
-        if requested_color_mode == "rgb":
-            if self.mock:
-                import tests.cameras.mock_cv2 as cv2
-            else:
-                import cv2
+        h, w, c = image.shape
 
-            color_image = cv2.cvtColor(color_image, cv2.COLOR_BGR2RGB)
-
-        h, w, _ = color_image.shape
         if h != self.capture_height or w != self.capture_width:
-            raise OSError(
-                f"Can't capture color image with expected height and width ({self.height} x {self.width}). ({h} x {w}) returned instead."
+            raise RuntimeError(
+                f"Captured frame dimensions ({h}x{w}) do not match configured capture dimensions ({self.capture_height}x{self.capture_width}) for {self}."
             )
 
-        if self.rotation is not None:
-            color_image = cv2.rotate(color_image, self.rotation)
+        processed_image = image
+        if requested_color_mode == ColorMode.RGB:
+            processed_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+            logger.debug(f"Converted frame from BGR to RGB for {self}.")
 
-        # log the number of seconds it took to read the image
-        self.logs["delta_timestamp_s"] = time.perf_counter() - start_time
+        if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+            processed_image = cv2.rotate(processed_image, self.rotation)
+            logger.debug(f"Rotated frame by {self.config.rotation} degrees for {self}.")
 
-        # log the utc time at which the image was received
-        self.logs["timestamp_utc"] = capture_timestamp_utc()
+        return processed_image
 
-        self.color_image = color_image
+    def _read_loop(self):
+        """
+        Internal loop run by the background thread for asynchronous reading.
 
-        return color_image
-
-    def read_loop(self):
+        Continuously reads frames from the camera using the synchronous `read()`
+        method and places the latest frame into the `frame_queue`. It overwrites
+        any previous frame in the queue.
+        """
+        logger.debug(f"Starting read loop thread for {self}.")
         while not self.stop_event.is_set():
             try:
-                self.color_image = self.read()
+                color_image = self.read()
+
+                with contextlib.suppress(queue.Empty):
+                    _ = self.frame_queue.get_nowait()
+                self.frame_queue.put(color_image)
+                logger.debug(f"Frame placed in queue for {self}.")
+
+            except DeviceNotConnectedError:
+                logger.error(f"Read loop for {self} stopped: Camera disconnected.")
+                break
             except Exception as e:
-                print(f"Error reading in thread: {e}")
+                logger.warning(f"Error reading frame in background thread for {self}: {e}")
 
-    def async_read(self):
+        logger.debug(f"Stopping read loop thread for {self}.")
+
+    def _start_read_thread(self) -> None:
+        """Starts or restarts the background read thread if it's not running."""
+        if self.thread is not None and self.thread.is_alive():
+            self.thread.join(timeout=0.1)
+        if self.stop_event is not None:
+            self.stop_event.set()
+
+        self.stop_event = Event()
+        self.thread = Thread(
+            target=self._read_loop, args=(), name=f"OpenCVCameraReadLoop-{self}-{self.index_or_path}"
+        )
+        self.thread.daemon = True
+        self.thread.start()
+        logger.debug(f"Read thread started for {self}.")
+
+    def _stop_read_thread(self) -> None:
+        """Signals the background read thread to stop and waits for it to join."""
+        if self.stop_event is not None:
+            self.stop_event.set()
+
+        if self.thread is not None and self.thread.is_alive():
+            self.thread.join(timeout=2.0)
+            if self.thread.is_alive():
+                logger.warning(f"Read thread for {self} did not terminate gracefully after 2 seconds.")
+            else:
+                logger.debug(f"Read thread for {self} joined successfully.")
+
+        self.thread = None
+        self.stop_event = None
+        logger.debug(f"Read thread stopped for {self}.")
+
+    def async_read(self, timeout_ms: float = 2000) -> np.ndarray:
+        """
+        Reads the latest available frame asynchronously.
+
+        This method retrieves the most recent frame captured by the background
+        read thread. It does not block waiting for the camera hardware directly,
+        only waits for a frame to appear in the internal queue up to the specified
+        timeout.
+
+        Args:
+            timeout_ms (float): Maximum time in milliseconds to wait for a frame
+                to become available in the queue. Defaults to 2000ms (2 seconds).
+
+        Returns:
+            np.ndarray: The latest captured frame as a NumPy array in the format
+                       (height, width, channels), processed according to configuration.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is not connected.
+            TimeoutError: If no frame becomes available within the specified timeout.
+            RuntimeError: If an unexpected error occurs while retrieving from the queue.
+        """
         if not self.is_connected:
-            raise DeviceNotConnectedError(
-                f"OpenCVCamera({self.camera_index}) is not connected. Try running `camera.connect()` first."
-            )
+            raise DeviceNotConnectedError(f"{self} is not connected.")
 
-        if self.thread is None:
-            self.stop_event = threading.Event()
-            self.thread = Thread(target=self.read_loop, args=())
-            self.thread.daemon = True
-            self.thread.start()
+        if self.thread is None or not self.thread.is_alive():
+            self._start_read_thread()
 
-        num_tries = 0
-        while True:
-            if self.color_image is not None:
-                return self.color_image
-
-            time.sleep(1 / self.fps)
-            num_tries += 1
-            if num_tries > self.fps * 2:
-                raise TimeoutError("Timed out waiting for async_read() to start.")
+        try:
+            return self.frame_queue.get(timeout=timeout_ms / 1000.0)
+        except queue.Empty as e:
+            thread_alive = self.thread is not None and self.thread.is_alive()
+            raise TimeoutError(
+                f"Timed out waiting for frame from camera {self} after {timeout_ms} ms. "
+                f"Read thread alive: {thread_alive}."
+            ) from e
+        except Exception as e:
+            logger.exception(f"Unexpected error getting frame from queue for {self}: {e}")
+            raise RuntimeError(f"Error getting frame from queue for camera {self.index_or_path}: {e}") from e
 
     def disconnect(self):
-        if not self.is_connected:
-            raise DeviceNotConnectedError(
-                f"OpenCVCamera({self.camera_index}) is not connected. Try running `camera.connect()` first."
-            )
+        """
+        Disconnects from the camera and cleans up resources.
+
+        Stops the background read thread (if running) and releases the OpenCV
+        VideoCapture object.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is already disconnected.
+        """
+        if not self.is_connected and self.thread is None:
+            raise DeviceNotConnectedError(f"{self} not connected.")
 
         if self.thread is not None:
-            self.stop_event.set()
-            self.thread.join()  # wait for the thread to finish
-            self.thread = None
-            self.stop_event = None
+            self._stop_read_thread()
 
-        self.camera.release()
-        self.camera = None
-        self.is_connected = False
+        if self.videocapture is not None:
+            self.videocapture.release()
+            self.videocapture = None
 
-    def __del__(self):
-        if getattr(self, "is_connected", False):
-            self.disconnect()
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(
-        description="Save a few frames using `OpenCVCamera` for all cameras connected to the computer, or a selected subset."
-    )
-    parser.add_argument(
-        "--camera-ids",
-        type=int,
-        nargs="*",
-        default=None,
-        help="List of camera indices used to instantiate the `OpenCVCamera`. If not provided, find and use all available camera indices.",
-    )
-    parser.add_argument(
-        "--fps",
-        type=int,
-        default=None,
-        help="Set the number of frames recorded per seconds for all cameras. If not provided, use the default fps of each camera.",
-    )
-    parser.add_argument(
-        "--width",
-        type=str,
-        default=None,
-        help="Set the width for all cameras. If not provided, use the default width of each camera.",
-    )
-    parser.add_argument(
-        "--height",
-        type=str,
-        default=None,
-        help="Set the height for all cameras. If not provided, use the default height of each camera.",
-    )
-    parser.add_argument(
-        "--images-dir",
-        type=Path,
-        default="outputs/images_from_opencv_cameras",
-        help="Set directory to save a few frames for each camera.",
-    )
-    parser.add_argument(
-        "--record-time-s",
-        type=float,
-        default=4.0,
-        help="Set the number of seconds used to record the frames. By default, 2 seconds.",
-    )
-    args = parser.parse_args()
-    save_images_from_cameras(**vars(args))
+        logger.info(f"{self} disconnected.")

lerobot/common/cameras/opencv/configuration_opencv.py

@@ -1,38 +1,71 @@
-from dataclasses import dataclass
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
 
-from ..configs import CameraConfig
+from dataclasses import dataclass
+from pathlib import Path
+
+from ..configs import CameraConfig, ColorMode, Cv2Rotation
 
 
 @CameraConfig.register_subclass("opencv")
 @dataclass
 class OpenCVCameraConfig(CameraConfig):
-    """
-    Example of tested options for Intel Real Sense D405:
+    """Configuration class for OpenCV-based camera devices or video files.
 
+    This class provides configuration options for cameras accessed through OpenCV,
+    supporting both physical camera devices and video files. It includes settings
+    for resolution, frame rate, color mode, and image rotation.
+
+    Example configurations:
     ```python
-    OpenCVCameraConfig(0, 30, 640, 480)
-    OpenCVCameraConfig(0, 60, 640, 480)
-    OpenCVCameraConfig(0, 90, 640, 480)
-    OpenCVCameraConfig(0, 30, 1280, 720)
+    # Basic configurations
+    OpenCVCameraConfig(0, 30, 1280, 720)   # 1280x720 @ 30FPS
+    OpenCVCameraConfig(/dev/video4, 60, 640, 480)   # 640x480 @ 60FPS
+
+    # Advanced configurations
+    OpenCVCameraConfig(128422271347, 30, 640, 480, rotation=Cv2Rotation.ROTATE_90)     # With 90° rotation
     ```
+
+    Attributes:
+        index_or_path: Either an integer representing the camera device index,
+                      or a Path object pointing to a video file.
+        fps: Requested frames per second for the color stream.
+        width: Requested frame width in pixels for the color stream.
+        height: Requested frame height in pixels for the color stream.
+        color_mode: Color mode for image output (RGB or BGR). Defaults to RGB.
+        rotation: Image rotation setting (0°, 90°, 180°, or 270°). Defaults to no rotation.
+
+    Note:
+        - Only 3-channel color output (RGB/BGR) is currently supported.
     """
 
-    camera_index: int
-    fps: int | None = None
-    width: int | None = None
-    height: int | None = None
-    color_mode: str = "rgb"
-    channels: int | None = None
-    rotation: int | None = None
-    mock: bool = False
+    index_or_path: int | Path
+    color_mode: ColorMode = ColorMode.RGB
+    rotation: Cv2Rotation = Cv2Rotation.NO_ROTATION
 
     def __post_init__(self):
-        if self.color_mode not in ["rgb", "bgr"]:
+        if self.color_mode not in (ColorMode.RGB, ColorMode.BGR):
             raise ValueError(
-                f"`color_mode` is expected to be 'rgb' or 'bgr', but {self.color_mode} is provided."
+                f"`color_mode` is expected to be {ColorMode.RGB.value} or {ColorMode.BGR.value}, but {self.color_mode} is provided."
             )
 
-        self.channels = 3
-
-        if self.rotation not in [-90, None, 90, 180]:
-            raise ValueError(f"`rotation` must be in [-90, None, 90, 180] (got {self.rotation})")
+        if self.rotation not in (
+            Cv2Rotation.NO_ROTATION,
+            Cv2Rotation.ROTATE_90,
+            Cv2Rotation.ROTATE_180,
+            Cv2Rotation.ROTATE_270,
+        ):
+            raise ValueError(
+                f"`rotation` is expected to be in {(Cv2Rotation.NO_ROTATION, Cv2Rotation.ROTATE_90, Cv2Rotation.ROTATE_180, Cv2Rotation.ROTATE_270)}, but {self.rotation} is provided."
+            )

lerobot/common/cameras/realsense/__init__.py

@@ -0,0 +1,16 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .camera_realsense import RealSenseCamera
+from .configuration_realsense import RealSenseCameraConfig

lerobot/common/cameras/realsense/camera_realsense.py

@@ -0,0 +1,612 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Provides the RealSenseCamera class for capturing frames from Intel RealSense cameras.
+"""
+
+import contextlib
+import logging
+import queue
+import time
+from threading import Event, Thread
+from typing import Any, Dict, List
+
+import cv2
+import numpy as np
+
+try:
+    import pyrealsense2 as rs
+except Exception as e:
+    logging.info(f"Could not import realsense: {e}")
+
+from lerobot.common.errors import DeviceAlreadyConnectedError, DeviceNotConnectedError
+
+from ..camera import Camera
+from ..configs import ColorMode
+from ..utils import get_cv2_rotation
+from .configuration_realsense import RealSenseCameraConfig
+
+logger = logging.getLogger(__name__)
+
+
+class RealSenseCamera(Camera):
+    """
+    Manages interactions with Intel RealSense cameras for frame and depth recording.
+
+    This class provides an interface similar to `OpenCVCamera` but tailored for
+    RealSense devices, leveraging the `pyrealsense2` library. It uses the camera's
+    unique serial number for identification, offering more stability than device
+    indices, especially on Linux. It also supports capturing depth maps alongside
+    color frames.
+
+    Use the provided utility script to find available camera indices and default profiles:
+    ```bash
+    python -m lerobot.find_cameras realsense
+    ```
+
+    A `RealSenseCamera` instance requires a configuration object specifying the
+    camera's serial number or a unique device name. If using the name, ensure only
+    one camera with that name is connected.
+
+    The camera's default settings (FPS, resolution, color mode) from the stream
+    profile are used unless overridden in the configuration.
+
+    Example:
+        ```python
+        from lerobot.common.cameras.realsense import RealSenseCamera, RealSenseCameraConfig
+        from lerobot.common.cameras import ColorMode, Cv2Rotation
+
+        # Basic usage with serial number
+        config = RealSenseCameraConfig(serial_number_or_name="1234567890") # Replace with actual SN
+        camera = RealSenseCamera(config)
+        camera.connect()
+
+        # Read 1 frame synchronously
+        color_image = camera.read()
+        print(color_image.shape)
+
+        # Read 1 frame asynchronously
+        async_image = camera.async_read()
+
+        # When done, properly disconnect the camera using
+        camera.disconnect()
+
+        # Example with depth capture and custom settings
+        custom_config = RealSenseCameraConfig(
+            serial_number_or_name="1234567890", # Replace with actual SN
+            fps=30,
+            width=1280,
+            height=720,
+            color_mode=ColorMode.BGR, # Request BGR output
+            rotation=Cv2Rotation.NO_ROTATION,
+            use_depth=True
+        )
+        depth_camera = RealSenseCamera(custom_config)
+        try:
+            depth_camera.connect()
+            depth_map = depth_camera.read_depth()
+            print(f"Depth shape: {depth_map.shape}")
+        finally:
+            depth_camera.disconnect()
+
+        # Example using a unique camera name
+        name_config = RealSenseCameraConfig(serial_number_or_name="Intel RealSense D435") # If unique
+        name_camera = RealSenseCamera(name_config)
+        # ... connect, read, disconnect ...
+        ```
+    """
+
+    def __init__(self, config: RealSenseCameraConfig):
+        """
+        Initializes the RealSenseCamera instance.
+
+        Args:
+            config: The configuration settings for the camera.
+        """
+
+        super().__init__(config)
+
+        self.config = config
+
+        if isinstance(config.serial_number_or_name, int):
+            self.serial_number = str(config.serial_number_or_name)
+        else:
+            self.serial_number = self._find_serial_number_from_name(config.serial_number_or_name)
+
+        self.fps = config.fps
+        self.color_mode = config.color_mode
+        self.use_depth = config.use_depth
+        self.warmup_time = config.warmup_time
+
+        self.rs_pipeline: rs.pipeline | None = None
+        self.rs_profile: rs.pipeline_profile | None = None
+
+        self.thread: Thread | None = None
+        self.stop_event: Event | None = None
+        self.frame_queue: queue.Queue = queue.Queue(maxsize=1)
+
+        self.rotation: int | None = get_cv2_rotation(config.rotation)
+
+        if self.height and self.width:
+            if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+                self.capture_width, self.capture_height = self.height, self.width
+            else:
+                self.capture_width, self.capture_height = self.width, self.height
+
+    def __str__(self) -> str:
+        return f"{self.__class__.__name__}({self.serial_number})"
+
+    @property
+    def is_connected(self) -> bool:
+        """Checks if the camera pipeline is started and streams are active."""
+        return self.rs_pipeline is not None and self.rs_profile is not None
+
+    def connect(self, warmup: bool = True):
+        """
+        Connects to the RealSense camera specified in the configuration.
+
+        Initializes the RealSense pipeline, configures the required streams (color
+        and optionally depth), starts the pipeline, and validates the actual stream settings.
+
+        Raises:
+            DeviceAlreadyConnectedError: If the camera is already connected.
+            ValueError: If the configuration is invalid (e.g., missing serial/name, name not unique).
+            ConnectionError: If the camera is found but fails to start the pipeline or no RealSense devices are detected at all.
+            RuntimeError: If the pipeline starts but fails to apply requested settings.
+        """
+        if self.is_connected:
+            raise DeviceAlreadyConnectedError(f"{self} is already connected.")
+
+        self.rs_pipeline = rs.pipeline()
+        rs_config = self._make_rs_pipeline_config()
+
+        try:
+            self.rs_profile = self.rs_pipeline.start(rs_config)
+            logger.debug(f"Successfully started pipeline for camera {self.serial_number}.")
+        except RuntimeError as e:
+            self.rs_profile = None
+            self.rs_pipeline = None
+            raise ConnectionError(
+                f"Failed to open {self} camera. Run 'python -m lerobot.find_cameras realsense' for details about the available cameras in your system."
+            ) from e
+
+        logger.debug(f"Validating stream configuration for {self}...")
+        self._validate_capture_settings()
+
+        if warmup:
+            if self.warmup_time is None:
+                raise ValueError(
+                    f"Warmup time is not set for {self}. Please set a warmup time in the configuration."
+                )
+            logger.debug(f"Reading a warm-up frames for {self} for {self.warmup_time} seconds...")
+            start_time = time.time()
+            while time.time() - start_time < self.warmup_time:
+                self.read()
+                time.sleep(0.1)
+
+        logger.info(f"{self} connected.")
+
+    @staticmethod
+    def find_cameras() -> List[Dict[str, Any]]:
+        """
+        Detects available Intel RealSense cameras connected to the system.
+
+        Returns:
+            List[Dict[str, Any]]: A list of dictionaries,
+            where each dictionary contains 'type', 'id' (serial number), 'name',
+            firmware version, USB type, and other available specs, and the default profile properties (width, height, fps, format).
+
+        Raises:
+            OSError: If pyrealsense2 is not installed.
+            ImportError: If pyrealsense2 is not installed.
+        """
+        found_cameras_info = []
+        context = rs.context()
+        devices = context.query_devices()
+
+        for device in devices:
+            camera_info = {
+                "name": device.get_info(rs.camera_info.name),
+                "type": "RealSense",
+                "id": device.get_info(rs.camera_info.serial_number),
+                "firmware_version": device.get_info(rs.camera_info.firmware_version),
+                "usb_type_descriptor": device.get_info(rs.camera_info.usb_type_descriptor),
+                "physical_port": device.get_info(rs.camera_info.physical_port),
+                "product_id": device.get_info(rs.camera_info.product_id),
+                "product_line": device.get_info(rs.camera_info.product_line),
+            }
+
+            # Get stream profiles for each sensor
+            sensors = device.query_sensors()
+            for sensor in sensors:
+                profiles = sensor.get_stream_profiles()
+
+                for profile in profiles:
+                    if profile.is_video_stream_profile() and profile.is_default():
+                        vprofile = profile.as_video_stream_profile()
+                        stream_info = {
+                            "stream_type": vprofile.stream_name(),
+                            "format": vprofile.format().name,
+                            "width": vprofile.width(),
+                            "height": vprofile.height(),
+                            "fps": vprofile.fps(),
+                        }
+                        camera_info["default_stream_profile"] = stream_info
+
+            found_cameras_info.append(camera_info)
+
+        logger.info(f"Detected RealSense cameras: {[cam['id'] for cam in found_cameras_info]}")
+        return found_cameras_info
+
+    def _find_serial_number_from_name(self, name: str) -> str:
+        """Finds the serial number for a given unique camera name."""
+        camera_infos = self.find_cameras()
+        found_devices = [cam for cam in camera_infos if str(cam["name"]) == name]
+
+        if not found_devices:
+            available_names = [cam["name"] for cam in camera_infos]
+            raise ValueError(
+                f"No RealSense camera found with name '{name}'. Available camera names: {available_names}"
+            )
+
+        if len(found_devices) > 1:
+            serial_numbers = [dev["serial_number"] for dev in found_devices]
+            raise ValueError(
+                f"Multiple RealSense cameras found with name '{name}'. "
+                f"Please use a unique serial number instead. Found SNs: {serial_numbers}"
+            )
+
+        serial_number = str(found_devices[0]["serial_number"])
+        logger.info(f"Found serial number '{serial_number}' for camera name '{name}'.")
+        return serial_number
+
+    def _make_rs_pipeline_config(self) -> rs.config:
+        """Creates and configures the RealSense pipeline configuration object."""
+        rs_config = rs.config()
+        rs.config.enable_device(rs_config, self.serial_number)
+
+        if self.width and self.height and self.fps:
+            logger.debug(
+                f"Requesting Color Stream: {self.capture_width}x{self.capture_height} @ {self.fps} FPS, Format: {rs.format.rgb8}"
+            )
+            rs_config.enable_stream(
+                rs.stream.color, self.capture_width, self.capture_height, rs.format.rgb8, self.fps
+            )
+            if self.use_depth:
+                logger.debug(
+                    f"Requesting Depth Stream: {self.capture_width}x{self.capture_height} @ {self.fps} FPS, Format: {rs.format.z16}"
+                )
+                rs_config.enable_stream(
+                    rs.stream.depth, self.capture_width, self.capture_height, rs.format.z16, self.fps
+                )
+        else:
+            logger.debug(f"Requesting Color Stream: Default settings, Format: {rs.stream.color}")
+            rs_config.enable_stream(rs.stream.color)
+            if self.use_depth:
+                logger.debug(f"Requesting Depth Stream: Default settings, Format: {rs.stream.depth}")
+                rs_config.enable_stream(rs.stream.depth)
+
+        return rs_config
+
+    def _validate_capture_settings(self) -> None:
+        """
+        Validates if the actual stream settings match the requested configuration.
+
+        This method compares the requested FPS, width, and height against the
+        actual settings obtained from the active RealSense profile after the
+        pipeline has started.
+
+        Raises:
+            RuntimeError: If the actual camera settings significantly deviate
+                          from the requested ones.
+            DeviceNotConnectedError: If the camera is not connected when attempting
+                                     to validate settings.
+        """
+        if not self.is_connected:
+            raise DeviceNotConnectedError(f"Cannot validate settings for {self} as it is not connected.")
+
+        stream = self.rs_profile.get_stream(rs.stream.color).as_video_stream_profile()
+
+        if self.fps is None:
+            self.fps = stream.fps()
+
+        if self.width is None or self.height is None:
+            actual_width = int(round(stream.width()))
+            actual_height = int(round(stream.height()))
+            if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+                self.width, self.height = actual_height, actual_width
+                self.capture_width, self.capture_height = actual_width, actual_height
+            else:
+                self.width, self.height = actual_width, actual_height
+                self.capture_width, self.capture_height = actual_width, actual_height
+            logger.info(f"Capture width set to camera default: {self.width}.")
+            logger.info(f"Capture height set to camera default: {self.height}.")
+        else:
+            self._validate_width_and_height(stream)
+
+        if self.use_depth:
+            stream = self.rs_profile.get_stream(rs.stream.depth).as_video_stream_profile()
+            self._validate_width_and_height(stream)
+
+    def _validate_width_and_height(self, stream) -> None:
+        """Validates and sets the internal capture width and height based on actual stream width."""
+        actual_width = int(round(stream.width()))
+        actual_height = int(round(stream.height()))
+
+        if self.capture_width != actual_width:
+            raise RuntimeError(
+                f"Failed to set requested capture width {self.capture_width} for {self}. Actual value: {actual_width}."
+            )
+        logger.debug(f"Capture width set to {actual_width} for {self}.")
+
+        if self.capture_height != actual_height:
+            raise RuntimeError(
+                f"Failed to set requested capture height {self.capture_height} for {self}. Actual value: {actual_height}."
+            )
+        logger.debug(f"Capture height set to {actual_height} for {self}.")
+
+    def read_depth(self, timeout_ms: int = 100) -> np.ndarray:
+        """
+        Reads a single frame (depth) synchronously from the camera.
+
+        This is a blocking call. It waits for a coherent set of frames (depth)
+        from the camera hardware via the RealSense pipeline.
+
+        Args:
+            timeout_ms (int): Maximum time in milliseconds to wait for a frame. Defaults to 100ms.
+
+        Returns:
+            np.ndarray: The depth map as a NumPy array (height, width)
+                  of type `np.uint16` (raw depth values in millimeters) and rotation.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is not connected.
+            RuntimeError: If reading frames from the pipeline fails or frames are invalid.
+        """
+
+        if not self.is_connected:
+            raise DeviceNotConnectedError(f"{self} is not connected.")
+        if not self.use_depth:
+            raise RuntimeError(
+                f"Failed to capture depth frame '.read_depth()'. Depth stream is not enabled for {self}."
+            )
+
+        start_time = time.perf_counter()
+
+        ret, frame = self.rs_pipeline.try_wait_for_frames(timeout_ms=timeout_ms)
+
+        if not ret or frame is None:
+            raise RuntimeError(f"{self} failed to capture frame. Returned status='{ret}'.")
+
+        depth_frame = frame.get_depth_frame()
+        depth_map = np.asanyarray(depth_frame.get_data())
+
+        depth_map_processed = self._postprocess_image(depth_map, depth_frame=True)
+
+        read_duration_ms = (time.perf_counter() - start_time) * 1e3
+        logger.debug(f"{self} synchronous read took: {read_duration_ms:.1f}ms")
+
+        return depth_map_processed
+
+    def read(self, color_mode: ColorMode | None = None, timeout_ms: int = 100) -> np.ndarray:
+        """
+        Reads a single frame (color) synchronously from the camera.
+
+        This is a blocking call. It waits for a coherent set of frames (color)
+        from the camera hardware via the RealSense pipeline.
+
+        Args:
+            timeout_ms (int): Maximum time in milliseconds to wait for a frame. Defaults to 100ms.
+
+        Returns:
+            np.ndarray: The captured color frame as a NumPy array
+              (height, width, channels), processed according to `color_mode` and rotation.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is not connected.
+            RuntimeError: If reading frames from the pipeline fails or frames are invalid.
+            ValueError: If an invalid `color_mode` is requested.
+        """
+
+        if not self.is_connected:
+            raise DeviceNotConnectedError(f"{self} is not connected.")
+
+        start_time = time.perf_counter()
+
+        ret, frame = self.rs_pipeline.try_wait_for_frames(
+            timeout_ms=timeout_ms
+        )  # NOTE(Steven): This read has a timeout while opencv doesn't
+
+        if not ret or frame is None:
+            raise RuntimeError(
+                f"Failed to capture frame from {self}. '.read()' returned status={ret} and frame is None."
+            )
+
+        color_frame = frame.get_color_frame()
+        color_image_raw = np.asanyarray(color_frame.get_data())
+
+        color_image_processed = self._postprocess_image(color_image_raw, color_mode)
+
+        read_duration_ms = (time.perf_counter() - start_time) * 1e3
+        logger.debug(f"{self} synchronous read took: {read_duration_ms:.1f}ms")
+
+        return color_image_processed
+
+    def _postprocess_image(
+        self, image: np.ndarray, color_mode: ColorMode | None = None, depth_frame: bool = False
+    ) -> np.ndarray:
+        """
+        Applies color conversion, dimension validation, and rotation to a raw color frame.
+
+        Args:
+            image (np.ndarray): The raw image frame (expected RGB format from RealSense).
+            color_mode (Optional[ColorMode]): The target color mode (RGB or BGR). If None,
+                                             uses the instance's default `self.color_mode`.
+
+        Returns:
+            np.ndarray: The processed image frame according to `self.color_mode` and `self.rotation`.
+
+        Raises:
+            ValueError: If the requested `color_mode` is invalid.
+            RuntimeError: If the raw frame dimensions do not match the configured
+                          `width` and `height`.
+        """
+
+        if color_mode and color_mode not in (ColorMode.RGB, ColorMode.BGR):
+            raise ValueError(
+                f"Invalid requested color mode '{color_mode}'. Expected {ColorMode.RGB} or {ColorMode.BGR}."
+            )
+
+        if depth_frame:
+            h, w = image.shape
+        else:
+            h, w, _c = image.shape
+
+        if h != self.capture_height or w != self.capture_width:
+            raise RuntimeError(
+                f"Captured frame dimensions ({h}x{w}) do not match configured capture dimensions ({self.capture_height}x{self.capture_width}) for {self}."
+            )
+
+        processed_image = image
+        if self.color_mode == ColorMode.BGR:
+            processed_image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
+            logger.debug(f"Converted frame from RGB to BGR for {self}.")
+
+        if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+            processed_image = cv2.rotate(processed_image, self.rotation)
+            logger.debug(f"Rotated frame by {self.config.rotation} degrees for {self}.")
+
+        return processed_image
+
+    def _read_loop(self):
+        """
+        Internal loop run by the background thread for asynchronous reading.
+
+        Continuously reads frames (color and optional depth) using `read()`
+        and places the latest result (single image or tuple) into the `frame_queue`.
+        It overwrites any previous frame in the queue.
+        """
+        logger.debug(f"Starting read loop thread for {self}.")
+        while not self.stop_event.is_set():
+            try:
+                frame_data = self.read(timeout_ms=500)
+
+                with contextlib.suppress(queue.Empty):
+                    _ = self.frame_queue.get_nowait()
+                self.frame_queue.put(frame_data)
+                logger.debug(f"Frame data placed in queue for {self}.")
+
+            except DeviceNotConnectedError:
+                logger.error(f"Read loop for {self} stopped: Camera disconnected.")
+                break
+            except Exception as e:
+                logger.warning(f"Error reading frame in background thread for {self}: {e}")
+
+        logger.debug(f"Stopping read loop thread for {self}.")
+
+    def _start_read_thread(self) -> None:
+        """Starts or restarts the background read thread if it's not running."""
+        if self.thread is not None and self.thread.is_alive():
+            self.thread.join(timeout=0.1)
+        if self.stop_event is not None:
+            self.stop_event.set()
+
+        self.stop_event = Event()
+        self.thread = Thread(target=self._read_loop, args=(), name=f"{self}_read_loop")
+        self.thread.daemon = True
+        self.thread.start()
+        logger.debug(f"Read thread started for {self}.")
+
+    def _stop_read_thread(self):
+        """Signals the background read thread to stop and waits for it to join."""
+        if self.stop_event is not None:
+            self.stop_event.set()
+
+        if self.thread is not None and self.thread.is_alive():
+            self.thread.join(timeout=2.0)
+            if self.thread.is_alive():
+                logger.warning(f"Read thread for {self} did not terminate gracefully after 2 seconds.")
+            else:
+                logger.debug(f"Read thread for {self} joined successfully.")
+
+        self.thread = None
+        self.stop_event = None
+        logger.debug(f"Read thread stopped for {self}.")
+
+    # NOTE(Steven): Missing implementation for depth for now
+    def async_read(self, timeout_ms: float = 100) -> np.ndarray:
+        """
+        Reads the latest available frame data (color or color+depth) asynchronously.
+
+        This method retrieves the most recent frame captured by the background
+        read thread. It does not block waiting for the camera hardware directly,
+        only waits for a frame to appear in the internal queue up to the specified
+        timeout.
+
+        Args:
+            timeout_ms (float): Maximum time in milliseconds to wait for a frame
+                to become available in the queue. Defaults to 100ms (0.1 seconds).
+
+        Returns:
+            np.ndarray:
+            The latest captured frame data (color image), processed according to configuration.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is not connected.
+            TimeoutError: If no frame data becomes available within the specified timeout.
+            RuntimeError: If the background thread died unexpectedly or another queue error occurs.
+        """
+        if not self.is_connected:
+            raise DeviceNotConnectedError(f"{self} is not connected.")
+
+        if self.thread is None or not self.thread.is_alive():
+            self._start_read_thread()
+
+        try:
+            return self.frame_queue.get(timeout=timeout_ms / 1000.0)
+        except queue.Empty as e:
+            thread_alive = self.thread is not None and self.thread.is_alive()
+            raise TimeoutError(
+                f"Timed out waiting for frame from camera {self} after {timeout_ms} ms. "
+                f"Read thread alive: {thread_alive}."
+            ) from e
+        except Exception as e:
+            raise RuntimeError(f"Error getting frame data from queue for camera {self}: {e}") from e
+
+    def disconnect(self):
+        """
+        Disconnects from the camera, stops the pipeline, and cleans up resources.
+
+        Stops the background read thread (if running) and stops the RealSense pipeline.
+
+        Raises:
+            DeviceNotConnectedError: If the camera is already disconnected (pipeline not running).
+        """
+
+        if not self.is_connected and self.thread is None:
+            raise DeviceNotConnectedError(
+                f"Attempted to disconnect {self}, but it appears already disconnected."
+            )
+
+        if self.thread is not None:
+            self._stop_read_thread()
+
+        if self.rs_pipeline is not None:
+            logger.debug(f"Stopping RealSense pipeline object for {self}.")
+            self.rs_pipeline.stop()
+            self.rs_pipeline = None
+            self.rs_profile = None
+
+        logger.info(f"{self} disconnected.")

lerobot/common/cameras/realsense/configuration_realsense.py

@@ -0,0 +1,80 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from dataclasses import dataclass
+
+from ..configs import CameraConfig, ColorMode, Cv2Rotation
+
+
+@CameraConfig.register_subclass("intelrealsense")
+@dataclass
+class RealSenseCameraConfig(CameraConfig):
+    """Configuration class for Intel RealSense cameras.
+
+    This class provides specialized configuration options for Intel RealSense cameras,
+    including support for depth sensing and device identification via serial number or name.
+
+    Example configurations for Intel RealSense D405:
+    ```python
+    # Basic configurations
+    RealSenseCameraConfig(128422271347, 30, 1280, 720)   # 1280x720 @ 30FPS
+    RealSenseCameraConfig(128422271347, 60, 640, 480)   # 640x480 @ 60FPS
+
+    # Advanced configurations
+    RealSenseCameraConfig(128422271347, 30, 640, 480, use_depth=True)  # With depth sensing
+    RealSenseCameraConfig(128422271347, 30, 640, 480, rotation=Cv2Rotation.ROTATE_90)     # With 90° rotation
+    ```
+
+    Attributes:
+        fps: Requested frames per second for the color stream.
+        width: Requested frame width in pixels for the color stream.
+        height: Requested frame height in pixels for the color stream.
+        serial_number_or_name: Unique serial number or human-readable name to identify the camera.
+        color_mode: Color mode for image output (RGB or BGR). Defaults to RGB.
+        use_depth: Whether to enable depth stream. Defaults to False.
+        rotation: Image rotation setting (0°, 90°, 180°, or 270°). Defaults to no rotation.
+
+    Note:
+        - Either name or serial_number must be specified.
+        - Depth stream configuration (if enabled) will use the same FPS as the color stream.
+        - The actual resolution and FPS may be adjusted by the camera to the nearest supported mode.
+        - For `fps`, `width` and `height`, either all of them need to be set, or none of them.
+    """
+
+    serial_number_or_name: int | str
+    color_mode: ColorMode = ColorMode.RGB
+    use_depth: bool = False
+    rotation: Cv2Rotation = Cv2Rotation.NO_ROTATION  # NOTE(Steven): Check if draccus can parse to an enum
+
+    def __post_init__(self):
+        if self.color_mode not in (ColorMode.RGB, ColorMode.BGR):
+            raise ValueError(
+                f"`color_mode` is expected to be {ColorMode.RGB.value} or {ColorMode.BGR.value}, but {self.color_mode} is provided."
+            )
+
+        if self.rotation not in (
+            Cv2Rotation.NO_ROTATION,
+            Cv2Rotation.ROTATE_90,
+            Cv2Rotation.ROTATE_180,
+            Cv2Rotation.ROTATE_270,
+        ):
+            raise ValueError(
+                f"`rotation` is expected to be in {(Cv2Rotation.NO_ROTATION, Cv2Rotation.ROTATE_90, Cv2Rotation.ROTATE_180, Cv2Rotation.ROTATE_270)}, but {self.rotation} is provided."
+            )
+
+        values = (self.fps, self.width, self.height)
+        if any(v is not None for v in values) and any(v is None for v in values):
+            raise ValueError(
+                "For `fps`, `width` and `height`, either all of them need to be set, or none of them."
+            )

lerobot/common/cameras/utils.py

@@ -1,5 +1,30 @@
+#!/usr/bin/env python
+
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import platform
+from pathlib import Path
+from typing import TypeAlias
+
+import numpy as np
+from PIL import Image
+
 from .camera import Camera
-from .configs import CameraConfig
+from .configs import CameraConfig, Cv2Rotation
+
+IndexOrPath: TypeAlias = int | Path
 
 
 def make_cameras_from_configs(camera_configs: dict[str, CameraConfig]) -> dict[str, Camera]:
@@ -12,10 +37,39 @@ def make_cameras_from_configs(camera_configs: dict[str, CameraConfig]) -> dict[s
             cameras[key] = OpenCVCamera(cfg)
 
         elif cfg.type == "intelrealsense":
-            from .intel.camera_realsense import RealSenseCamera
+            from .realsense.camera_realsense import RealSenseCamera
 
             cameras[key] = RealSenseCamera(cfg)
         else:
             raise ValueError(f"The motor type '{cfg.type}' is not valid.")
 
     return cameras
+
+
+def get_cv2_rotation(rotation: Cv2Rotation) -> int | None:
+    import cv2
+
+    if rotation == Cv2Rotation.ROTATE_90:
+        return cv2.ROTATE_90_CLOCKWISE
+    elif rotation == Cv2Rotation.ROTATE_180:
+        return cv2.ROTATE_180
+    elif rotation == Cv2Rotation.ROTATE_270:
+        return cv2.ROTATE_90_COUNTERCLOCKWISE
+    else:
+        return None
+
+
+def get_cv2_backend() -> int:
+    import cv2
+
+    if platform.system() == "Windows":
+        return cv2.CAP_AVFOUNDATION
+    else:
+        return cv2.CAP_ANY
+
+
+def save_image(img_array: np.ndarray, camera_index: int, frame_index: int, images_dir: Path):
+    img = Image.fromarray(img_array)
+    path = images_dir / f"camera_{camera_index:02d}_frame_{frame_index:06d}.png"
+    path.parent.mkdir(parents=True, exist_ok=True)
+    img.save(str(path), quality=100)

lerobot/common/constants.py

@@ -48,5 +48,5 @@ default_cache_path = Path(HF_HOME) / "lerobot"
 HF_LEROBOT_HOME = Path(os.getenv("HF_LEROBOT_HOME", default_cache_path)).expanduser()
 
 # calibration dir
-default_calibration_path = HF_LEROBOT_HOME / ".calibration"
+default_calibration_path = HF_LEROBOT_HOME / "calibration"
 HF_LEROBOT_CALIBRATION = Path(os.getenv("HF_LEROBOT_CALIBRATION", default_calibration_path)).expanduser()

lerobot/common/datasets/factory.py

@@ -49,7 +49,7 @@ def resolve_delta_timestamps(
                 "observation.state": [-0.04, -0.02, 0]
                 "observation.action": [-0.02, 0, 0.02]
             }
-            returns `None` if the the resulting dict is empty.
+            returns `None` if the resulting dict is empty.
     """
     delta_timestamps = {}
     for key in ds_meta.features:

lerobot/common/datasets/lerobot_dataset.py

@@ -48,7 +48,6 @@ from lerobot.common.datasets.utils import (
     embed_images,
     get_delta_indices,
     get_episode_data_index,
-    get_features_from_robot,
     get_hf_features_from_features,
     get_safe_version,
     hf_transform_to_torch,
@@ -67,11 +66,11 @@ from lerobot.common.datasets.utils import (
 )
 from lerobot.common.datasets.video_utils import (
     VideoFrame,
-    decode_video_frames_torchvision,
+    decode_video_frames,
     encode_video_frames,
+    get_safe_default_codec,
     get_video_info,
 )
-from lerobot.common.robots.utils import Robot
 
 CODEBASE_VERSION = "v2.1"
 
@@ -303,10 +302,9 @@ class LeRobotDatasetMetadata:
         cls,
         repo_id: str,
         fps: int,
-        root: str | Path | None = None,
-        robot: Robot | None = None,
+        features: dict,
         robot_type: str | None = None,
-        features: dict | None = None,
+        root: str | Path | None = None,
         use_videos: bool = True,
     ) -> "LeRobotDatasetMetadata":
         """Creates metadata for a LeRobotDataset."""
@@ -316,33 +314,27 @@ class LeRobotDatasetMetadata:
 
         obj.root.mkdir(parents=True, exist_ok=False)
 
-        if robot is not None:
-            features = get_features_from_robot(robot, use_videos)
-            robot_type = robot.robot_type
-            if not all(cam.fps == fps for cam in robot.cameras.values()):
-                logging.warning(
-                    f"Some cameras in your {robot.robot_type} robot don't have an fps matching the fps of your dataset."
-                    "In this case, frames from lower fps cameras will be repeated to fill in the blanks."
-                )
-        elif features is None:
-            raise ValueError(
-                "Dataset features must either come from a Robot or explicitly passed upon creation."
-            )
-        else:
-            # TODO(aliberts, rcadene): implement sanity check for features
-            features = {**features, **DEFAULT_FEATURES}
+        # if robot is not None:
+        #     features = get_features_from_robot(robot, use_videos)
+        #     robot_type = robot.robot_type
+        #     if not all(cam.fps == fps for cam in robot.cameras.values()):
+        #         logging.warning(
+        #             f"Some cameras in your {robot.robot_type} robot don't have an fps matching the fps of your dataset."
+        #             "In this case, frames from lower fps cameras will be repeated to fill in the blanks."
+        #         )
 
-            # check if none of the features contains a "/" in their names,
-            # as this would break the dict flattening in the stats computation, which uses '/' as separator
-            for key in features:
-                if "/" in key:
-                    raise ValueError(f"Feature names should not contain '/'. Found '/' in feature '{key}'.")
+        # TODO(aliberts, rcadene): implement sanity check for features
+        features = {**features, **DEFAULT_FEATURES}
 
-            features = {**features, **DEFAULT_FEATURES}
+        # check if none of the features contains a "/" in their names,
+        # as this would break the dict flattening in the stats computation, which uses '/' as separator
+        for key in features:
+            if "/" in key:
+                raise ValueError(f"Feature names should not contain '/'. Found '/' in feature '{key}'.")
 
         obj.tasks, obj.task_to_task_index = {}, {}
         obj.episodes_stats, obj.stats, obj.episodes = {}, {}, {}
-        obj.info = create_empty_dataset_info(CODEBASE_VERSION, fps, robot_type, features, use_videos)
+        obj.info = create_empty_dataset_info(CODEBASE_VERSION, fps, features, use_videos, robot_type)
         if len(obj.video_keys) > 0 and not use_videos:
             raise ValueError()
         write_json(obj.info, obj.root / INFO_PATH)
@@ -462,8 +454,8 @@ class LeRobotDataset(torch.utils.data.Dataset):
             download_videos (bool, optional): Flag to download the videos. Note that when set to True but the
                 video files are already present on local disk, they won't be downloaded again. Defaults to
                 True.
-            video_backend (str | None, optional): Video backend to use for decoding videos. There is currently
-                a single option which is the pyav decoder used by Torchvision. Defaults to pyav.
+            video_backend (str | None, optional): Video backend to use for decoding videos. Defaults to torchcodec when available int the platform; otherwise, defaults to 'pyav'.
+                You can also use the 'pyav' decoder used by Torchvision, which used to be the default option, or 'video_reader' which is another decoder of Torchvision.
         """
         super().__init__()
         self.repo_id = repo_id
@@ -473,7 +465,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         self.episodes = episodes
         self.tolerance_s = tolerance_s
         self.revision = revision if revision else CODEBASE_VERSION
-        self.video_backend = video_backend if video_backend else "pyav"
+        self.video_backend = video_backend if video_backend else get_safe_default_codec()
         self.delta_indices = None
 
         # Unused attributes
@@ -707,9 +699,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         item = {}
         for vid_key, query_ts in query_timestamps.items():
             video_path = self.root / self.meta.get_video_file_path(ep_idx, vid_key)
-            frames = decode_video_frames_torchvision(
-                video_path, query_ts, self.tolerance_s, self.video_backend
-            )
+            frames = decode_video_frames(video_path, query_ts, self.tolerance_s, self.video_backend)
             item[vid_key] = frames.squeeze(0)
 
         return item
@@ -786,7 +776,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         else:
             self.image_writer.save_image(image=image, fpath=fpath)
 
-    def add_frame(self, frame: dict) -> None:
+    def add_frame(self, frame: dict, task: str, timestamp: float | None = None) -> None:
         """
         This function only adds the frame to the episode_buffer. Apart from images — which are written in a
         temporary directory — nothing is written to disk. To save those frames, the 'save_episode()' method
@@ -804,17 +794,14 @@ class LeRobotDataset(torch.utils.data.Dataset):
 
         # Automatically add frame_index and timestamp to episode buffer
         frame_index = self.episode_buffer["size"]
-        timestamp = frame.pop("timestamp") if "timestamp" in frame else frame_index / self.fps
+        if timestamp is None:
+            timestamp = frame_index / self.fps
         self.episode_buffer["frame_index"].append(frame_index)
         self.episode_buffer["timestamp"].append(timestamp)
+        self.episode_buffer["task"].append(task)
 
         # Add frame features to episode_buffer
         for key in frame:
-            if key == "task":
-                # Note: we associate the task in natural language to its task index during `save_episode`
-                self.episode_buffer["task"].append(frame["task"])
-                continue
-
             if key not in self.features:
                 raise ValueError(
                     f"An element of the frame is not in the features. '{key}' not in '{self.features.keys()}'."
@@ -990,10 +977,9 @@ class LeRobotDataset(torch.utils.data.Dataset):
         cls,
         repo_id: str,
         fps: int,
+        features: dict,
         root: str | Path | None = None,
-        robot: Robot | None = None,
         robot_type: str | None = None,
-        features: dict | None = None,
         use_videos: bool = True,
         tolerance_s: float = 1e-4,
         image_writer_processes: int = 0,
@@ -1005,10 +991,9 @@ class LeRobotDataset(torch.utils.data.Dataset):
         obj.meta = LeRobotDatasetMetadata.create(
             repo_id=repo_id,
             fps=fps,
-            root=root,
-            robot=robot,
             robot_type=robot_type,
             features=features,
+            root=root,
             use_videos=use_videos,
         )
         obj.repo_id = obj.meta.repo_id
@@ -1029,7 +1014,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         obj.delta_timestamps = None
         obj.delta_indices = None
         obj.episode_data_index = None
-        obj.video_backend = video_backend if video_backend is not None else "pyav"
+        obj.video_backend = video_backend if video_backend is not None else get_safe_default_codec()
         return obj
 
 
@@ -1054,7 +1039,7 @@ class MultiLeRobotDataset(torch.utils.data.Dataset):
         super().__init__()
         self.repo_ids = repo_ids
         self.root = Path(root) if root else HF_LEROBOT_HOME
-        self.tolerances_s = tolerances_s if tolerances_s else {repo_id: 1e-4 for repo_id in repo_ids}
+        self.tolerances_s = tolerances_s if tolerances_s else dict.fromkeys(repo_ids, 0.0001)
         # Construct the underlying datasets passing everything but `transform` and `delta_timestamps` which
         # are handled by this class.
         self._datasets = [

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/mobile_cabinet.txt

@@ -1,85 +0,0 @@
-https://drive.google.com/file/d/1_SOJkgfP5yZyVjMhTt3nwhvyUjcnlI51/view?usp=drive_link
-https://drive.google.com/file/d/1rmgN8UUzph1qwJnzG1d-uOafodn-gLvb/view?usp=drive_link
-https://drive.google.com/file/d/1NYQ-XxsBVinB6dUoZmVWweT83367P3i2/view?usp=drive_link
-https://drive.google.com/file/d/1oAv_j74zxxCJieMG7r5Vl2BeHK1__3s3/view?usp=drive_link
-https://drive.google.com/file/d/1wFUJQROsrTJt64YRuIeExhFjr2wnK5uu/view?usp=drive_link
-https://drive.google.com/file/d/1KzL3Tt0Le7jVl58XVRUcmigmXjyiuhbK/view?usp=drive_link
-https://drive.google.com/file/d/1qy_YBladeHtianSSGtgAPSHtMin7msvf/view?usp=drive_link
-https://drive.google.com/file/d/1rA_F0V_qL_nyuC_0aBKCisF4-0TIkF2Y/view?usp=drive_link
-https://drive.google.com/file/d/1hw-8qMpz9VgSt62XoASqNRuPECpCwJQP/view?usp=drive_link
-https://drive.google.com/file/d/1BpHOl9rKMzdvNGka6js7C0s40hH6vnDA/view?usp=drive_link
-https://drive.google.com/file/d/1PazhkhiDnJ-OUMyDVDFxEZNKQQqHiNWS/view?usp=drive_link
-https://drive.google.com/file/d/1lZ665R6ATl57dypxH4dGJ2NSt6XYnbuz/view?usp=drive_link
-https://drive.google.com/file/d/1V9HzLaf-tlG15wUzT7KrTDCS_z1vi5NV/view?usp=drive_link
-https://drive.google.com/file/d/1aKauWiXoKqbNwn_2xs4MrmLlaNYlVNmO/view?usp=drive_link
-https://drive.google.com/file/d/1WVD5DFhriO1YmmOgiVHhacR6HWoTPxav/view?usp=drive_link
-https://drive.google.com/file/d/1_X43WgeBAsfkhH9EmpyPki8U9joMeAGC/view?usp=drive_link
-https://drive.google.com/file/d/1t8x0GqWoNKWtnBsB7_D40Z34nL9ak4kf/view?usp=drive_link
-https://drive.google.com/file/d/15V_f26WaKOXjKnq2T3HRWAmtQUi4lbu2/view?usp=drive_link
-https://drive.google.com/file/d/11VFIAsiSDsMOBANgrOcZBpKB9AFWnLy7/view?usp=drive_link
-https://drive.google.com/file/d/1M0NS7vVaxJv3FHnuRYtdwTFYF7We4LxP/view?usp=drive_link
-https://drive.google.com/file/d/1mR0OItTNqFnVLoczcyKYlm6drAy778lO/view?usp=drive_link
-https://drive.google.com/file/d/1NbVFWDQAh-z4JJ4D-Zw6Lps9kdvpqh2j/view?usp=drive_link
-https://drive.google.com/file/d/1JQoZGBzl4W3QG26-n39tefcGN0fDRMbB/view?usp=drive_link
-https://drive.google.com/file/d/1VBjHl-TvZpncopvasIP5G9gecbB2a5f6/view?usp=drive_link
-https://drive.google.com/file/d/1VzSf6zaB21nahm7MsPwroXbJ84NIwq0b/view?usp=drive_link
-https://drive.google.com/file/d/1OtNnfMEydNtZOcivs4k6E_uJSpf8PkGy/view?usp=drive_link
-https://drive.google.com/file/d/14nVvpvsrFr_03Pa_N7MKzwnRwibOUYM6/view?usp=drive_link
-https://drive.google.com/file/d/1M8li6duiO2r3lv_9HhF_XJn0oZUIEK5F/view?usp=drive_link
-https://drive.google.com/file/d/1Cpzea6fO14lxAaNfSBifqoa4ekhCiLD1/view?usp=drive_link
-https://drive.google.com/file/d/1mbxRTm5vlbsY9UJ0jfjM6j9D7kPJjBpG/view?usp=drive_link
-https://drive.google.com/file/d/1RXD1i6IfWsHRlCxVmG04h2h5Ycm_WwZN/view?usp=drive_link
-https://drive.google.com/file/d/1QFqFSwDGOk1BkgGmqgCcc2BRWnJ6R3MA/view?usp=drive_link
-https://drive.google.com/file/d/1bFqWR8DQM0ZUxxtS2bl-RANQvukeFLzp/view?usp=drive_link
-https://drive.google.com/file/d/1pR-rH3yNGoyPdD4hJ6-3lXQ-PstBx9du/view?usp=drive_link
-https://drive.google.com/file/d/107OAwLY-hva9HeQLIK7VCh-ytdDabVjr/view?usp=drive_link
-https://drive.google.com/file/d/1Tpl08QOaSZ37GTO4awFWSdD8wBR9xdlT/view?usp=drive_link
-https://drive.google.com/file/d/1MR164AOM-0S1T6RX8xKTV2IHyaCvpqAW/view?usp=drive_link
-https://drive.google.com/file/d/1_wknJfVnStIhJ82lU_QtcrwahsqYIsr8/view?usp=drive_link
-https://drive.google.com/file/d/1ZuEktWrbYkTx0l5pj3WiZ2CJrfbDOHNo/view?usp=drive_link
-https://drive.google.com/file/d/15G_10hkkkq6yxvyI5NGZirlF-RzduR2F/view?usp=drive_link
-https://drive.google.com/file/d/1DBKxg3ONqh7dhLuX6oh1Yyo2x383V1Hp/view?usp=drive_link
-https://drive.google.com/file/d/1B5iDBkTUr5vopDddV_fHud18SqAHhauS/view?usp=drive_link
-https://drive.google.com/file/d/1acwFV0eenRkki1QcjSKH5xqOtys-P3Pr/view?usp=drive_link
-https://drive.google.com/file/d/1S47BI83xyrh-FKXsvAQqer98Biu_p8XK/view?usp=drive_link
-https://drive.google.com/file/d/1JL6DmBZl3uyq9dyLfgSqtGF06e7E9JwM/view?usp=drive_link
-https://drive.google.com/file/d/16WvRS4Kjog8Pxgr0E3sGGnI01YwL9Uql/view?usp=drive_link
-https://drive.google.com/file/d/12ttGqL33IPWg0-s1SD44rr22M6LiSQBr/view?usp=drive_link
-https://drive.google.com/file/d/1OyZqqnldTU_DliRbr6x0C4a_iWPwIN7j/view?usp=drive_link
-https://drive.google.com/file/d/1oYk00IpLnR9fesLfD15Ebe7nVBffEbcS/view?usp=drive_link
-https://drive.google.com/file/d/1eyE2-MQduCEqCd-5_kl5zsoOEERAzpZD/view?usp=drive_link
-https://drive.google.com/file/d/1ir1Ya-vO0d97pfvbePlUeuKTTRc0qIMU/view?usp=drive_link
-https://drive.google.com/file/d/1hOi-JnqlMt47gVnLZHMTqeojyYVErohl/view?usp=drive_link
-https://drive.google.com/file/d/1NFFw5_PqigQ7xGqsL-MNq2B1r5yAscCf/view?usp=drive_link
-https://drive.google.com/file/d/1uftq1-Zlh8d2sNLWrlVcKYQUwZTD7o24/view?usp=drive_link
-https://drive.google.com/file/d/1-ax19dSLPacVgk000T-m3l4flPcg07pM/view?usp=drive_link
-https://drive.google.com/file/d/126y-lgn86-ZmCz8hooF1THKJGGObw3OB/view?usp=drive_link
-https://drive.google.com/file/d/1JiDniK0VmDIkk92AbBILb8J2Ba59PWML/view?usp=drive_link
-https://drive.google.com/file/d/1kr8nPIRljiU0R4J9SMgj80o1FPQxzu9z/view?usp=drive_link
-https://drive.google.com/file/d/1bbThWRij1pKBh_kFgV8FwK0sXtTHBoLX/view?usp=drive_link
-https://drive.google.com/file/d/1WenzDW6lxk1xkOFm-OiGFfc0ROskAuKU/view?usp=drive_link
-https://drive.google.com/file/d/1MiKRzuzUn1yN-k_6kPJJzIGy7dT-nnsD/view?usp=drive_link
-https://drive.google.com/file/d/17rRg2tcmB-gNhQ0KoZJQmNfyFeoij1jH/view?usp=drive_link
-https://drive.google.com/file/d/11mokBpvrY3ld6sY5WztREtJ1jgqfQV70/view?usp=drive_link
-https://drive.google.com/file/d/1Il_6IOx9NDp1bX_KHizJfBwzTufTmn86/view?usp=drive_link
-https://drive.google.com/file/d/1KswtJGsxJ7eeBDAmNA_aeLjOxcH6MIxa/view?usp=drive_link
-https://drive.google.com/file/d/1gzMhi5uWu4C3Y6WbQ3L-08V96GxTZrRR/view?usp=drive_link
-https://drive.google.com/file/d/1nRQFtaBxfUCYc2W90Qibh0kHCt6YQCfc/view?usp=drive_link
-https://drive.google.com/file/d/1vs-gyW-KheqHbUATwAhA2mmR9GOGw7f_/view?usp=drive_link
-https://drive.google.com/file/d/1MuxzGOA2fgLaHryq82KkQumtuRJGcUOC/view?usp=drive_link
-https://drive.google.com/file/d/1IIwxZnGlqrXLUXqG6yMO0r7uhCvhpk9e/view?usp=drive_link
-https://drive.google.com/file/d/1vE7XPyaFcXP4DtTY5Y9WKIt7zWgmX-Cr/view?usp=drive_link
-https://drive.google.com/file/d/1j-bIV09gr21RC3-x1N_pK4RPLV3fmWKz/view?usp=drive_link
-https://drive.google.com/file/d/1t3nW1rD3S-EL0Oymb5U7ZAj5UMkydkln/view?usp=drive_link
-https://drive.google.com/file/d/14hbfHCdMKtJZ41F9CQReMec2jeRFTOqR/view?usp=drive_link
-https://drive.google.com/file/d/1x-hUyOSne5BW0AzQ3W6_Pf4g5yXQWi9M/view?usp=drive_link
-https://drive.google.com/file/d/1sw9JqRg6E-3P84I3ZhzTrJMu0vuiaMmP/view?usp=drive_link
-https://drive.google.com/file/d/1LuqhQlL4MGZhB_6THmkovRxrlP26BbdC/view?usp=drive_link
-https://drive.google.com/file/d/15C5K6v_lkjnMSmUvVyqHQKwh2N166e7K/view?usp=drive_link
-https://drive.google.com/file/d/1ns_9eSsQeeoZ10nlbkLy8tu0GmJFSnkt/view?usp=drive_link
-https://drive.google.com/file/d/1NpzWJeK6CqjxzjIMYe6aYdX8xGsQwD4o/view?usp=drive_link
-https://drive.google.com/file/d/1NMLezwufKJ9_8xTc9KQThSzVVD71B9Ui/view?usp=drive_link
-https://drive.google.com/file/d/1aa71DCUqs6oXlIxX35jgsmsgm-NlDxPV/view?usp=drive_link
-https://drive.google.com/file/d/1UJzkIZzAL0j-D5YQBnoq7mHvttASy12O/view?usp=drive_link
-https://drive.google.com/file/d/1nPgx36HIJFb7oI94VbRzWjpPP2GANxzG/view?usp=drive_link
-https://drive.google.com/file/d/1NovAP-KVJjqcuvWy3d6G4ptGGAIDqcCx/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/mobile_chair.txt

@@ -1,55 +0,0 @@
-https://drive.google.com/file/d/11M3Ye0r5agMaaicPbVGD0q2Hb3rGklbb/view?usp=drive_link
-https://drive.google.com/file/d/1-tx7SvYYgSvXCvnf_EI2OVdwK-CkFY6S/view?usp=drive_link
-https://drive.google.com/file/d/1EWJunmOpMHaU1hE106wwpbkGYcjQXYAF/view?usp=drive_link
-https://drive.google.com/file/d/1IDn95Z7FSiCckrSENtGV4u3RyFHNQSDY/view?usp=drive_link
-https://drive.google.com/file/d/1CwzvWj1i7QOtqrZvsCZ6BdZaKNDfpN32/view?usp=drive_link
-https://drive.google.com/file/d/1HvAvlhm77nAD3Td24QPSeq8lw-Rl_aOh/view?usp=drive_link
-https://drive.google.com/file/d/1t-suKYOPhXH666RpAYNRp2QU_DOy3AeM/view?usp=drive_link
-https://drive.google.com/file/d/18xpKgWh7RWyjMN5PkLTOo-AxsAadAuRw/view?usp=drive_link
-https://drive.google.com/file/d/1oci5Eto-ztv-AQNz8EnwZveBIhxvk-xJ/view?usp=drive_link
-https://drive.google.com/file/d/1Y-t_4vxdE6NpHO0DLJR8f3mD0Q-Wj5-c/view?usp=drive_link
-https://drive.google.com/file/d/1lylRqbbbB8bgtpsBWMPACmHJreuKmllv/view?usp=drive_link
-https://drive.google.com/file/d/1yliSyMig_NXShWfQx6qyW7Ijf2Y5lFK6/view?usp=drive_link
-https://drive.google.com/file/d/1XXhwJsJbeb7KXAooGvJapnm9bjnGUmxS/view?usp=drive_link
-https://drive.google.com/file/d/1_xs1f3hW2JArKyvfF7UWubWjyROGTLs6/view?usp=drive_link
-https://drive.google.com/file/d/1WVEHpr6EqKCZbkHapQSTXJq4xE4SWFT-/view?usp=drive_link
-https://drive.google.com/file/d/1RqOHv9pEQGvW8NUA7ynffFmG999TL_Az/view?usp=drive_link
-https://drive.google.com/file/d/1cu5AgD2gh-uA3PFJmzxxzNaF3qOSlYY1/view?usp=drive_link
-https://drive.google.com/file/d/1SsrXqiPclNrnYToPZ9Uq-k3y0C4qdHT1/view?usp=drive_link
-https://drive.google.com/file/d/1-J7EXf0vjkLIfSqT8ICEsP6CTjzSLBop/view?usp=drive_link
-https://drive.google.com/file/d/11O7ewUmoZXfyyKjy_6B5RW4DpjICxqBT/view?usp=drive_link
-https://drive.google.com/file/d/1iic44kZoCsjNsfAz2cMstZ9-WQvAhblF/view?usp=drive_link
-https://drive.google.com/file/d/1yLV1lVX-2WnWQldGlnQZ0x7QBuDiVkL3/view?usp=drive_link
-https://drive.google.com/file/d/1Tybp9ru98TTbGn4eyROpUQwDFuALWXmk/view?usp=drive_link
-https://drive.google.com/file/d/13E9OTMiipVJByDs5-J19oWwAz7l94LTN/view?usp=drive_link
-https://drive.google.com/file/d/1EeTpJQdMSliw4JzSMtJ6CyTvVdexjM4M/view?usp=drive_link
-https://drive.google.com/file/d/1NHyNwoFqzeAu-1_PSpq5JfxaiD_xbpn9/view?usp=drive_link
-https://drive.google.com/file/d/1fJcS0phDp4xm_FyGaJ5wr9Pe4KqtHaxD/view?usp=drive_link
-https://drive.google.com/file/d/12AqrLUaewDPEcFRqPZeZFb_TQ0Lfi3At/view?usp=drive_link
-https://drive.google.com/file/d/1x_hd4Qsq1oJS-aj2t3qM7WbbV7KZj05b/view?usp=drive_link
-https://drive.google.com/file/d/14OUSUArmsB068hs6BuEIXQhI1Cyz8Sf0/view?usp=drive_link
-https://drive.google.com/file/d/16zlzh1T5zeUJQnFf382NXkFEKEnDub4O/view?usp=drive_link
-https://drive.google.com/file/d/1IbDltmN-NEFCNtr1TO4ILxEgQ94rtjWv/view?usp=drive_link
-https://drive.google.com/file/d/15gmlf8Gx9455pZ1AlqcCSwh3nDPxMzSr/view?usp=drive_link
-https://drive.google.com/file/d/1qHpRL1oZfIMo_vxnm8qfwQ-7l0BZIVva/view?usp=drive_link
-https://drive.google.com/file/d/1H1xskIgiFZivkYn23rMzH3xePGOh3VTC/view?usp=drive_link
-https://drive.google.com/file/d/1avls6Pv0kYiCMNVknbc1zQsgy64MUDMM/view?usp=drive_link
-https://drive.google.com/file/d/1MmWVgCj5khc8KMIifmt3EzF1o-CtPyyn/view?usp=drive_link
-https://drive.google.com/file/d/1U0kCc_xqW0WNppf4sbnK14euWKdPZtzB/view?usp=drive_link
-https://drive.google.com/file/d/16CaEyQscOuhLj23PEGDTL9DeyNkohkMn/view?usp=drive_link
-https://drive.google.com/file/d/1Iu8uM6UUJ0zW8tvN-9UiOe_4oSNzEutg/view?usp=drive_link
-https://drive.google.com/file/d/1UImqiBaIxCR-1DNJaZhHqeHhaySOtVIr/view?usp=drive_link
-https://drive.google.com/file/d/1VpU2V_leIoRIyv_lAvE7eLHBG8DxCTnp/view?usp=drive_link
-https://drive.google.com/file/d/1_Q8J27OT3Xby7QY6yHvIJauFRWEMxkRm/view?usp=drive_link
-https://drive.google.com/file/d/1bantmVo1L9Xz4tbiNw_a1UC2Z_HPO1wT/view?usp=drive_link
-https://drive.google.com/file/d/1IRIXMJMCBDkBjbaHvAlEiBogSvZ1jK_3/view?usp=drive_link
-https://drive.google.com/file/d/1mAHXKjiFbjwydypW2t5Lv8_H5x6nHegl/view?usp=drive_link
-https://drive.google.com/file/d/1SfyY796fLrBCMY39OcyuxZafqSCRZPZk/view?usp=drive_link
-https://drive.google.com/file/d/1X-44sZ8CcfzIskc0dvSx882o1yFhHaZB/view?usp=drive_link
-https://drive.google.com/file/d/1BOIWCCCk6DLD4Bmvc75ZbbLi9AQm-1ao/view?usp=drive_link
-https://drive.google.com/file/d/1RuyDtRE1kk76sw-wP8vx5SgLoPF3PA_H/view?usp=drive_link
-https://drive.google.com/file/d/1c4eoQiBbGuy3CTAQDUSkd84Ponh1roAQ/view?usp=drive_link
-https://drive.google.com/file/d/19PXB9z4Ljq6dsbf9TqcOrrP5SRbw2Tc_/view?usp=drive_link
-https://drive.google.com/file/d/1nn1VVZVoIXWdYDozR7XHXE4mPLQG80PQ/view?usp=drive_link
-https://drive.google.com/file/d/1MBdFGOKPV8GUhwoSsJ_Ky3qAMLM2Bv3K/view?usp=drive_link
-https://drive.google.com/file/d/1of3k_M-7Nh3I1TndcWedxK4ca9dn8Sc5/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/mobile_elevator.txt

@@ -1,20 +0,0 @@
-https://drive.google.com/file/d/12ctkOAdkCNGN1JLbZb5ww3XTBn2LFpGI/view?usp=drive_link
-https://drive.google.com/file/d/1G_Vd46_4fq6O64gHHjUbJX5Ld44ZZx0y/view?usp=drive_link
-https://drive.google.com/file/d/1uKgUy73B3xBogQAOUhfZjO0X5qZGsi2c/view?usp=drive_link
-https://drive.google.com/file/d/1fu9cIrfI-fE2LhdGUxbx7-8Ci_PF8Ypm/view?usp=drive_link
-https://drive.google.com/file/d/1Ygk9ZPJzx8xw2A9JF3NHbJ44TqnvSTQR/view?usp=drive_link
-https://drive.google.com/file/d/18m5xPuccNsEB20WPshm3zhxmXc6k63ED/view?usp=drive_link
-https://drive.google.com/file/d/1DiqqxC44rriviRQpqogcv0-EB-Y6nr9g/view?usp=drive_link
-https://drive.google.com/file/d/1qPdaoTVDizJXkfXLioWU7iJ8hqCXSyOQ/view?usp=drive_link
-https://drive.google.com/file/d/1Fj9kIA_mG7f67WFfACJEaZ7izcHG7vUm/view?usp=drive_link
-https://drive.google.com/file/d/1WpYehZnI2P7dUdJPfkE-ij1rqCnjZEbB/view?usp=drive_link
-https://drive.google.com/file/d/1_zwWkT4jPyzB38STWb6whlzsPzXmfA9r/view?usp=drive_link
-https://drive.google.com/file/d/1U6-J4I_fPlSFFGfhZPxS5_YzKXwXIZYp/view?usp=drive_link
-https://drive.google.com/file/d/1pRhxxcTfZp5tQo_EScvJUwfc3amiS6Vk/view?usp=drive_link
-https://drive.google.com/file/d/1lWLntqra83RlYU_gN7Vostnfydf6gutd/view?usp=drive_link
-https://drive.google.com/file/d/1vIBKo0x-NYEHV1FvRpco1lQMpRdAWAIL/view?usp=drive_link
-https://drive.google.com/file/d/1pdrLV3JTQou_XH0Aap61Ssf60iVKm1jJ/view?usp=drive_link
-https://drive.google.com/file/d/1QTsLoQ7SwmKdQHjBGVDaR2uTwfFwtrOf/view?usp=drive_link
-https://drive.google.com/file/d/1Gytai8M_12J36GY6L_TulEcOC-035jwS/view?usp=drive_link
-https://drive.google.com/file/d/14LJudNc629NT-i8xreXtzl27ce_DxOFJ/view?usp=drive_link
-https://drive.google.com/file/d/1sBvPCODbzxGAI0S3lgN5cSG9Go3lRi00/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/mobile_shrimp.txt

@@ -1,18 +0,0 @@
-https://drive.google.com/file/d/1MJn9GbC8p9lN4gC9KDMLEkTkP_gGpXj0/view?usp=drive_link
-https://drive.google.com/file/d/1-4LXgjl7ZCOgp-8GCJmFRD8OeqN5Jf7-/view?usp=drive_link
-https://drive.google.com/file/d/1Ho06Ce0SPbqU3juaMxNUwAt3zCRLGC8W/view?usp=drive_link
-https://drive.google.com/file/d/1ivHoj7_7olBSxH-Y8kqXEW7ttITK-45j/view?usp=drive_link
-https://drive.google.com/file/d/1qjY4hM_IvZ8cq2II_n9MeJbvyeuN4oBP/view?usp=drive_link
-https://drive.google.com/file/d/1rKVhO_f92-7sw13T8hTVrza3B9oAVgoy/view?usp=drive_link
-https://drive.google.com/file/d/1pcLPHO8fBkc1-CRa88tyQtEueE4xiXNi/view?usp=drive_link
-https://drive.google.com/file/d/1Vev_chCsIeEdvQ8poEYNsOJFGy_QU8kZ/view?usp=drive_link
-https://drive.google.com/file/d/1l5G4zpRkxSLCQjvGPYSN4zfCvVRQuzMz/view?usp=drive_link
-https://drive.google.com/file/d/14vgthE1eoakXkr2-DRw50E6lAqYOiUuE/view?usp=drive_link
-https://drive.google.com/file/d/17nPSmKKmgQ2B7zkzWrZYiLM3RBuFod82/view?usp=drive_link
-https://drive.google.com/file/d/1QcDsxplVvb_ID9BVrihl5FvlC-j7waXi/view?usp=drive_link
-https://drive.google.com/file/d/18pEejBpI-eEVaWAAjBCyC0vgbX3T1Esj/view?usp=drive_link
-https://drive.google.com/file/d/1H8eH6_IRODtEFT6WoM77ltR5OoOrqXmI/view?usp=drive_link
-https://drive.google.com/file/d/1IWlpFRZhoxyG4nS13CWK4leZVk5wbNx4/view?usp=drive_link
-https://drive.google.com/file/d/1PbZA8_OCGmMLxNP9xbkLRSChniL4uGxl/view?usp=drive_link
-https://drive.google.com/file/d/1p9XAdmG2f_WeflNO4DIJ_tr1rK6M9B4B/view?usp=drive_link
-https://drive.google.com/file/d/1nS59Et1cNAvKo3Y4SeSGRuZD5TvBbCF3/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/mobile_wash_pan.txt

@@ -1 +0,0 @@
-https://drive.google.com/drive/folders/1S8eFg98IaGAIKVZ8QFWG1bx4mHa-O204

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/mobile_wipe_wine.txt

@@ -1,4 +0,0 @@
-https://drive.google.com/drive/folders/1tC_g1AJ8lglBLY-fjsQrG6DMBa3Ucp-0
-https://drive.google.com/file/d/1fG_Yi2MJrFjiUVN3XoiWXLtTxHlwwaDv/view?usp=drive_link
-https://drive.google.com/file/d/1WX32VWfzzX3Blmd06DRxLwFbMJfVe7P4/view?usp=drive_link
-https://drive.google.com/file/d/18onsX3vXg3xkFwP5bVUCjdV4n9TRn0C9/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/sim_insertion_human.txt

@@ -1,3 +0,0 @@
-https://drive.google.com/drive/folders/1RgyD0JgTX30H4IM5XZn8I3zSV_mr8pyF
-https://drive.google.com/file/d/18Cudl6nikDtgRolea7je8iF_gGKzynOP/view?usp=drive_link
-https://drive.google.com/file/d/1C1kZYyROzs-PrLc0SkDgUgMi4-L3lauE/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/sim_insertion_scripted.txt

@@ -1,3 +0,0 @@
-https://drive.google.com/drive/folders/1TsojQQSXtHEoGnqgJ3gmpPQR2DPLtS2N
-https://drive.google.com/file/d/1wfMSZ24oOh5KR_0aaP3Cnu_c4ZCveduB/view?usp=drive_link
-https://drive.google.com/file/d/17EuCUWS6uCCr6yyNzpXdcdE-_TTNCKtf/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/sim_transfer_cube_human.txt

@@ -1,3 +0,0 @@
-https://drive.google.com/drive/folders/1sc-E4QYW7A0o23m1u2VWNGVq5smAsfCo
-https://drive.google.com/file/d/18smMymtr8tIxaNUQ61gW6dG50pt3MvGq/view?usp=drive_link
-https://drive.google.com/file/d/1Nk7l53d9sJoGDBKAOnNrExX5nLacATc6/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/sim_transfer_cube_scripted.txt

@@ -1,3 +0,0 @@
-https://drive.google.com/drive/folders/1aRyoOhQwxhyt1J8XgEig4s6kzaw__LXj
-https://drive.google.com/file/d/1pnGIOd-E4-rhz2P3VxpknMKRZCoKt6eI/view?usp=drive_link
-https://drive.google.com/file/d/1GKReZHrXU73NMiC5zKCq_UtqPVtYq8eo/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_battery.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/19qS_n7vKgDcPeTMnvDHQ5-n73xEbJz5D
-https://drive.google.com/file/d/1oC31By0A2bsBeHyUwBdQw1z4ng6yi9Za/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_candy.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1m5rQ6UVH8Q9RQp_6c0CxkQ88-L-ScO7q
-https://drive.google.com/file/d/1wHz2qcmwcVG0C0CZ9MjQDQcmj4OY9_a3/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_coffee.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1seQGay470nGQ-knBI5TjsTr8iL9Qws5q
-https://drive.google.com/file/d/1T89hSX5U99wLGvGTE7yUBaQPOpyj6Sai/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_coffee_new.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1t3eDc5Rg0DveyRe8oTm6Dia_FYU5mXyf
-https://drive.google.com/file/d/1TXFaduTakvS0ZWJqKCX-HIvYglum_5CY/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_cups_open.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1Z9X3DNzd6LS0FFjQemNUMoMA5yk5VQOh
-https://drive.google.com/file/d/1Wlyc0vTkjXuWB6zbaVOWhEfD7BmPgUV_/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_fork_pick_up.txt

@@ -1,53 +0,0 @@
-https://drive.google.com/drive/folders/1DYgB4ifX4uIid9m9jnC0Zdz8Nf7ZC0fc
-https://drive.google.com/file/d/1Eb-NRNk_FmVleCbU_Ng5Y4dfcjTKN7Rv/view?usp=drive_link
-https://drive.google.com/file/d/1dkhjEADakT-44l9jf-nK4x89kr4yG_qb/view?usp=drive_link
-https://drive.google.com/file/d/14hDhgcZkVqNExGb4tIXpSjMshhqZETch/view?usp=drive_link
-https://drive.google.com/file/d/1zVMEHpHbuNyP5A_lYU7RPSLB-4V0yfZw/view?usp=drive_link
-https://drive.google.com/file/d/1JtgDjBvy7FnRpFzrx_foC3quorYQFAR-/view?usp=drive_link
-https://drive.google.com/file/d/1EHdneB6F-PP0dQlX8qPaXbxmKoBy_YwO/view?usp=drive_link
-https://drive.google.com/file/d/17Z0jjVBy1OPKREPu77_n_rQzorDiapji/view?usp=drive_link
-https://drive.google.com/file/d/1F4i23qPJ_qTf5jWjfLo4ARGJChznYWt3/view?usp=drive_link
-https://drive.google.com/file/d/1kZtXWM3uS0-rLblydBfJ0mMcVnMMXw9w/view?usp=drive_link
-https://drive.google.com/file/d/1mNODox87xFfY5Z_o5mcLsr8SHb39jDik/view?usp=drive_link
-https://drive.google.com/file/d/1Ob44VdmEUA93FKDECiRb5Ogz2xQg5IWp/view?usp=drive_link
-https://drive.google.com/file/d/1fdQLdjj3Cwv33R1wZhfrLz9Del8mqgHb/view?usp=drive_link
-https://drive.google.com/file/d/1Yu3L3ft21zP__XL8pCfhb788ZleuW1n5/view?usp=drive_link
-https://drive.google.com/file/d/1ozBBWXVZ9hXDh9ooHUNroHdYm8UDqnhJ/view?usp=drive_link
-https://drive.google.com/file/d/1o0TGqvfWw_Lunxb5ubKDS21Lr_WC0h75/view?usp=drive_link
-https://drive.google.com/file/d/1jZnd5eP5L6BH5l98BPN6OnoQx3fu8e9n/view?usp=drive_link
-https://drive.google.com/file/d/1S5sYbz8wcLYp0V67v13i4PRcBxodn4Hg/view?usp=drive_link
-https://drive.google.com/file/d/1rFeg_x6ftJYwPtBv34D3h2L2cpDLeR4G/view?usp=drive_link
-https://drive.google.com/file/d/1GvS3lcm4o6nm_scUk0XxKeVFNmzjucDZ/view?usp=drive_link
-https://drive.google.com/file/d/1-9i0riphC7NhhDahcQfD1QoBXP5gF90A/view?usp=drive_link
-https://drive.google.com/file/d/15p_IqGsMbKuvzMS872THAZr-3SBtb1Fr/view?usp=drive_link
-https://drive.google.com/file/d/1ToyYcBfJL8gbQn0q_59zPLsFmm7dmMJo/view?usp=drive_link
-https://drive.google.com/file/d/1e_7PNH7CYafE4pAebP7ZdI7XFbmEcy_i/view?usp=drive_link
-https://drive.google.com/file/d/1JoabvGVsIQdug2xOhUIhetEIyDM91y_Y/view?usp=drive_link
-https://drive.google.com/file/d/1kOMw1y0lmnVaCjwZICfzCsx6e0Z8MNGR/view?usp=drive_link
-https://drive.google.com/file/d/16it_wd1JOevUQTK2_CvF_pBACTgpIPgM/view?usp=drive_link
-https://drive.google.com/file/d/1IRcCj9HnJSfbyMgr5XEERGlEnWeZQwOc/view?usp=drive_link
-https://drive.google.com/file/d/1Z2dIJfq_S3liGmPN9Rphvkmucnmw7tlb/view?usp=drive_link
-https://drive.google.com/file/d/1J3NoAjzndGx9yNyaBOJHdNny1epzUoBt/view?usp=drive_link
-https://drive.google.com/file/d/18nOvxV1k8FSmBrhT4TPo2sKKSZXougyx/view?usp=drive_link
-https://drive.google.com/file/d/1CT8FxclafFMjSd7gCWVw3VSeryeiF04i/view?usp=drive_link
-https://drive.google.com/file/d/16M9KVqQMFfSsXfypK0bocFft8Nz3j2Rt/view?usp=drive_link
-https://drive.google.com/file/d/18QPVkw6bj6HW8LTPrQLWrrUX4R6RcF42/view?usp=drive_link
-https://drive.google.com/file/d/1hQTVtA5hBTE_StXpJafTZJ3tgt2VQQ_t/view?usp=drive_link
-https://drive.google.com/file/d/1Dn-d5g69H6EgAWgsFdrcbJKtz7ySsCQ8/view?usp=drive_link
-https://drive.google.com/file/d/13hMr16483P7ALYv73yMRUN37fJdVQM62/view?usp=drive_link
-https://drive.google.com/file/d/1848yN3XMN5zJMEgApt6KzrWgfRPfimtv/view?usp=drive_link
-https://drive.google.com/file/d/1oAD9kSnS0fTgj-CjD4u9VdZ5X67IOIMa/view?usp=drive_link
-https://drive.google.com/file/d/1ilzIWLCCG5b_KgF5s0wdN2I5-lFNpwC1/view?usp=drive_link
-https://drive.google.com/file/d/1rjsT2YBjnidxod1s9s-myAYz8boHr-WB/view?usp=drive_link
-https://drive.google.com/file/d/18Gg48HTub15bd8qzbhiCUufbVy0fbN5G/view?usp=drive_link
-https://drive.google.com/file/d/1WsSnQSqmMTVSRwrhT1Y-v782My2zcjLm/view?usp=drive_link
-https://drive.google.com/file/d/1ea9ZCvoyc-xqiFXgeDcA_mOWsw7VUuoi/view?usp=drive_link
-https://drive.google.com/file/d/1wv1v3-XhPgbNzp62BXbJTDzMPu2tlDUc/view?usp=drive_link
-https://drive.google.com/file/d/18-ikzt8LoZ83Gi3goKCELs4U4z8hrRoF/view?usp=drive_link
-https://drive.google.com/file/d/16Bjhp7JNCXkGuLvyNcZowAx3W-Y-15DV/view?usp=drive_link
-https://drive.google.com/file/d/1Gc-KRI-xwcp1fMR55ugbrLg_5y3SPde-/view?usp=drive_link
-https://drive.google.com/file/d/1oP72Q386Z4Sy5MMm-t5yNogIe5Van_9k/view?usp=drive_link
-https://drive.google.com/file/d/112T90eDUDVH-SyOV7UnZl5bscAH2hcfq/view?usp=drive_link
-https://drive.google.com/file/d/1y-uKOesRRhjgDtFbG_j65f4SGg0v8XDg/view?usp=drive_link
-https://drive.google.com/file/d/1LOP05OagoI3km-ZKQBrS204A85UVk7Ok/view?usp=drive_link
-https://drive.google.com/file/d/1QkHQKgasVzWsmdPvkXgGhWyQ84d93_Az/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_pingpong_test.txt

@@ -1 +0,0 @@
-https://drive.google.com/drive/folders/1Ut2cv6o6Pkfgg46DgwVUM7Z5PkNG8eJ-

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_pro_pencil.txt

@@ -1 +0,0 @@
-https://drive.google.com/drive/folders/1FqxPV0PgvgIu8XFjtvZSPSExuNcxVVAY

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_screw_driver.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1SKtG0ct9q0nVdYssJNMWSOjikcXliT58
-https://drive.google.com/file/d/1nchD21O30B3i3LDoqramo1zgW5YvpJIN/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_tape.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1_4DHf2cma0xsChLQFghwigX6Ukti5-zQ
-https://drive.google.com/file/d/1_8vS4hDNDgUQY-SmekrNaa7dF67QJYU-/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_thread_velcro.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1_4DHf2cma0xsChLQFghwigX6Ukti5-zQ
-https://drive.google.com/file/d/1_8vS4hDNDgUQY-SmekrNaa7dF67QJYU-/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_towel.txt

@@ -1,2 +0,0 @@
-https://drive.google.com/drive/folders/1fAD7vkyTGTFB_nGXIKofCU1U05oE3MFv
-https://drive.google.com/file/d/1XzyQ2B6LLvcurIonOpEu4nij2qwNWshH/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_vinh_cup.txt

@@ -1,53 +0,0 @@
-https://drive.google.com/drive/folders/13EQsVsnxT86K20QAoyE_YpsFbQ7fZQdu
-https://drive.google.com/file/d/1-W_JHghZG65FNTVhw1SXhtQrazdLL3Ue/view?usp=drive_link
-https://drive.google.com/file/d/1VwRJgdWUo-2nQaNM7Bs77-fsm8iwUxEo/view?usp=drive_link
-https://drive.google.com/file/d/1wFzGRo5iYA13WLi6IV1ry64RyahQBFio/view?usp=drive_link
-https://drive.google.com/file/d/1IKtQzQ-n-UTv64hYpReu2R4cqUvmNQqD/view?usp=drive_link
-https://drive.google.com/file/d/1GicVci9OiuuZZH79i5Mg7AtWod94MzwT/view?usp=drive_link
-https://drive.google.com/file/d/1JVnIoR7EIQp70T4eAf9RX65JcTrzsjQc/view?usp=drive_link
-https://drive.google.com/file/d/1W2xr4h23ucjPrc-mBEeqnACsfaImpc0p/view?usp=drive_link
-https://drive.google.com/file/d/10xj_0V7A07o3uCa7v5omUrTC0YlPW8H3/view?usp=drive_link
-https://drive.google.com/file/d/1FOc3EMaCy8Mb0_a7PuXLAwKwvxkbKmwU/view?usp=drive_link
-https://drive.google.com/file/d/143PgDXBcf2GQ0Q07ZPMVMfBgZDd5sLJG/view?usp=drive_link
-https://drive.google.com/file/d/1pE5Tyj0LlGbGWvUzuhixp86Ibu55Ez3I/view?usp=drive_link
-https://drive.google.com/file/d/141668b1VzX80ncrVJPzhkoAeIFB4MEK9/view?usp=drive_link
-https://drive.google.com/file/d/1bw12lo37p1ZvRvErHsll7cEYi2OxscvZ/view?usp=drive_link
-https://drive.google.com/file/d/1zfnMFvbgBjl6SzYhksbaOzfbwLrCN6tb/view?usp=drive_link
-https://drive.google.com/file/d/1-GIszA6mUJMaNB-tdh9r9skc77SWA0VX/view?usp=drive_link
-https://drive.google.com/file/d/1fTB0zWFYU6zh4IIUFT2zX_OkwYqmElwY/view?usp=drive_link
-https://drive.google.com/file/d/1gPIPNKGmrO9c7gKF7SP0SuUYbIBBq8z1/view?usp=drive_link
-https://drive.google.com/file/d/12JeJ-dQd5lYyn6PlDOGdE-ChVeiZ-Uv0/view?usp=drive_link
-https://drive.google.com/file/d/100_20cgCqerU6qoh3TfTbwLy9mlDAFEG/view?usp=drive_link
-https://drive.google.com/file/d/111oAGJ76ku_pYgbBoIdZAC1_XEQcPI__/view?usp=drive_link
-https://drive.google.com/file/d/1UhC8L-354ZQ2gblPFGI35EMsVwfpuKa0/view?usp=drive_link
-https://drive.google.com/file/d/1sIXQSgUR_xdrNtGrL6QGBnkLMKErsIp1/view?usp=drive_link
-https://drive.google.com/file/d/16Ax77bDSIXnsn4GFL8XYKKT1P6bPpfMd/view?usp=drive_link
-https://drive.google.com/file/d/1pgRVYwwVIsWq_qsWqZpe1UBzZfF5Fa9D/view?usp=drive_link
-https://drive.google.com/file/d/1jtimaZkWsY1P5gC2bbS64H_WCUU7HXN2/view?usp=drive_link
-https://drive.google.com/file/d/1N6Bh02P-RiTEgtx1YH1Db_X3TGpP-X_r/view?usp=drive_link
-https://drive.google.com/file/d/14Fy8EwJ8d9Vh97Yt1VOvUChSCrfIjBij/view?usp=drive_link
-https://drive.google.com/file/d/1IRuv42dvIMPuKhcMZmuXaBjJ-lPFOmQd/view?usp=drive_link
-https://drive.google.com/file/d/16XWzNY2D8ucVVn5geBgsVdhm3ppO4que/view?usp=drive_link
-https://drive.google.com/file/d/1xsVOoQgthK_L_SDrmq_JvQgUpAvPEAY8/view?usp=drive_link
-https://drive.google.com/file/d/1bZbw66DyEMvnJnzkdUUNbKjvNKg8KFYM/view?usp=drive_link
-https://drive.google.com/file/d/1CyTVkdrNGGpouCXr4CfhKbMzE6Ah3oo3/view?usp=drive_link
-https://drive.google.com/file/d/1hDRyeM-XEDpHXpptbT8LvNnlQUR3PWOh/view?usp=drive_link
-https://drive.google.com/file/d/1XhHWxbra8Iy5irQZ83IvxwaJqHq9x4s1/view?usp=drive_link
-https://drive.google.com/file/d/1haZcn6aM1o4JlmP9tJj3x2enrxiPaDSD/view?usp=drive_link
-https://drive.google.com/file/d/1ypDyuUTbljaBZ34f-t7lj3O_0bRmyX2n/view?usp=drive_link
-https://drive.google.com/file/d/1ILEEZo_tA9_ChIAprr2mPaNVKZi5vXsO/view?usp=drive_link
-https://drive.google.com/file/d/1U7nVYFaGE8vVTfLCW33D74xOjDcqfgyJ/view?usp=drive_link
-https://drive.google.com/file/d/1rZ93_rmCov5SMDxPkfM3qthcRELZrQX6/view?usp=drive_link
-https://drive.google.com/file/d/1mYO1b_csddtyE3qT6cwLiw-m2w2_1Lxh/view?usp=drive_link
-https://drive.google.com/file/d/1xz7Q5x2jikY8wJQjMRQpRws6AnfWlHm5/view?usp=drive_link
-https://drive.google.com/file/d/1OO8GaO-0FrSZRd1kxMYwBmubyiLOWnbl/view?usp=drive_link
-https://drive.google.com/file/d/1EXn4NVDmf-4_HCy34mYwT-vwK2CFI9ev/view?usp=drive_link
-https://drive.google.com/file/d/10hH70XhXRL9C5SnAG4toHtfHqfJUJo4H/view?usp=drive_link
-https://drive.google.com/file/d/18tiBcxea0guUai4lwsXQvt0q2LZ8ZnnJ/view?usp=drive_link
-https://drive.google.com/file/d/1Q8R8qv37vk5PQ5kQ2ibx6BFLOySD0VpX/view?usp=drive_link
-https://drive.google.com/file/d/17aNriHzjhdibCyuUjQoMFZqjybJZtggG/view?usp=drive_link
-https://drive.google.com/file/d/1LVjEYHSdeKm6CotU1QguIeNEPaIaFl_1/view?usp=drive_link
-https://drive.google.com/file/d/1ufAhE_EkgJ85slg2EW8aW_grOzE_Lmxd/view?usp=drive_link
-https://drive.google.com/file/d/1wtzLtXrkw9eXRGESTPIOlpl1tInu-b2m/view?usp=drive_link
-https://drive.google.com/file/d/1Mk5qvVtD_QHwGOUApRq76TUw2T5THu6f/view?usp=drive_link
-https://drive.google.com/file/d/1y1WQ3hboWVJ68KEYQQ3OhreGuaUpSgwc/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_vinh_cup_left.txt

@@ -1,52 +0,0 @@
-https://drive.google.com/drive/folders/1dxWh6YFZUDt6qXIoxgD9bla3CiFjZ11C
-https://drive.google.com/file/d/1hNBJN00SCAlOl0ZEgm7RRGbAGDjyBs0p/view?usp=drive_link
-https://drive.google.com/file/d/17He0CVwXGeoMmXg4SHKo-osNn7YPKVL7/view?usp=drive_link
-https://drive.google.com/file/d/1laNKUVID1x2CV6a2O2WQjwFewKu4lidL/view?usp=drive_link
-https://drive.google.com/file/d/1pNf36xbZJGRArYLmNAvRj5y6CoqdC6kB/view?usp=drive_link
-https://drive.google.com/file/d/1_4E1-y3JXk5I0ebycLYM70YDPK9g52gZ/view?usp=drive_link
-https://drive.google.com/file/d/1PHfzhGPdbolKyOpS3FnR2w7Q8zUlJXSk/view?usp=drive_link
-https://drive.google.com/file/d/17ls2PPN-Pi3tEuK059cwV2_iDT8aGhOO/view?usp=drive_link
-https://drive.google.com/file/d/1LWsg6PmCT00Kv_N_slrmcwKmQPGoBT3k/view?usp=drive_link
-https://drive.google.com/file/d/12LckrchoHTUVH7rxi8J7zD9dA19GXvoW/view?usp=drive_link
-https://drive.google.com/file/d/1VqrJKjAIkj5gtFXL69grdSeu9CyaqnSw/view?usp=drive_link
-https://drive.google.com/file/d/1g5rQYDBZvW-kUtYPeyF3qmd53v6k7kXu/view?usp=drive_link
-https://drive.google.com/file/d/10kUgaSJ0TS7teaG83G3Rf_DG4XGrBt6A/view?usp=drive_link
-https://drive.google.com/file/d/1je9XmneZQZvTma5adMJICUPDovW3ppei/view?usp=drive_link
-https://drive.google.com/file/d/1v28r6bedwZGbUPVVTVImXhK-42XdtGfj/view?usp=drive_link
-https://drive.google.com/file/d/1-TEEx9sGVvzMMaNXYfQMtY2JJ6cvl0dT/view?usp=drive_link
-https://drive.google.com/file/d/1YdBKdJFP9rJWBUX7qrOYL_gfUA8o6J9M/view?usp=drive_link
-https://drive.google.com/file/d/1X9vffwQHNUSKLXr2RlYNtbWDIFCIDfdF/view?usp=drive_link
-https://drive.google.com/file/d/11hqesqa5kvEe5FABUnZRcvmOhR373cYM/view?usp=drive_link
-https://drive.google.com/file/d/1ltTTECjEcbQPgS3UPRgMzaE2x9n6H7dC/view?usp=drive_link
-https://drive.google.com/file/d/1Zxqfa29JdwT-bfMpivi6IG2vz34d21dD/view?usp=drive_link
-https://drive.google.com/file/d/11LQlVxS5hz494dYUJ_PNRPx2NHIJbQns/view?usp=drive_link
-https://drive.google.com/file/d/1i1JhNtnZpO_E8rAv8gxBP3ZTZRvcvsZi/view?usp=drive_link
-https://drive.google.com/file/d/11jOXAr2EULUO4Qkm748634lg4UUFho5U/view?usp=drive_link
-https://drive.google.com/file/d/1rj67wur8DdB_Pipwx24bY43xu4X1eQ5e/view?usp=drive_link
-https://drive.google.com/file/d/15ZTm6lO6f_JQy_4SNfrOu3iPYn1Ro8mh/view?usp=drive_link
-https://drive.google.com/file/d/1q4gBtqWPJtCwXEvknGgN0WHGp7Vfn1b9/view?usp=drive_link
-https://drive.google.com/file/d/1t17keyre47AYqm8GgXiQ7EcvcUkeSiDQ/view?usp=drive_link
-https://drive.google.com/file/d/1OYUPGxtZgOF86Ng_BEOTXm_XOYpuQPsO/view?usp=drive_link
-https://drive.google.com/file/d/1cBjbGHi3dwWHtx6r9EQJi0JT_CE3LuHt/view?usp=drive_link
-https://drive.google.com/file/d/14qaMyF0mcbCB-fCYKNyo5_2NahSC6D5u/view?usp=drive_link
-https://drive.google.com/file/d/12FgX86eA7Y5co9ULBVK80XMsiKQSs-Ri/view?usp=drive_link
-https://drive.google.com/file/d/1yvoHWidf-jdBVw6qCCXOFfkVwKj_2hPk/view?usp=drive_link
-https://drive.google.com/file/d/1a2SugsSDlC8UtUrFzp-_KAwyZckQOvdQ/view?usp=drive_link
-https://drive.google.com/file/d/1l8pILBFSAosypWJMza2K09Vm7rug9axm/view?usp=drive_link
-https://drive.google.com/file/d/1hfPQ8dBCk97PnOhq6_MIISm3IEzcOxJG/view?usp=drive_link
-https://drive.google.com/file/d/1PPAUwlJCFKpms8cqF_k1v2_fCgDBOc3S/view?usp=drive_link
-https://drive.google.com/file/d/1lVKQZeqFfK3amEmLuFhYLUFQ2eyE8rOW/view?usp=drive_link
-https://drive.google.com/file/d/1K9iPMLfDowcIFoyzpvgn88dQ6x6kVwNG/view?usp=drive_link
-https://drive.google.com/file/d/1PNvMqG9tL7QxeLaYBGHiWYR6SYb5iIct/view?usp=drive_link
-https://drive.google.com/file/d/1xkRtzbvIkUsylx9hrFLGQsJn0h1EYu-5/view?usp=drive_link
-https://drive.google.com/file/d/1nxMRrJlSayjDIfr5CmHO1NzAw3COhsLi/view?usp=drive_link
-https://drive.google.com/file/d/1Qs3WEyMGrmagiHIkkFEueWNnJhkUeR1s/view?usp=drive_link
-https://drive.google.com/file/d/1D-G2_Q0SS3M8zyJbg_XzkF2ANPw1HTuX/view?usp=drive_link
-https://drive.google.com/file/d/1mdmJsDGO-YtJAOF_yPKl6lq4PJOIbQhT/view?usp=drive_link
-https://drive.google.com/file/d/11m9bwfop_sPmnQr_8amB6EEsrbAeG_z5/view?usp=drive_link
-https://drive.google.com/file/d/19tyYt5FMn5kru0g9o2nMJhKPnsDqkIZv/view?usp=drive_link
-https://drive.google.com/file/d/1XvTpUdsVTZ-vydvdYYmynbma--HfUGSl/view?usp=drive_link
-https://drive.google.com/file/d/1MO3hFu68J6NohTzr9aB_fY02VA6QSOqj/view?usp=drive_link
-https://drive.google.com/file/d/1Lh-UjwAk__04YOTWINF_QGVU8SjetVaY/view?usp=drive_link
-https://drive.google.com/file/d/1jkSOUwZV5GJ7rZlVeErjcu0DBQs8Np0d/view?usp=drive_link
-https://drive.google.com/file/d/1VIN1eLI-93WrVQwCjsv6XQr353DqqBYA/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_aloha_raw_urls/static_ziploc_slide.txt

@@ -1,8 +0,0 @@
-https://drive.google.com/drive/folders/1EgKar7rWBmTIRmeJYZciSwjZx3uP2mHO
-https://drive.google.com/file/d/12eYWQO15atK2hBjXhynPJd9MKAj_42pz/view?usp=drive_link
-https://drive.google.com/file/d/1Ul4oEeICJDjgfYTl4H1uaisTzVYIM6wd/view?usp=drive_link
-https://drive.google.com/file/d/1WSF-OG8lKSe2wVYCv5D1aJNipxpgddk-/view?usp=drive_link
-https://drive.google.com/file/d/1_ppD5j5sFh26aWW0JmhLzJMeNB-lCArk/view?usp=drive_link
-https://drive.google.com/file/d/1WUp846dgWXYhu4oJfhHxiU6YL_7N6s4W/view?usp=drive_link
-https://drive.google.com/file/d/1HRZNAIoAQw_uYiPwnBvtBioQoqiqoXdA/view?usp=drive_link
-https://drive.google.com/file/d/1hedGq-QDMnIn8GlXXBC3GiEJ_Y-LTxyt/view?usp=drive_link

lerobot/common/datasets/push_dataset_to_hub/_diffusion_policy_replay_buffer.py

@@ -1,634 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Helper code for loading PushT dataset from Diffusion Policy (https://diffusion-policy.cs.columbia.edu/)
-
-Copied from the original Diffusion Policy repository and used in our `download_and_upload_dataset.py` script.
-"""
-
-from __future__ import annotations
-
-import math
-import numbers
-import os
-from functools import cached_property
-
-import numcodecs
-import numpy as np
-import zarr
-
-
-def check_chunks_compatible(chunks: tuple, shape: tuple):
-    assert len(shape) == len(chunks)
-    for c in chunks:
-        assert isinstance(c, numbers.Integral)
-        assert c > 0
-
-
-def rechunk_recompress_array(group, name, chunks=None, chunk_length=None, compressor=None, tmp_key="_temp"):
-    old_arr = group[name]
-    if chunks is None:
-        chunks = (chunk_length,) + old_arr.chunks[1:] if chunk_length is not None else old_arr.chunks
-    check_chunks_compatible(chunks, old_arr.shape)
-
-    if compressor is None:
-        compressor = old_arr.compressor
-
-    if (chunks == old_arr.chunks) and (compressor == old_arr.compressor):
-        # no change
-        return old_arr
-
-    # rechunk recompress
-    group.move(name, tmp_key)
-    old_arr = group[tmp_key]
-    n_copied, n_skipped, n_bytes_copied = zarr.copy(
-        source=old_arr,
-        dest=group,
-        name=name,
-        chunks=chunks,
-        compressor=compressor,
-    )
-    del group[tmp_key]
-    arr = group[name]
-    return arr
-
-
-def get_optimal_chunks(shape, dtype, target_chunk_bytes=2e6, max_chunk_length=None):
-    """
-    Common shapes
-    T,D
-    T,N,D
-    T,H,W,C
-    T,N,H,W,C
-    """
-    itemsize = np.dtype(dtype).itemsize
-    # reversed
-    rshape = list(shape[::-1])
-    if max_chunk_length is not None:
-        rshape[-1] = int(max_chunk_length)
-    split_idx = len(shape) - 1
-    for i in range(len(shape) - 1):
-        this_chunk_bytes = itemsize * np.prod(rshape[:i])
-        next_chunk_bytes = itemsize * np.prod(rshape[: i + 1])
-        if this_chunk_bytes <= target_chunk_bytes and next_chunk_bytes > target_chunk_bytes:
-            split_idx = i
-
-    rchunks = rshape[:split_idx]
-    item_chunk_bytes = itemsize * np.prod(rshape[:split_idx])
-    this_max_chunk_length = rshape[split_idx]
-    next_chunk_length = min(this_max_chunk_length, math.ceil(target_chunk_bytes / item_chunk_bytes))
-    rchunks.append(next_chunk_length)
-    len_diff = len(shape) - len(rchunks)
-    rchunks.extend([1] * len_diff)
-    chunks = tuple(rchunks[::-1])
-    # print(np.prod(chunks) * itemsize / target_chunk_bytes)
-    return chunks
-
-
-class ReplayBuffer:
-    """
-    Zarr-based temporal datastructure.
-    Assumes first dimension to be time. Only chunk in time dimension.
-    """
-
-    def __init__(self, root: zarr.Group | dict[str, dict]):
-        """
-        Dummy constructor. Use copy_from* and create_from* class methods instead.
-        """
-        assert "data" in root
-        assert "meta" in root
-        assert "episode_ends" in root["meta"]
-        for value in root["data"].values():
-            assert value.shape[0] == root["meta"]["episode_ends"][-1]
-        self.root = root
-
-    # ============= create constructors ===============
-    @classmethod
-    def create_empty_zarr(cls, storage=None, root=None):
-        if root is None:
-            if storage is None:
-                storage = zarr.MemoryStore()
-            root = zarr.group(store=storage)
-        root.require_group("data", overwrite=False)
-        meta = root.require_group("meta", overwrite=False)
-        if "episode_ends" not in meta:
-            meta.zeros("episode_ends", shape=(0,), dtype=np.int64, compressor=None, overwrite=False)
-        return cls(root=root)
-
-    @classmethod
-    def create_empty_numpy(cls):
-        root = {"data": {}, "meta": {"episode_ends": np.zeros((0,), dtype=np.int64)}}
-        return cls(root=root)
-
-    @classmethod
-    def create_from_group(cls, group, **kwargs):
-        if "data" not in group:
-            # create from stratch
-            buffer = cls.create_empty_zarr(root=group, **kwargs)
-        else:
-            # already exist
-            buffer = cls(root=group, **kwargs)
-        return buffer
-
-    @classmethod
-    def create_from_path(cls, zarr_path, mode="r", **kwargs):
-        """
-        Open a on-disk zarr directly (for dataset larger than memory).
-        Slower.
-        """
-        group = zarr.open(os.path.expanduser(zarr_path), mode)
-        return cls.create_from_group(group, **kwargs)
-
-    # ============= copy constructors ===============
-    @classmethod
-    def copy_from_store(
-        cls,
-        src_store,
-        store=None,
-        keys=None,
-        chunks: dict[str, tuple] | None = None,
-        compressors: dict | str | numcodecs.abc.Codec | None = None,
-        if_exists="replace",
-        **kwargs,
-    ):
-        """
-        Load to memory.
-        """
-        src_root = zarr.group(src_store)
-        if chunks is None:
-            chunks = {}
-        if compressors is None:
-            compressors = {}
-        root = None
-        if store is None:
-            # numpy backend
-            meta = {}
-            for key, value in src_root["meta"].items():
-                if len(value.shape) == 0:
-                    meta[key] = np.array(value)
-                else:
-                    meta[key] = value[:]
-
-            if keys is None:
-                keys = src_root["data"].keys()
-            data = {}
-            for key in keys:
-                arr = src_root["data"][key]
-                data[key] = arr[:]
-
-            root = {"meta": meta, "data": data}
-        else:
-            root = zarr.group(store=store)
-            # copy without recompression
-            n_copied, n_skipped, n_bytes_copied = zarr.copy_store(
-                source=src_store, dest=store, source_path="/meta", dest_path="/meta", if_exists=if_exists
-            )
-            data_group = root.create_group("data", overwrite=True)
-            if keys is None:
-                keys = src_root["data"].keys()
-            for key in keys:
-                value = src_root["data"][key]
-                cks = cls._resolve_array_chunks(chunks=chunks, key=key, array=value)
-                cpr = cls._resolve_array_compressor(compressors=compressors, key=key, array=value)
-                if cks == value.chunks and cpr == value.compressor:
-                    # copy without recompression
-                    this_path = "/data/" + key
-                    n_copied, n_skipped, n_bytes_copied = zarr.copy_store(
-                        source=src_store,
-                        dest=store,
-                        source_path=this_path,
-                        dest_path=this_path,
-                        if_exists=if_exists,
-                    )
-                else:
-                    # copy with recompression
-                    n_copied, n_skipped, n_bytes_copied = zarr.copy(
-                        source=value,
-                        dest=data_group,
-                        name=key,
-                        chunks=cks,
-                        compressor=cpr,
-                        if_exists=if_exists,
-                    )
-        buffer = cls(root=root)
-        return buffer
-
-    @classmethod
-    def copy_from_path(
-        cls,
-        zarr_path,
-        backend=None,
-        store=None,
-        keys=None,
-        chunks: dict[str, tuple] | None = None,
-        compressors: dict | str | numcodecs.abc.Codec | None = None,
-        if_exists="replace",
-        **kwargs,
-    ):
-        """
-        Copy a on-disk zarr to in-memory compressed.
-        Recommended
-        """
-        if chunks is None:
-            chunks = {}
-        if compressors is None:
-            compressors = {}
-        if backend == "numpy":
-            print("backend argument is deprecated!")
-            store = None
-        group = zarr.open(os.path.expanduser(zarr_path), "r")
-        return cls.copy_from_store(
-            src_store=group.store,
-            store=store,
-            keys=keys,
-            chunks=chunks,
-            compressors=compressors,
-            if_exists=if_exists,
-            **kwargs,
-        )
-
-    # ============= save methods ===============
-    def save_to_store(
-        self,
-        store,
-        chunks: dict[str, tuple] | None = None,
-        compressors: str | numcodecs.abc.Codec | dict | None = None,
-        if_exists="replace",
-        **kwargs,
-    ):
-        root = zarr.group(store)
-        if chunks is None:
-            chunks = {}
-        if compressors is None:
-            compressors = {}
-        if self.backend == "zarr":
-            # recompression free copy
-            n_copied, n_skipped, n_bytes_copied = zarr.copy_store(
-                source=self.root.store,
-                dest=store,
-                source_path="/meta",
-                dest_path="/meta",
-                if_exists=if_exists,
-            )
-        else:
-            meta_group = root.create_group("meta", overwrite=True)
-            # save meta, no chunking
-            for key, value in self.root["meta"].items():
-                _ = meta_group.array(name=key, data=value, shape=value.shape, chunks=value.shape)
-
-        # save data, chunk
-        data_group = root.create_group("data", overwrite=True)
-        for key, value in self.root["data"].items():
-            cks = self._resolve_array_chunks(chunks=chunks, key=key, array=value)
-            cpr = self._resolve_array_compressor(compressors=compressors, key=key, array=value)
-            if isinstance(value, zarr.Array):
-                if cks == value.chunks and cpr == value.compressor:
-                    # copy without recompression
-                    this_path = "/data/" + key
-                    n_copied, n_skipped, n_bytes_copied = zarr.copy_store(
-                        source=self.root.store,
-                        dest=store,
-                        source_path=this_path,
-                        dest_path=this_path,
-                        if_exists=if_exists,
-                    )
-                else:
-                    # copy with recompression
-                    n_copied, n_skipped, n_bytes_copied = zarr.copy(
-                        source=value,
-                        dest=data_group,
-                        name=key,
-                        chunks=cks,
-                        compressor=cpr,
-                        if_exists=if_exists,
-                    )
-            else:
-                # numpy
-                _ = data_group.array(name=key, data=value, chunks=cks, compressor=cpr)
-        return store
-
-    def save_to_path(
-        self,
-        zarr_path,
-        chunks: dict[str, tuple] | None = None,
-        compressors: str | numcodecs.abc.Codec | dict | None = None,
-        if_exists="replace",
-        **kwargs,
-    ):
-        if chunks is None:
-            chunks = {}
-        if compressors is None:
-            compressors = {}
-        store = zarr.DirectoryStore(os.path.expanduser(zarr_path))
-        return self.save_to_store(
-            store, chunks=chunks, compressors=compressors, if_exists=if_exists, **kwargs
-        )
-
-    @staticmethod
-    def resolve_compressor(compressor="default"):
-        if compressor == "default":
-            compressor = numcodecs.Blosc(cname="lz4", clevel=5, shuffle=numcodecs.Blosc.NOSHUFFLE)
-        elif compressor == "disk":
-            compressor = numcodecs.Blosc("zstd", clevel=5, shuffle=numcodecs.Blosc.BITSHUFFLE)
-        return compressor
-
-    @classmethod
-    def _resolve_array_compressor(cls, compressors: dict | str | numcodecs.abc.Codec, key, array):
-        # allows compressor to be explicitly set to None
-        cpr = "nil"
-        if isinstance(compressors, dict):
-            if key in compressors:
-                cpr = cls.resolve_compressor(compressors[key])
-            elif isinstance(array, zarr.Array):
-                cpr = array.compressor
-        else:
-            cpr = cls.resolve_compressor(compressors)
-        # backup default
-        if cpr == "nil":
-            cpr = cls.resolve_compressor("default")
-        return cpr
-
-    @classmethod
-    def _resolve_array_chunks(cls, chunks: dict | tuple, key, array):
-        cks = None
-        if isinstance(chunks, dict):
-            if key in chunks:
-                cks = chunks[key]
-            elif isinstance(array, zarr.Array):
-                cks = array.chunks
-        elif isinstance(chunks, tuple):
-            cks = chunks
-        else:
-            raise TypeError(f"Unsupported chunks type {type(chunks)}")
-        # backup default
-        if cks is None:
-            cks = get_optimal_chunks(shape=array.shape, dtype=array.dtype)
-        # check
-        check_chunks_compatible(chunks=cks, shape=array.shape)
-        return cks
-
-    # ============= properties =================
-    @cached_property
-    def data(self):
-        return self.root["data"]
-
-    @cached_property
-    def meta(self):
-        return self.root["meta"]
-
-    def update_meta(self, data):
-        # sanitize data
-        np_data = {}
-        for key, value in data.items():
-            if isinstance(value, np.ndarray):
-                np_data[key] = value
-            else:
-                arr = np.array(value)
-                if arr.dtype == object:
-                    raise TypeError(f"Invalid value type {type(value)}")
-                np_data[key] = arr
-
-        meta_group = self.meta
-        if self.backend == "zarr":
-            for key, value in np_data.items():
-                _ = meta_group.array(
-                    name=key, data=value, shape=value.shape, chunks=value.shape, overwrite=True
-                )
-        else:
-            meta_group.update(np_data)
-
-        return meta_group
-
-    @property
-    def episode_ends(self):
-        return self.meta["episode_ends"]
-
-    def get_episode_idxs(self):
-        import numba
-
-        numba.jit(nopython=True)
-
-        def _get_episode_idxs(episode_ends):
-            result = np.zeros((episode_ends[-1],), dtype=np.int64)
-            for i in range(len(episode_ends)):
-                start = 0
-                if i > 0:
-                    start = episode_ends[i - 1]
-                end = episode_ends[i]
-                for idx in range(start, end):
-                    result[idx] = i
-            return result
-
-        return _get_episode_idxs(self.episode_ends)
-
-    @property
-    def backend(self):
-        backend = "numpy"
-        if isinstance(self.root, zarr.Group):
-            backend = "zarr"
-        return backend
-
-    # =========== dict-like API ==============
-    def __repr__(self) -> str:
-        if self.backend == "zarr":
-            return str(self.root.tree())
-        else:
-            return super().__repr__()
-
-    def keys(self):
-        return self.data.keys()
-
-    def values(self):
-        return self.data.values()
-
-    def items(self):
-        return self.data.items()
-
-    def __getitem__(self, key):
-        return self.data[key]
-
-    def __contains__(self, key):
-        return key in self.data
-
-    # =========== our API ==============
-    @property
-    def n_steps(self):
-        if len(self.episode_ends) == 0:
-            return 0
-        return self.episode_ends[-1]
-
-    @property
-    def n_episodes(self):
-        return len(self.episode_ends)
-
-    @property
-    def chunk_size(self):
-        if self.backend == "zarr":
-            return next(iter(self.data.arrays()))[-1].chunks[0]
-        return None
-
-    @property
-    def episode_lengths(self):
-        ends = self.episode_ends[:]
-        ends = np.insert(ends, 0, 0)
-        lengths = np.diff(ends)
-        return lengths
-
-    def add_episode(
-        self,
-        data: dict[str, np.ndarray],
-        chunks: dict[str, tuple] | None = None,
-        compressors: str | numcodecs.abc.Codec | dict | None = None,
-    ):
-        if chunks is None:
-            chunks = {}
-        if compressors is None:
-            compressors = {}
-        assert len(data) > 0
-        is_zarr = self.backend == "zarr"
-
-        curr_len = self.n_steps
-        episode_length = None
-        for value in data.values():
-            assert len(value.shape) >= 1
-            if episode_length is None:
-                episode_length = len(value)
-            else:
-                assert episode_length == len(value)
-        new_len = curr_len + episode_length
-
-        for key, value in data.items():
-            new_shape = (new_len,) + value.shape[1:]
-            # create array
-            if key not in self.data:
-                if is_zarr:
-                    cks = self._resolve_array_chunks(chunks=chunks, key=key, array=value)
-                    cpr = self._resolve_array_compressor(compressors=compressors, key=key, array=value)
-                    arr = self.data.zeros(
-                        name=key, shape=new_shape, chunks=cks, dtype=value.dtype, compressor=cpr
-                    )
-                else:
-                    # copy data to prevent modify
-                    arr = np.zeros(shape=new_shape, dtype=value.dtype)
-                    self.data[key] = arr
-            else:
-                arr = self.data[key]
-                assert value.shape[1:] == arr.shape[1:]
-                # same method for both zarr and numpy
-                if is_zarr:
-                    arr.resize(new_shape)
-                else:
-                    arr.resize(new_shape, refcheck=False)
-            # copy data
-            arr[-value.shape[0] :] = value
-
-        # append to episode ends
-        episode_ends = self.episode_ends
-        if is_zarr:
-            episode_ends.resize(episode_ends.shape[0] + 1)
-        else:
-            episode_ends.resize(episode_ends.shape[0] + 1, refcheck=False)
-        episode_ends[-1] = new_len
-
-        # rechunk
-        if is_zarr and episode_ends.chunks[0] < episode_ends.shape[0]:
-            rechunk_recompress_array(self.meta, "episode_ends", chunk_length=int(episode_ends.shape[0] * 1.5))
-
-    def drop_episode(self):
-        is_zarr = self.backend == "zarr"
-        episode_ends = self.episode_ends[:].copy()
-        assert len(episode_ends) > 0
-        start_idx = 0
-        if len(episode_ends) > 1:
-            start_idx = episode_ends[-2]
-        for value in self.data.values():
-            new_shape = (start_idx,) + value.shape[1:]
-            if is_zarr:
-                value.resize(new_shape)
-            else:
-                value.resize(new_shape, refcheck=False)
-        if is_zarr:
-            self.episode_ends.resize(len(episode_ends) - 1)
-        else:
-            self.episode_ends.resize(len(episode_ends) - 1, refcheck=False)
-
-    def pop_episode(self):
-        assert self.n_episodes > 0
-        episode = self.get_episode(self.n_episodes - 1, copy=True)
-        self.drop_episode()
-        return episode
-
-    def extend(self, data):
-        self.add_episode(data)
-
-    def get_episode(self, idx, copy=False):
-        idx = list(range(len(self.episode_ends)))[idx]
-        start_idx = 0
-        if idx > 0:
-            start_idx = self.episode_ends[idx - 1]
-        end_idx = self.episode_ends[idx]
-        result = self.get_steps_slice(start_idx, end_idx, copy=copy)
-        return result
-
-    def get_episode_slice(self, idx):
-        start_idx = 0
-        if idx > 0:
-            start_idx = self.episode_ends[idx - 1]
-        end_idx = self.episode_ends[idx]
-        return slice(start_idx, end_idx)
-
-    def get_steps_slice(self, start, stop, step=None, copy=False):
-        _slice = slice(start, stop, step)
-
-        result = {}
-        for key, value in self.data.items():
-            x = value[_slice]
-            if copy and isinstance(value, np.ndarray):
-                x = x.copy()
-            result[key] = x
-        return result
-
-    # =========== chunking =============
-    def get_chunks(self) -> dict:
-        assert self.backend == "zarr"
-        chunks = {}
-        for key, value in self.data.items():
-            chunks[key] = value.chunks
-        return chunks
-
-    def set_chunks(self, chunks: dict):
-        assert self.backend == "zarr"
-        for key, value in chunks.items():
-            if key in self.data:
-                arr = self.data[key]
-                if value != arr.chunks:
-                    check_chunks_compatible(chunks=value, shape=arr.shape)
-                    rechunk_recompress_array(self.data, key, chunks=value)
-
-    def get_compressors(self) -> dict:
-        assert self.backend == "zarr"
-        compressors = {}
-        for key, value in self.data.items():
-            compressors[key] = value.compressor
-        return compressors
-
-    def set_compressors(self, compressors: dict):
-        assert self.backend == "zarr"
-        for key, value in compressors.items():
-            if key in self.data:
-                arr = self.data[key]
-                compressor = self.resolve_compressor(value)
-                if compressor != arr.compressor:
-                    rechunk_recompress_array(self.data, key, compressor=compressor)

lerobot/common/datasets/push_dataset_to_hub/_download_raw.py

@@ -1,202 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-This file contains download scripts for raw datasets.
-
-Example of usage:
-```
-python lerobot/common/datasets/push_dataset_to_hub/_download_raw.py \
---raw-dir data/lerobot-raw/pusht_raw \
---repo-id lerobot-raw/pusht_raw
-```
-"""
-
-import argparse
-import logging
-import warnings
-from pathlib import Path
-
-from huggingface_hub import snapshot_download
-
-from lerobot.common.datasets.push_dataset_to_hub.utils import check_repo_id
-
-# {raw_repo_id: raw_format}
-AVAILABLE_RAW_REPO_IDS = {
-    "lerobot-raw/aloha_mobile_cabinet_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_mobile_chair_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_mobile_elevator_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_mobile_shrimp_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_mobile_wash_pan_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_mobile_wipe_wine_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_sim_insertion_human_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_sim_insertion_scripted_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_sim_transfer_cube_human_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_sim_transfer_cube_scripted_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_battery_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_candy_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_coffee_new_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_coffee_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_cups_open_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_fork_pick_up_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_pingpong_test_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_pro_pencil_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_screw_driver_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_tape_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_thread_velcro_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_towel_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_vinh_cup_left_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_vinh_cup_raw": "aloha_hdf5",
-    "lerobot-raw/aloha_static_ziploc_slide_raw": "aloha_hdf5",
-    "lerobot-raw/umi_cup_in_the_wild_raw": "umi_zarr",
-    "lerobot-raw/pusht_raw": "pusht_zarr",
-    "lerobot-raw/unitreeh1_fold_clothes_raw": "aloha_hdf5",
-    "lerobot-raw/unitreeh1_rearrange_objects_raw": "aloha_hdf5",
-    "lerobot-raw/unitreeh1_two_robot_greeting_raw": "aloha_hdf5",
-    "lerobot-raw/unitreeh1_warehouse_raw": "aloha_hdf5",
-    "lerobot-raw/xarm_lift_medium_raw": "xarm_pkl",
-    "lerobot-raw/xarm_lift_medium_replay_raw": "xarm_pkl",
-    "lerobot-raw/xarm_push_medium_raw": "xarm_pkl",
-    "lerobot-raw/xarm_push_medium_replay_raw": "xarm_pkl",
-    "lerobot-raw/fractal20220817_data_raw": "openx_rlds.fractal20220817_data",
-    "lerobot-raw/kuka_raw": "openx_rlds.kuka",
-    "lerobot-raw/bridge_openx_raw": "openx_rlds.bridge_openx",
-    "lerobot-raw/taco_play_raw": "openx_rlds.taco_play",
-    "lerobot-raw/jaco_play_raw": "openx_rlds.jaco_play",
-    "lerobot-raw/berkeley_cable_routing_raw": "openx_rlds.berkeley_cable_routing",
-    "lerobot-raw/roboturk_raw": "openx_rlds.roboturk",
-    "lerobot-raw/nyu_door_opening_surprising_effectiveness_raw": "openx_rlds.nyu_door_opening_surprising_effectiveness",
-    "lerobot-raw/viola_raw": "openx_rlds.viola",
-    "lerobot-raw/berkeley_autolab_ur5_raw": "openx_rlds.berkeley_autolab_ur5",
-    "lerobot-raw/toto_raw": "openx_rlds.toto",
-    "lerobot-raw/language_table_raw": "openx_rlds.language_table",
-    "lerobot-raw/columbia_cairlab_pusht_real_raw": "openx_rlds.columbia_cairlab_pusht_real",
-    "lerobot-raw/stanford_kuka_multimodal_dataset_raw": "openx_rlds.stanford_kuka_multimodal_dataset",
-    "lerobot-raw/nyu_rot_dataset_raw": "openx_rlds.nyu_rot_dataset",
-    "lerobot-raw/io_ai_tech_raw": "openx_rlds.io_ai_tech",
-    "lerobot-raw/stanford_hydra_dataset_raw": "openx_rlds.stanford_hydra_dataset",
-    "lerobot-raw/austin_buds_dataset_raw": "openx_rlds.austin_buds_dataset",
-    "lerobot-raw/nyu_franka_play_dataset_raw": "openx_rlds.nyu_franka_play_dataset",
-    "lerobot-raw/maniskill_dataset_raw": "openx_rlds.maniskill_dataset",
-    "lerobot-raw/furniture_bench_dataset_raw": "openx_rlds.furniture_bench_dataset",
-    "lerobot-raw/cmu_franka_exploration_dataset_raw": "openx_rlds.cmu_franka_exploration_dataset",
-    "lerobot-raw/ucsd_kitchen_dataset_raw": "openx_rlds.ucsd_kitchen_dataset",
-    "lerobot-raw/ucsd_pick_and_place_dataset_raw": "openx_rlds.ucsd_pick_and_place_dataset",
-    "lerobot-raw/spoc_raw": "openx_rlds.spoc",
-    "lerobot-raw/austin_sailor_dataset_raw": "openx_rlds.austin_sailor_dataset",
-    "lerobot-raw/austin_sirius_dataset_raw": "openx_rlds.austin_sirius_dataset",
-    "lerobot-raw/bc_z_raw": "openx_rlds.bc_z",
-    "lerobot-raw/utokyo_pr2_opening_fridge_raw": "openx_rlds.utokyo_pr2_opening_fridge",
-    "lerobot-raw/utokyo_pr2_tabletop_manipulation_raw": "openx_rlds.utokyo_pr2_tabletop_manipulation",
-    "lerobot-raw/utokyo_xarm_pick_and_place_raw": "openx_rlds.utokyo_xarm_pick_and_place",
-    "lerobot-raw/utokyo_xarm_bimanual_raw": "openx_rlds.utokyo_xarm_bimanual",
-    "lerobot-raw/utokyo_saytap_raw": "openx_rlds.utokyo_saytap",
-    "lerobot-raw/robo_net_raw": "openx_rlds.robo_net",
-    "lerobot-raw/robo_set_raw": "openx_rlds.robo_set",
-    "lerobot-raw/berkeley_mvp_raw": "openx_rlds.berkeley_mvp",
-    "lerobot-raw/berkeley_rpt_raw": "openx_rlds.berkeley_rpt",
-    "lerobot-raw/kaist_nonprehensile_raw": "openx_rlds.kaist_nonprehensile",
-    "lerobot-raw/stanford_mask_vit_raw": "openx_rlds.stanford_mask_vit",
-    "lerobot-raw/tokyo_u_lsmo_raw": "openx_rlds.tokyo_u_lsmo",
-    "lerobot-raw/dlr_sara_pour_raw": "openx_rlds.dlr_sara_pour",
-    "lerobot-raw/dlr_sara_grid_clamp_raw": "openx_rlds.dlr_sara_grid_clamp",
-    "lerobot-raw/dlr_edan_shared_control_raw": "openx_rlds.dlr_edan_shared_control",
-    "lerobot-raw/asu_table_top_raw": "openx_rlds.asu_table_top",
-    "lerobot-raw/stanford_robocook_raw": "openx_rlds.stanford_robocook",
-    "lerobot-raw/imperialcollege_sawyer_wrist_cam_raw": "openx_rlds.imperialcollege_sawyer_wrist_cam",
-    "lerobot-raw/iamlab_cmu_pickup_insert_raw": "openx_rlds.iamlab_cmu_pickup_insert",
-    "lerobot-raw/uiuc_d3field_raw": "openx_rlds.uiuc_d3field",
-    "lerobot-raw/utaustin_mutex_raw": "openx_rlds.utaustin_mutex",
-    "lerobot-raw/berkeley_fanuc_manipulation_raw": "openx_rlds.berkeley_fanuc_manipulation",
-    "lerobot-raw/cmu_playing_with_food_raw": "openx_rlds.cmu_playing_with_food",
-    "lerobot-raw/cmu_play_fusion_raw": "openx_rlds.cmu_play_fusion",
-    "lerobot-raw/cmu_stretch_raw": "openx_rlds.cmu_stretch",
-    "lerobot-raw/berkeley_gnm_recon_raw": "openx_rlds.berkeley_gnm_recon",
-    "lerobot-raw/berkeley_gnm_cory_hall_raw": "openx_rlds.berkeley_gnm_cory_hall",
-    "lerobot-raw/berkeley_gnm_sac_son_raw": "openx_rlds.berkeley_gnm_sac_son",
-    "lerobot-raw/droid_raw": "openx_rlds.droid",
-    "lerobot-raw/droid_100_raw": "openx_rlds.droid100",
-    "lerobot-raw/fmb_raw": "openx_rlds.fmb",
-    "lerobot-raw/dobbe_raw": "openx_rlds.dobbe",
-    "lerobot-raw/usc_cloth_sim_raw": "openx_rlds.usc_cloth_sim",
-    "lerobot-raw/plex_robosuite_raw": "openx_rlds.plex_robosuite",
-    "lerobot-raw/conq_hose_manipulation_raw": "openx_rlds.conq_hose_manipulation",
-    "lerobot-raw/vima_raw": "openx_rlds.vima",
-    "lerobot-raw/robot_vqa_raw": "openx_rlds.robot_vqa",
-    "lerobot-raw/mimic_play_raw": "openx_rlds.mimic_play",
-    "lerobot-raw/tidybot_raw": "openx_rlds.tidybot",
-    "lerobot-raw/eth_agent_affordances_raw": "openx_rlds.eth_agent_affordances",
-}
-
-
-def download_raw(raw_dir: Path, repo_id: str):
-    check_repo_id(repo_id)
-    user_id, dataset_id = repo_id.split("/")
-
-    if not dataset_id.endswith("_raw"):
-        warnings.warn(
-            f"""`dataset_id` ({dataset_id}) doesn't end with '_raw' (e.g. 'lerobot/pusht_raw'). Following this
-             naming convention by renaming your repository is advised, but not mandatory.""",
-            stacklevel=1,
-        )
-
-    # Send warning if raw_dir isn't well formatted
-    if raw_dir.parts[-2] != user_id or raw_dir.parts[-1] != dataset_id:
-        warnings.warn(
-            f"""`raw_dir` ({raw_dir}) doesn't contain a community or user id `/` the name of the dataset that
-             match the `repo_id` (e.g. 'data/lerobot/pusht_raw'). Following this naming convention is advised,
-             but not mandatory.""",
-            stacklevel=1,
-        )
-    raw_dir.mkdir(parents=True, exist_ok=True)
-
-    logging.info(f"Start downloading from huggingface.co/{user_id} for {dataset_id}")
-    snapshot_download(repo_id, repo_type="dataset", local_dir=raw_dir)
-    logging.info(f"Finish downloading from huggingface.co/{user_id} for {dataset_id}")
-
-
-def download_all_raw_datasets(data_dir: Path | None = None):
-    if data_dir is None:
-        data_dir = Path("data")
-    for repo_id in AVAILABLE_RAW_REPO_IDS:
-        raw_dir = data_dir / repo_id
-        download_raw(raw_dir, repo_id)
-
-
-def main():
-    parser = argparse.ArgumentParser(
-        description=f"""A script to download raw datasets from Hugging Face hub to a local directory. Here is a
-            non exhaustive list of available repositories to use in `--repo-id`: {list(AVAILABLE_RAW_REPO_IDS.keys())}""",
-    )
-
-    parser.add_argument(
-        "--raw-dir",
-        type=Path,
-        required=True,
-        help="Directory containing input raw datasets (e.g. `data/aloha_mobile_chair_raw` or `data/pusht_raw).",
-    )
-    parser.add_argument(
-        "--repo-id",
-        type=str,
-        required=True,
-        help="""Repositery identifier on Hugging Face: a community or a user name `/` the name of
-        the dataset (e.g. `lerobot/pusht_raw`, `cadene/aloha_sim_insertion_human_raw`).""",
-    )
-    args = parser.parse_args()
-    download_raw(**vars(args))
-
-
-if __name__ == "__main__":
-    main()

lerobot/common/datasets/push_dataset_to_hub/_encode_datasets.py

@@ -1,184 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-Use this script to batch encode lerobot dataset from their raw format to LeRobotDataset and push their updated
-version to the hub. Under the hood, this script reuses 'push_dataset_to_hub.py'. It assumes that you already
-downloaded raw datasets, which you can do with the related '_download_raw.py' script.
-
-For instance, for codebase_version = 'v1.6', the following command was run, assuming raw datasets from
-lerobot-raw were downloaded in 'raw/datasets/directory':
-```bash
-python lerobot/common/datasets/push_dataset_to_hub/_encode_datasets.py \
-  --raw-dir raw/datasets/directory \
-  --raw-repo-ids lerobot-raw \
-  --local-dir push/datasets/directory \
-  --tests-data-dir tests/data \
-  --push-repo lerobot \
-  --vcodec libsvtav1 \
-  --pix-fmt yuv420p \
-  --g 2 \
-  --crf 30
-```
-"""
-
-import argparse
-from pathlib import Path
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub._download_raw import AVAILABLE_RAW_REPO_IDS
-from lerobot.common.datasets.push_dataset_to_hub.utils import check_repo_id
-from lerobot.scripts.push_dataset_to_hub import push_dataset_to_hub
-
-
-def get_push_repo_id_from_raw(raw_repo_id: str, push_repo: str) -> str:
-    dataset_id_raw = raw_repo_id.split("/")[1]
-    dataset_id = dataset_id_raw.removesuffix("_raw")
-    return f"{push_repo}/{dataset_id}"
-
-
-def encode_datasets(
-    raw_dir: Path,
-    raw_repo_ids: list[str],
-    push_repo: str,
-    vcodec: str,
-    pix_fmt: str,
-    g: int,
-    crf: int,
-    local_dir: Path | None = None,
-    tests_data_dir: Path | None = None,
-    raw_format: str | None = None,
-    dry_run: bool = False,
-) -> None:
-    if len(raw_repo_ids) == 1 and raw_repo_ids[0].lower() == "lerobot-raw":
-        raw_repo_ids_format = AVAILABLE_RAW_REPO_IDS
-    else:
-        if raw_format is None:
-            raise ValueError(raw_format)
-        raw_repo_ids_format = {id_: raw_format for id_ in raw_repo_ids}
-
-    for raw_repo_id, repo_raw_format in raw_repo_ids_format.items():
-        check_repo_id(raw_repo_id)
-        dataset_repo_id_push = get_push_repo_id_from_raw(raw_repo_id, push_repo)
-        dataset_raw_dir = raw_dir / raw_repo_id
-        dataset_dir = local_dir / dataset_repo_id_push if local_dir is not None else None
-        encoding = {
-            "vcodec": vcodec,
-            "pix_fmt": pix_fmt,
-            "g": g,
-            "crf": crf,
-        }
-
-        if not (dataset_raw_dir).is_dir():
-            raise NotADirectoryError(dataset_raw_dir)
-
-        if not dry_run:
-            push_dataset_to_hub(
-                dataset_raw_dir,
-                raw_format=repo_raw_format,
-                repo_id=dataset_repo_id_push,
-                local_dir=dataset_dir,
-                resume=True,
-                encoding=encoding,
-                tests_data_dir=tests_data_dir,
-            )
-        else:
-            print(
-                f"DRY RUN: {dataset_raw_dir}  -->  {dataset_dir}  -->  {dataset_repo_id_push}@{CODEBASE_VERSION}"
-            )
-
-
-def main():
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--raw-dir",
-        type=Path,
-        default=Path("data"),
-        help="Directory where raw datasets are located.",
-    )
-    parser.add_argument(
-        "--raw-repo-ids",
-        type=str,
-        nargs="*",
-        default=["lerobot-raw"],
-        help="""Raw dataset repo ids. if 'lerobot-raw', the keys from `AVAILABLE_RAW_REPO_IDS` will be
-            used and raw datasets will be fetched from the 'lerobot-raw/' repo and pushed with their
-            associated format. It is assumed that each dataset is located at `raw_dir / raw_repo_id` """,
-    )
-    parser.add_argument(
-        "--raw-format",
-        type=str,
-        default=None,
-        help="""Raw format to use for the raw repo-ids. Must be specified if --raw-repo-ids is not
-            'lerobot-raw'""",
-    )
-    parser.add_argument(
-        "--local-dir",
-        type=Path,
-        default=None,
-        help="""When provided, writes the dataset converted to LeRobotDataset format in this directory
-        (e.g. `data/lerobot/aloha_mobile_chair`).""",
-    )
-    parser.add_argument(
-        "--push-repo",
-        type=str,
-        default="lerobot",
-        help="Repo to upload datasets to",
-    )
-    parser.add_argument(
-        "--vcodec",
-        type=str,
-        default="libsvtav1",
-        help="Codec to use for encoding videos",
-    )
-    parser.add_argument(
-        "--pix-fmt",
-        type=str,
-        default="yuv420p",
-        help="Pixel formats (chroma subsampling) to be used for encoding",
-    )
-    parser.add_argument(
-        "--g",
-        type=int,
-        default=2,
-        help="Group of pictures sizes to be used for encoding.",
-    )
-    parser.add_argument(
-        "--crf",
-        type=int,
-        default=30,
-        help="Constant rate factors to be used for encoding.",
-    )
-    parser.add_argument(
-        "--tests-data-dir",
-        type=Path,
-        default=None,
-        help=(
-            "When provided, save tests artifacts into the given directory "
-            "(e.g. `--tests-data-dir tests/data` will save to tests/data/{--repo-id})."
-        ),
-    )
-    parser.add_argument(
-        "--dry-run",
-        type=int,
-        default=0,
-        help="If not set to 0, this script won't download or upload anything.",
-    )
-    args = parser.parse_args()
-    encode_datasets(**vars(args))
-
-
-if __name__ == "__main__":
-    main()

lerobot/common/datasets/push_dataset_to_hub/_umi_imagecodecs_numcodecs.py

@@ -1,326 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# imagecodecs/numcodecs.py
-
-# Copyright (c) 2021-2022, Christoph Gohlke
-# All rights reserved.
-#
-# Redistribution and use in source and binary forms, with or without
-# modification, are permitted provided that the following conditions are met:
-#
-# 1. Redistributions of source code must retain the above copyright notice,
-#    this list of conditions and the following disclaimer.
-#
-# 2. Redistributions in binary form must reproduce the above copyright notice,
-#    this list of conditions and the following disclaimer in the documentation
-#    and/or other materials provided with the distribution.
-#
-# 3. Neither the name of the copyright holder nor the names of its
-#    contributors may be used to endorse or promote products derived from
-#    this software without specific prior written permission.
-#
-# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
-# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
-# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
-# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
-# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
-# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
-# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
-# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
-# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
-# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
-# POSSIBILITY OF SUCH DAMAGE.
-
-# Copied from: https://github.com/real-stanford/universal_manipulation_interface/blob/298776ce251f33b6b3185a98d6e7d1f9ad49168b/diffusion_policy/codecs/imagecodecs_numcodecs.py#L1
-"""Additional numcodecs implemented using imagecodecs."""
-
-__version__ = "2022.9.26"
-
-__all__ = ("register_codecs",)
-
-import imagecodecs
-import numpy
-from numcodecs.abc import Codec
-from numcodecs.registry import get_codec, register_codec
-
-# TODO (azouitine): Remove useless codecs
-
-
-def protective_squeeze(x: numpy.ndarray):
-    """
-    Squeeze dim only if it's not the last dim.
-    Image dim expected to be *, H, W, C
-    """
-    img_shape = x.shape[-3:]
-    if len(x.shape) > 3:
-        n_imgs = numpy.prod(x.shape[:-3])
-        if n_imgs > 1:
-            img_shape = (-1,) + img_shape
-    return x.reshape(img_shape)
-
-
-def get_default_image_compressor(**kwargs):
-    if imagecodecs.JPEGXL:
-        # has JPEGXL
-        this_kwargs = {
-            "effort": 3,
-            "distance": 0.3,
-            # bug in libjxl, invalid codestream for non-lossless
-            # when decoding speed > 1
-            "decodingspeed": 1,
-        }
-        this_kwargs.update(kwargs)
-        return JpegXl(**this_kwargs)
-    else:
-        this_kwargs = {"level": 50}
-        this_kwargs.update(kwargs)
-        return Jpeg2k(**this_kwargs)
-
-
-class Jpeg2k(Codec):
-    """JPEG 2000 codec for numcodecs."""
-
-    codec_id = "imagecodecs_jpeg2k"
-
-    def __init__(
-        self,
-        level=None,
-        codecformat=None,
-        colorspace=None,
-        tile=None,
-        reversible=None,
-        bitspersample=None,
-        resolutions=None,
-        numthreads=None,
-        verbose=0,
-    ):
-        self.level = level
-        self.codecformat = codecformat
-        self.colorspace = colorspace
-        self.tile = None if tile is None else tuple(tile)
-        self.reversible = reversible
-        self.bitspersample = bitspersample
-        self.resolutions = resolutions
-        self.numthreads = numthreads
-        self.verbose = verbose
-
-    def encode(self, buf):
-        buf = protective_squeeze(numpy.asarray(buf))
-        return imagecodecs.jpeg2k_encode(
-            buf,
-            level=self.level,
-            codecformat=self.codecformat,
-            colorspace=self.colorspace,
-            tile=self.tile,
-            reversible=self.reversible,
-            bitspersample=self.bitspersample,
-            resolutions=self.resolutions,
-            numthreads=self.numthreads,
-            verbose=self.verbose,
-        )
-
-    def decode(self, buf, out=None):
-        return imagecodecs.jpeg2k_decode(buf, verbose=self.verbose, numthreads=self.numthreads, out=out)
-
-
-class JpegXl(Codec):
-    """JPEG XL codec for numcodecs."""
-
-    codec_id = "imagecodecs_jpegxl"
-
-    def __init__(
-        self,
-        # encode
-        level=None,
-        effort=None,
-        distance=None,
-        lossless=None,
-        decodingspeed=None,
-        photometric=None,
-        planar=None,
-        usecontainer=None,
-        # decode
-        index=None,
-        keeporientation=None,
-        # both
-        numthreads=None,
-    ):
-        """
-        Return JPEG XL image from numpy array.
-        Float must be in nominal range 0..1.
-
-        Currently L, LA, RGB, RGBA images are supported in contig mode.
-        Extra channels are only supported for grayscale images in planar mode.
-
-        Parameters
-        ----------
-        level : Default to None, i.e. not overwriting lossess and decodingspeed options.
-            When < 0: Use lossless compression
-            When in [0,1,2,3,4]: Sets the decoding speed tier for the provided options.
-                Minimum is 0 (slowest to decode, best quality/density), and maximum
-                is 4 (fastest to decode, at the cost of some quality/density).
-        effort : Default to 3.
-            Sets encoder effort/speed level without affecting decoding speed.
-            Valid values are, from faster to slower speed: 1:lightning 2:thunder
-                3:falcon 4:cheetah 5:hare 6:wombat 7:squirrel 8:kitten 9:tortoise.
-            Speed: lightning, thunder, falcon, cheetah, hare, wombat, squirrel, kitten, tortoise
-            control the encoder effort in ascending order.
-            This also affects memory usage: using lower effort will typically reduce memory
-            consumption during encoding.
-            lightning and thunder are fast modes useful for lossless mode (modular).
-            falcon disables all of the following tools.
-            cheetah enables coefficient reordering, context clustering, and heuristics for selecting DCT sizes and quantization steps.
-            hare enables Gaborish filtering, chroma from luma, and an initial estimate of quantization steps.
-            wombat enables error diffusion quantization and full DCT size selection heuristics.
-            squirrel (default) enables dots, patches, and spline detection, and full context clustering.
-            kitten optimizes the adaptive quantization for a psychovisual metric.
-            tortoise enables a more thorough adaptive quantization search.
-        distance : Default to 1.0
-            Sets the distance level for lossy compression: target max butteraugli distance,
-            lower = higher quality. Range: 0 .. 15. 0.0 = mathematically lossless
-            (however, use JxlEncoderSetFrameLossless instead to use true lossless,
-            as setting distance to 0 alone is not the only requirement).
-            1.0 = visually lossless. Recommended range: 0.5 .. 3.0.
-        lossess : Default to False.
-            Use lossess encoding.
-        decodingspeed : Default to 0.
-            Duplicate to level. [0,4]
-        photometric : Return JxlColorSpace value.
-            Default logic is quite complicated but works most of the time.
-            Accepted value:
-                int: [-1,3]
-                str: ['RGB',
-                    'WHITEISZERO', 'MINISWHITE',
-                    'BLACKISZERO', 'MINISBLACK', 'GRAY',
-                    'XYB', 'KNOWN']
-        planar : Enable multi-channel mode.
-            Default to false.
-        usecontainer :
-            Forces the encoder to use the box-based container format (BMFF)
-            even when not necessary.
-            When using JxlEncoderUseBoxes, JxlEncoderStoreJPEGMetadata or
-            JxlEncoderSetCodestreamLevel with level 10, the encoder will
-            automatically also use the container format, it is not necessary
-            to use JxlEncoderUseContainer for those use cases.
-            By default this setting is disabled.
-        index : Selectively decode frames for animation.
-            Default to 0, decode all frames.
-            When set to > 0, decode that frame index only.
-        keeporientation :
-            Enables or disables preserving of as-in-bitstream pixeldata orientation.
-            Some images are encoded with an Orientation tag indicating that the
-            decoder must perform a rotation and/or mirroring to the encoded image data.
-
-            If skip_reorientation is JXL_FALSE (the default): the decoder will apply
-            the transformation from the orientation setting, hence rendering the image
-            according to its specified intent. When producing a JxlBasicInfo, the decoder
-            will always set the orientation field to JXL_ORIENT_IDENTITY (matching the
-            returned pixel data) and also align xsize and ysize so that they correspond
-            to the width and the height of the returned pixel data.
-
-            If skip_reorientation is JXL_TRUE: the decoder will skip applying the
-            transformation from the orientation setting, returning the image in
-            the as-in-bitstream pixeldata orientation. This may be faster to decode
-            since the decoder doesnt have to apply the transformation, but can
-            cause wrong display of the image if the orientation tag is not correctly
-            taken into account by the user.
-
-            By default, this option is disabled, and the returned pixel data is
-            re-oriented according to the images Orientation setting.
-        threads : Default to 1.
-            If <= 0, use all cores.
-            If > 32, clipped to 32.
-        """
-
-        self.level = level
-        self.effort = effort
-        self.distance = distance
-        self.lossless = bool(lossless)
-        self.decodingspeed = decodingspeed
-        self.photometric = photometric
-        self.planar = planar
-        self.usecontainer = usecontainer
-        self.index = index
-        self.keeporientation = keeporientation
-        self.numthreads = numthreads
-
-    def encode(self, buf):
-        # TODO: only squeeze all but last dim
-        buf = protective_squeeze(numpy.asarray(buf))
-        return imagecodecs.jpegxl_encode(
-            buf,
-            level=self.level,
-            effort=self.effort,
-            distance=self.distance,
-            lossless=self.lossless,
-            decodingspeed=self.decodingspeed,
-            photometric=self.photometric,
-            planar=self.planar,
-            usecontainer=self.usecontainer,
-            numthreads=self.numthreads,
-        )
-
-    def decode(self, buf, out=None):
-        return imagecodecs.jpegxl_decode(
-            buf,
-            index=self.index,
-            keeporientation=self.keeporientation,
-            numthreads=self.numthreads,
-            out=out,
-        )
-
-
-def _flat(out):
-    """Return numpy array as contiguous view of bytes if possible."""
-    if out is None:
-        return None
-    view = memoryview(out)
-    if view.readonly or not view.contiguous:
-        return None
-    return view.cast("B")
-
-
-def register_codecs(codecs=None, force=False, verbose=True):
-    """Register codecs in this module with numcodecs."""
-    for name, cls in globals().items():
-        if not hasattr(cls, "codec_id") or name == "Codec":
-            continue
-        if codecs is not None and cls.codec_id not in codecs:
-            continue
-        try:
-            try:  # noqa: SIM105
-                get_codec({"id": cls.codec_id})
-            except TypeError:
-                # registered, but failed
-                pass
-        except ValueError:
-            # not registered yet
-            pass
-        else:
-            if not force:
-                if verbose:
-                    log_warning(f"numcodec {cls.codec_id!r} already registered")
-                continue
-            if verbose:
-                log_warning(f"replacing registered numcodec {cls.codec_id!r}")
-        register_codec(cls)
-
-
-def log_warning(msg, *args, **kwargs):
-    """Log message with level WARNING."""
-    import logging
-
-    logging.getLogger(__name__).warning(msg, *args, **kwargs)

lerobot/common/datasets/push_dataset_to_hub/aloha_hdf5_format.py

@@ -1,233 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-Contains utilities to process raw data format of HDF5 files like in: https://github.com/tonyzhaozh/act
-"""
-
-import gc
-import shutil
-from pathlib import Path
-
-import h5py
-import numpy as np
-import torch
-import tqdm
-from datasets import Dataset, Features, Image, Sequence, Value
-from PIL import Image as PILImage
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub.utils import (
-    calculate_episode_data_index,
-    concatenate_episodes,
-    get_default_encoding,
-    save_images_concurrently,
-)
-from lerobot.common.datasets.utils import (
-    hf_transform_to_torch,
-)
-from lerobot.common.datasets.video_utils import VideoFrame, encode_video_frames
-
-
-def get_cameras(hdf5_data):
-    # ignore depth channel, not currently handled
-    # TODO(rcadene): add depth
-    rgb_cameras = [key for key in hdf5_data["/observations/images"].keys() if "depth" not in key]  # noqa: SIM118
-    return rgb_cameras
-
-
-def check_format(raw_dir) -> bool:
-    # only frames from simulation are uncompressed
-    compressed_images = "sim" not in raw_dir.name
-
-    hdf5_paths = list(raw_dir.glob("episode_*.hdf5"))
-    assert len(hdf5_paths) != 0
-    for hdf5_path in hdf5_paths:
-        with h5py.File(hdf5_path, "r") as data:
-            assert "/action" in data
-            assert "/observations/qpos" in data
-
-            assert data["/action"].ndim == 2
-            assert data["/observations/qpos"].ndim == 2
-
-            num_frames = data["/action"].shape[0]
-            assert num_frames == data["/observations/qpos"].shape[0]
-
-            for camera in get_cameras(data):
-                assert num_frames == data[f"/observations/images/{camera}"].shape[0]
-
-                if compressed_images:
-                    assert data[f"/observations/images/{camera}"].ndim == 2
-                else:
-                    assert data[f"/observations/images/{camera}"].ndim == 4
-                    b, h, w, c = data[f"/observations/images/{camera}"].shape
-                    assert c < h and c < w, f"Expect (h,w,c) image format but ({h=},{w=},{c=}) provided."
-
-
-def load_from_raw(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int,
-    video: bool,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    # only frames from simulation are uncompressed
-    compressed_images = "sim" not in raw_dir.name
-
-    hdf5_files = sorted(raw_dir.glob("episode_*.hdf5"))
-    num_episodes = len(hdf5_files)
-
-    ep_dicts = []
-    ep_ids = episodes if episodes else range(num_episodes)
-    for ep_idx in tqdm.tqdm(ep_ids):
-        ep_path = hdf5_files[ep_idx]
-        with h5py.File(ep_path, "r") as ep:
-            num_frames = ep["/action"].shape[0]
-
-            # last step of demonstration is considered done
-            done = torch.zeros(num_frames, dtype=torch.bool)
-            done[-1] = True
-
-            state = torch.from_numpy(ep["/observations/qpos"][:])
-            action = torch.from_numpy(ep["/action"][:])
-            if "/observations/qvel" in ep:
-                velocity = torch.from_numpy(ep["/observations/qvel"][:])
-            if "/observations/effort" in ep:
-                effort = torch.from_numpy(ep["/observations/effort"][:])
-
-            ep_dict = {}
-
-            for camera in get_cameras(ep):
-                img_key = f"observation.images.{camera}"
-
-                if compressed_images:
-                    import cv2
-
-                    # load one compressed image after the other in RAM and uncompress
-                    imgs_array = []
-                    for data in ep[f"/observations/images/{camera}"]:
-                        imgs_array.append(cv2.imdecode(data, 1))
-                    imgs_array = np.array(imgs_array)
-
-                else:
-                    # load all images in RAM
-                    imgs_array = ep[f"/observations/images/{camera}"][:]
-
-                if video:
-                    # save png images in temporary directory
-                    tmp_imgs_dir = videos_dir / "tmp_images"
-                    save_images_concurrently(imgs_array, tmp_imgs_dir)
-
-                    # encode images to a mp4 video
-                    fname = f"{img_key}_episode_{ep_idx:06d}.mp4"
-                    video_path = videos_dir / fname
-                    encode_video_frames(tmp_imgs_dir, video_path, fps, **(encoding or {}))
-
-                    # clean temporary images directory
-                    shutil.rmtree(tmp_imgs_dir)
-
-                    # store the reference to the video frame
-                    ep_dict[img_key] = [
-                        {"path": f"videos/{fname}", "timestamp": i / fps} for i in range(num_frames)
-                    ]
-                else:
-                    ep_dict[img_key] = [PILImage.fromarray(x) for x in imgs_array]
-
-            ep_dict["observation.state"] = state
-            if "/observations/velocity" in ep:
-                ep_dict["observation.velocity"] = velocity
-            if "/observations/effort" in ep:
-                ep_dict["observation.effort"] = effort
-            ep_dict["action"] = action
-            ep_dict["episode_index"] = torch.tensor([ep_idx] * num_frames)
-            ep_dict["frame_index"] = torch.arange(0, num_frames, 1)
-            ep_dict["timestamp"] = torch.arange(0, num_frames, 1) / fps
-            ep_dict["next.done"] = done
-            # TODO(rcadene): add reward and success by computing them in sim
-
-            assert isinstance(ep_idx, int)
-            ep_dicts.append(ep_dict)
-
-        gc.collect()
-
-    data_dict = concatenate_episodes(ep_dicts)
-
-    total_frames = data_dict["frame_index"].shape[0]
-    data_dict["index"] = torch.arange(0, total_frames, 1)
-    return data_dict
-
-
-def to_hf_dataset(data_dict, video) -> Dataset:
-    features = {}
-
-    keys = [key for key in data_dict if "observation.images." in key]
-    for key in keys:
-        if video:
-            features[key] = VideoFrame()
-        else:
-            features[key] = Image()
-
-    features["observation.state"] = Sequence(
-        length=data_dict["observation.state"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    if "observation.velocity" in data_dict:
-        features["observation.velocity"] = Sequence(
-            length=data_dict["observation.velocity"].shape[1], feature=Value(dtype="float32", id=None)
-        )
-    if "observation.effort" in data_dict:
-        features["observation.effort"] = Sequence(
-            length=data_dict["observation.effort"].shape[1], feature=Value(dtype="float32", id=None)
-        )
-    features["action"] = Sequence(
-        length=data_dict["action"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["episode_index"] = Value(dtype="int64", id=None)
-    features["frame_index"] = Value(dtype="int64", id=None)
-    features["timestamp"] = Value(dtype="float32", id=None)
-    features["next.done"] = Value(dtype="bool", id=None)
-    features["index"] = Value(dtype="int64", id=None)
-
-    hf_dataset = Dataset.from_dict(data_dict, features=Features(features))
-    hf_dataset.set_transform(hf_transform_to_torch)
-    return hf_dataset
-
-
-def from_raw_to_lerobot_format(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int | None = None,
-    video: bool = True,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    # sanity check
-    check_format(raw_dir)
-
-    if fps is None:
-        fps = 50
-
-    data_dict = load_from_raw(raw_dir, videos_dir, fps, video, episodes, encoding)
-    hf_dataset = to_hf_dataset(data_dict, video)
-    episode_data_index = calculate_episode_data_index(hf_dataset)
-    info = {
-        "codebase_version": CODEBASE_VERSION,
-        "fps": fps,
-        "video": video,
-    }
-    if video:
-        info["encoding"] = get_default_encoding()
-
-    return hf_dataset, episode_data_index, info

lerobot/common/datasets/push_dataset_to_hub/cam_png_format.py

@@ -1,107 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-Contains utilities to process raw data format of png images files recorded with capture_camera_feed.py
-"""
-
-from pathlib import Path
-
-import torch
-from datasets import Dataset, Features, Image, Value
-from PIL import Image as PILImage
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub.utils import (
-    calculate_episode_data_index,
-    concatenate_episodes,
-)
-from lerobot.common.datasets.utils import hf_transform_to_torch
-from lerobot.common.datasets.video_utils import VideoFrame
-
-
-def check_format(raw_dir: Path) -> bool:
-    image_paths = list(raw_dir.glob("frame_*.png"))
-    if len(image_paths) == 0:
-        raise ValueError
-
-
-def load_from_raw(raw_dir: Path, fps: int, episodes: list[int] | None = None):
-    if episodes is not None:
-        # TODO(aliberts): add support for multi-episodes.
-        raise NotImplementedError()
-
-    ep_dict = {}
-    ep_idx = 0
-
-    image_paths = sorted(raw_dir.glob("frame_*.png"))
-    num_frames = len(image_paths)
-
-    ep_dict["observation.image"] = [PILImage.open(x) for x in image_paths]
-    ep_dict["episode_index"] = torch.tensor([ep_idx] * num_frames)
-    ep_dict["frame_index"] = torch.arange(0, num_frames, 1)
-    ep_dict["timestamp"] = torch.arange(0, num_frames, 1) / fps
-
-    ep_dicts = [ep_dict]
-    data_dict = concatenate_episodes(ep_dicts)
-    total_frames = data_dict["frame_index"].shape[0]
-    data_dict["index"] = torch.arange(0, total_frames, 1)
-    return data_dict
-
-
-def to_hf_dataset(data_dict, video) -> Dataset:
-    features = {}
-    if video:
-        features["observation.image"] = VideoFrame()
-    else:
-        features["observation.image"] = Image()
-
-    features["episode_index"] = Value(dtype="int64", id=None)
-    features["frame_index"] = Value(dtype="int64", id=None)
-    features["timestamp"] = Value(dtype="float32", id=None)
-    features["index"] = Value(dtype="int64", id=None)
-
-    hf_dataset = Dataset.from_dict(data_dict, features=Features(features))
-    hf_dataset.set_transform(hf_transform_to_torch)
-    return hf_dataset
-
-
-def from_raw_to_lerobot_format(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int | None = None,
-    video: bool = True,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    if video or episodes or encoding is not None:
-        # TODO(aliberts): support this
-        raise NotImplementedError
-
-    # sanity check
-    check_format(raw_dir)
-
-    if fps is None:
-        fps = 30
-
-    data_dict = load_from_raw(raw_dir, videos_dir, fps, video, episodes)
-    hf_dataset = to_hf_dataset(data_dict, video)
-    episode_data_index = calculate_episode_data_index(hf_dataset)
-    info = {
-        "codebase_version": CODEBASE_VERSION,
-        "fps": fps,
-        "video": video,
-    }
-    return hf_dataset, episode_data_index, info

lerobot/common/datasets/push_dataset_to_hub/dora_parquet_format.py

@@ -1,233 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-Contains utilities to process raw data format from dora-record
-"""
-
-import re
-import warnings
-from pathlib import Path
-
-import pandas as pd
-import torch
-from datasets import Dataset, Features, Image, Sequence, Value
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub.utils import calculate_episode_data_index
-from lerobot.common.datasets.utils import (
-    hf_transform_to_torch,
-)
-from lerobot.common.datasets.video_utils import VideoFrame
-
-
-def check_format(raw_dir) -> bool:
-    assert raw_dir.exists()
-
-    leader_file = list(raw_dir.glob("*.parquet"))
-    if len(leader_file) == 0:
-        raise ValueError(f"Missing parquet files in '{raw_dir}'")
-    return True
-
-
-def load_from_raw(raw_dir: Path, videos_dir: Path, fps: int, video: bool, episodes: list[int] | None = None):
-    # Load data stream that will be used as reference for the timestamps synchronization
-    reference_files = list(raw_dir.glob("observation.images.cam_*.parquet"))
-    if len(reference_files) == 0:
-        raise ValueError(f"Missing reference files for camera, starting with  in '{raw_dir}'")
-    # select first camera in alphanumeric order
-    reference_key = sorted(reference_files)[0].stem
-    reference_df = pd.read_parquet(raw_dir / f"{reference_key}.parquet")
-    reference_df = reference_df[["timestamp_utc", reference_key]]
-
-    # Merge all data stream using nearest backward strategy
-    df = reference_df
-    for path in raw_dir.glob("*.parquet"):
-        key = path.stem  # action or observation.state or ...
-        if key == reference_key:
-            continue
-        if "failed_episode_index" in key:
-            # TODO(rcadene): add support for removing episodes that are tagged as "failed"
-            continue
-        modality_df = pd.read_parquet(path)
-        modality_df = modality_df[["timestamp_utc", key]]
-        df = pd.merge_asof(
-            df,
-            modality_df,
-            on="timestamp_utc",
-            # "nearest" is the best option over "backward", since the latter can desynchronizes camera timestamps by
-            # matching timestamps that are too far apart, in order to fit the backward constraints. It's not the case for "nearest".
-            # However, note that "nearest" might synchronize the reference camera with other cameras on slightly future timestamps.
-            # are too far apart.
-            direction="nearest",
-            tolerance=pd.Timedelta(f"{1 / fps} seconds"),
-        )
-    # Remove rows with episode_index -1 which indicates data that correspond to in-between episodes
-    df = df[df["episode_index"] != -1]
-
-    image_keys = [key for key in df if "observation.images." in key]
-
-    def get_episode_index(row):
-        episode_index_per_cam = {}
-        for key in image_keys:
-            path = row[key][0]["path"]
-            match = re.search(r"_(\d{6}).mp4", path)
-            if not match:
-                raise ValueError(path)
-            episode_index = int(match.group(1))
-            episode_index_per_cam[key] = episode_index
-        if len(set(episode_index_per_cam.values())) != 1:
-            raise ValueError(
-                f"All cameras are expected to belong to the same episode, but getting {episode_index_per_cam}"
-            )
-        return episode_index
-
-    df["episode_index"] = df.apply(get_episode_index, axis=1)
-
-    # dora only use arrays, so single values are encapsulated into a list
-    df["frame_index"] = df.groupby("episode_index").cumcount()
-    df = df.reset_index()
-    df["index"] = df.index
-
-    # set 'next.done' to True for the last frame of each episode
-    df["next.done"] = False
-    df.loc[df.groupby("episode_index").tail(1).index, "next.done"] = True
-
-    df["timestamp"] = df["timestamp_utc"].map(lambda x: x.timestamp())
-    # each episode starts with timestamp 0 to match the ones from the video
-    df["timestamp"] = df.groupby("episode_index")["timestamp"].transform(lambda x: x - x.iloc[0])
-
-    del df["timestamp_utc"]
-
-    # sanity check
-    has_nan = df.isna().any().any()
-    if has_nan:
-        raise ValueError("Dataset contains Nan values.")
-
-    # sanity check episode indices go from 0 to n-1
-    ep_ids = [ep_idx for ep_idx, _ in df.groupby("episode_index")]
-    expected_ep_ids = list(range(df["episode_index"].max() + 1))
-    if ep_ids != expected_ep_ids:
-        raise ValueError(f"Episodes indices go from {ep_ids} instead of {expected_ep_ids}")
-
-    # Create symlink to raw videos directory (that needs to be absolute not relative)
-    videos_dir.parent.mkdir(parents=True, exist_ok=True)
-    videos_dir.symlink_to((raw_dir / "videos").absolute())
-
-    # sanity check the video paths are well formatted
-    for key in df:
-        if "observation.images." not in key:
-            continue
-        for ep_idx in ep_ids:
-            video_path = videos_dir / f"{key}_episode_{ep_idx:06d}.mp4"
-            if not video_path.exists():
-                raise ValueError(f"Video file not found in {video_path}")
-
-    data_dict = {}
-    for key in df:
-        # is video frame
-        if "observation.images." in key:
-            # we need `[0] because dora only use arrays, so single values are encapsulated into a list.
-            # it is the case for video_frame dictionary = [{"path": ..., "timestamp": ...}]
-            data_dict[key] = [video_frame[0] for video_frame in df[key].values]
-
-            # sanity check the video path is well formatted
-            video_path = videos_dir.parent / data_dict[key][0]["path"]
-            if not video_path.exists():
-                raise ValueError(f"Video file not found in {video_path}")
-        # is number
-        elif df[key].iloc[0].ndim == 0 or df[key].iloc[0].shape[0] == 1:
-            data_dict[key] = torch.from_numpy(df[key].values)
-        # is vector
-        elif df[key].iloc[0].shape[0] > 1:
-            data_dict[key] = torch.stack([torch.from_numpy(x.copy()) for x in df[key].values])
-        else:
-            raise ValueError(key)
-
-    return data_dict
-
-
-def to_hf_dataset(data_dict, video) -> Dataset:
-    features = {}
-
-    keys = [key for key in data_dict if "observation.images." in key]
-    for key in keys:
-        if video:
-            features[key] = VideoFrame()
-        else:
-            features[key] = Image()
-
-    features["observation.state"] = Sequence(
-        length=data_dict["observation.state"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    if "observation.velocity" in data_dict:
-        features["observation.velocity"] = Sequence(
-            length=data_dict["observation.velocity"].shape[1], feature=Value(dtype="float32", id=None)
-        )
-    if "observation.effort" in data_dict:
-        features["observation.effort"] = Sequence(
-            length=data_dict["observation.effort"].shape[1], feature=Value(dtype="float32", id=None)
-        )
-    features["action"] = Sequence(
-        length=data_dict["action"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["episode_index"] = Value(dtype="int64", id=None)
-    features["frame_index"] = Value(dtype="int64", id=None)
-    features["timestamp"] = Value(dtype="float32", id=None)
-    features["next.done"] = Value(dtype="bool", id=None)
-    features["index"] = Value(dtype="int64", id=None)
-
-    hf_dataset = Dataset.from_dict(data_dict, features=Features(features))
-    hf_dataset.set_transform(hf_transform_to_torch)
-    return hf_dataset
-
-
-def from_raw_to_lerobot_format(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int | None = None,
-    video: bool = True,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    # sanity check
-    check_format(raw_dir)
-
-    if fps is None:
-        fps = 30
-    else:
-        raise NotImplementedError()
-
-    if not video:
-        raise NotImplementedError()
-
-    if encoding is not None:
-        warnings.warn(
-            "Video encoding is currently done outside of LeRobot for the dora_parquet format.",
-            stacklevel=1,
-        )
-
-    data_df = load_from_raw(raw_dir, videos_dir, fps, episodes)
-    hf_dataset = to_hf_dataset(data_df, video)
-    episode_data_index = calculate_episode_data_index(hf_dataset)
-    info = {
-        "codebase_version": CODEBASE_VERSION,
-        "fps": fps,
-        "video": video,
-    }
-    if video:
-        info["encoding"] = "unknown"
-
-    return hf_dataset, episode_data_index, info

lerobot/common/datasets/push_dataset_to_hub/openx_rlds_format.py

@@ -1,312 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-For all datasets in the RLDS format.
-For https://github.com/google-deepmind/open_x_embodiment (OPENX) datasets.
-
-NOTE: You need to install tensorflow and tensorflow_datasets before running this script.
-
-Example:
-    python lerobot/scripts/push_dataset_to_hub.py \
-        --raw-dir /path/to/data/bridge_dataset/1.0.0/ \
-        --repo-id your_hub/sampled_bridge_data_v2 \
-        --raw-format rlds \
-        --episodes 3 4 5 8 9
-
-Exact dataset fps defined in openx/config.py, obtained from:
-    https://docs.google.com/spreadsheets/d/1rPBD77tk60AEIGZrGSODwyyzs5FgCU9Uz3h-3_t2A9g/edit?gid=0#gid=0&range=R:R
-"""
-
-import shutil
-from pathlib import Path
-
-import numpy as np
-import tensorflow as tf
-import tensorflow_datasets as tfds
-import torch
-import tqdm
-from datasets import Dataset, Features, Image, Sequence, Value
-from PIL import Image as PILImage
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub.utils import (
-    calculate_episode_data_index,
-    concatenate_episodes,
-    get_default_encoding,
-    save_images_concurrently,
-)
-from lerobot.common.datasets.utils import (
-    hf_transform_to_torch,
-)
-from lerobot.common.datasets.video_utils import VideoFrame, encode_video_frames
-
-np.set_printoptions(precision=2)
-
-
-def tf_to_torch(data):
-    return torch.from_numpy(data.numpy())
-
-
-def tf_img_convert(img):
-    if img.dtype == tf.string:
-        img = tf.io.decode_image(img, expand_animations=False, dtype=tf.uint8)
-    elif img.dtype != tf.uint8:
-        raise ValueError(f"Unsupported image dtype: found with dtype {img.dtype}")
-    return img.numpy()
-
-
-def _broadcast_metadata_rlds(i: tf.Tensor, traj: dict) -> dict:
-    """
-    In the RLDS format, each trajectory has some top-level metadata that is explicitly separated out, and a "steps"
-    entry. This function moves the "steps" entry to the top level, broadcasting any metadata to the length of the
-    trajectory. This function also adds the extra metadata fields `_len`, `_traj_index`, and `_frame_index`.
-
-    NOTE: adapted from DLimp library https://github.com/kvablack/dlimp/
-    """
-    steps = traj.pop("steps")
-
-    traj_len = tf.shape(tf.nest.flatten(steps)[0])[0]
-
-    # broadcast metadata to the length of the trajectory
-    metadata = tf.nest.map_structure(lambda x: tf.repeat(x, traj_len), traj)
-
-    # put steps back in
-    assert "traj_metadata" not in steps
-    traj = {**steps, "traj_metadata": metadata}
-
-    assert "_len" not in traj
-    assert "_traj_index" not in traj
-    assert "_frame_index" not in traj
-    traj["_len"] = tf.repeat(traj_len, traj_len)
-    traj["_traj_index"] = tf.repeat(i, traj_len)
-    traj["_frame_index"] = tf.range(traj_len)
-
-    return traj
-
-
-def load_from_raw(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int,
-    video: bool,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    """
-    Args:
-        raw_dir (Path): _description_
-        videos_dir (Path): _description_
-        fps (int): _description_
-        video (bool): _description_
-        episodes (list[int] | None, optional): _description_. Defaults to None.
-    """
-    ds_builder = tfds.builder_from_directory(str(raw_dir))
-    dataset = ds_builder.as_dataset(
-        split="all",
-        decoders={"steps": tfds.decode.SkipDecoding()},
-    )
-
-    dataset_info = ds_builder.info
-    print("dataset_info: ", dataset_info)
-
-    ds_length = len(dataset)
-    dataset = dataset.take(ds_length)
-    # "flatten" the dataset as such we can apply trajectory level map() easily
-    # each [obs][key] has a shape of (frame_size, ...)
-    dataset = dataset.enumerate().map(_broadcast_metadata_rlds)
-
-    # we will apply the standardization transform if the dataset_name is provided
-    # if the dataset name is not provided and the goal is to convert any rlds formatted dataset
-    # search for 'image' keys in the observations
-    image_keys = []
-    state_keys = []
-    observation_info = dataset_info.features["steps"]["observation"]
-    for key in observation_info:
-        # check whether the key is for an image or a vector observation
-        if len(observation_info[key].shape) == 3:
-            # only adding uint8 images discards depth images
-            if observation_info[key].dtype == tf.uint8:
-                image_keys.append(key)
-        else:
-            state_keys.append(key)
-
-    lang_key = "language_instruction" if "language_instruction" in dataset.element_spec else None
-
-    print(" - image_keys: ", image_keys)
-    print(" - lang_key: ", lang_key)
-
-    it = iter(dataset)
-
-    ep_dicts = []
-    # Init temp path to save ep_dicts in case of crash
-    tmp_ep_dicts_dir = videos_dir.parent.joinpath("ep_dicts")
-    tmp_ep_dicts_dir.mkdir(parents=True, exist_ok=True)
-
-    # check if ep_dicts have already been saved in /tmp
-    starting_ep_idx = 0
-    saved_ep_dicts = [ep.__str__() for ep in tmp_ep_dicts_dir.iterdir()]
-    if len(saved_ep_dicts) > 0:
-        saved_ep_dicts.sort()
-        # get last ep_idx number
-        starting_ep_idx = int(saved_ep_dicts[-1][-13:-3]) + 1
-        for i in range(starting_ep_idx):
-            episode = next(it)
-            ep_dicts.append(torch.load(saved_ep_dicts[i]))
-
-    # if we user specified episodes, skip the ones not in the list
-    if episodes is not None:
-        if ds_length == 0:
-            raise ValueError("No episodes found.")
-        # convert episodes index to sorted list
-        episodes = sorted(episodes)
-
-    for ep_idx in tqdm.tqdm(range(starting_ep_idx, ds_length)):
-        episode = next(it)
-
-        # if user specified episodes, skip the ones not in the list
-        if episodes is not None:
-            if len(episodes) == 0:
-                break
-            if ep_idx == episodes[0]:
-                # process this episode
-                print(" selecting episode idx: ", ep_idx)
-                episodes.pop(0)
-            else:
-                continue  # skip
-
-        num_frames = episode["action"].shape[0]
-
-        ep_dict = {}
-        for key in state_keys:
-            ep_dict[f"observation.{key}"] = tf_to_torch(episode["observation"][key])
-
-        ep_dict["action"] = tf_to_torch(episode["action"])
-        ep_dict["next.reward"] = tf_to_torch(episode["reward"]).float()
-        ep_dict["next.done"] = tf_to_torch(episode["is_last"])
-        ep_dict["is_terminal"] = tf_to_torch(episode["is_terminal"])
-        ep_dict["is_first"] = tf_to_torch(episode["is_first"])
-        ep_dict["discount"] = tf_to_torch(episode["discount"])
-
-        # If lang_key is present, convert the entire tensor at once
-        if lang_key is not None:
-            ep_dict["language_instruction"] = [x.numpy().decode("utf-8") for x in episode[lang_key]]
-
-        ep_dict["timestamp"] = torch.arange(0, num_frames, 1) / fps
-        ep_dict["episode_index"] = torch.tensor([ep_idx] * num_frames)
-        ep_dict["frame_index"] = torch.arange(0, num_frames, 1)
-
-        image_array_dict = {key: [] for key in image_keys}
-
-        for im_key in image_keys:
-            imgs = episode["observation"][im_key]
-            image_array_dict[im_key] = [tf_img_convert(img) for img in imgs]
-
-        # loop through all cameras
-        for im_key in image_keys:
-            img_key = f"observation.images.{im_key}"
-            imgs_array = image_array_dict[im_key]
-            imgs_array = np.array(imgs_array)
-            if video:
-                # save png images in temporary directory
-                tmp_imgs_dir = videos_dir / "tmp_images"
-                save_images_concurrently(imgs_array, tmp_imgs_dir)
-
-                # encode images to a mp4 video
-                fname = f"{img_key}_episode_{ep_idx:06d}.mp4"
-                video_path = videos_dir / fname
-                encode_video_frames(tmp_imgs_dir, video_path, fps, **(encoding or {}))
-
-                # clean temporary images directory
-                shutil.rmtree(tmp_imgs_dir)
-
-                # store the reference to the video frame
-                ep_dict[img_key] = [
-                    {"path": f"videos/{fname}", "timestamp": i / fps} for i in range(num_frames)
-                ]
-            else:
-                ep_dict[img_key] = [PILImage.fromarray(x) for x in imgs_array]
-
-        path_ep_dict = tmp_ep_dicts_dir.joinpath(
-            "ep_dict_" + "0" * (10 - len(str(ep_idx))) + str(ep_idx) + ".pt"
-        )
-        torch.save(ep_dict, path_ep_dict)
-
-        ep_dicts.append(ep_dict)
-
-    data_dict = concatenate_episodes(ep_dicts)
-
-    total_frames = data_dict["frame_index"].shape[0]
-    data_dict["index"] = torch.arange(0, total_frames, 1)
-    return data_dict
-
-
-def to_hf_dataset(data_dict, video) -> Dataset:
-    features = {}
-
-    for key in data_dict:
-        # check if vector state obs
-        if key.startswith("observation.") and "observation.images." not in key:
-            features[key] = Sequence(length=data_dict[key].shape[1], feature=Value(dtype="float32", id=None))
-        # check if image obs
-        elif "observation.images." in key:
-            if video:
-                features[key] = VideoFrame()
-            else:
-                features[key] = Image()
-
-    if "language_instruction" in data_dict:
-        features["language_instruction"] = Value(dtype="string", id=None)
-
-    features["action"] = Sequence(
-        length=data_dict["action"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-
-    features["is_terminal"] = Value(dtype="bool", id=None)
-    features["is_first"] = Value(dtype="bool", id=None)
-    features["discount"] = Value(dtype="float32", id=None)
-
-    features["episode_index"] = Value(dtype="int64", id=None)
-    features["frame_index"] = Value(dtype="int64", id=None)
-    features["timestamp"] = Value(dtype="float32", id=None)
-    features["next.reward"] = Value(dtype="float32", id=None)
-    features["next.done"] = Value(dtype="bool", id=None)
-    features["index"] = Value(dtype="int64", id=None)
-
-    hf_dataset = Dataset.from_dict(data_dict, features=Features(features))
-    hf_dataset.set_transform(hf_transform_to_torch)
-    return hf_dataset
-
-
-def from_raw_to_lerobot_format(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int | None = None,
-    video: bool = True,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    data_dict = load_from_raw(raw_dir, videos_dir, fps, video, episodes, encoding)
-    hf_dataset = to_hf_dataset(data_dict, video)
-    episode_data_index = calculate_episode_data_index(hf_dataset)
-    info = {
-        "codebase_version": CODEBASE_VERSION,
-        "fps": fps,
-        "video": video,
-    }
-    if video:
-        info["encoding"] = get_default_encoding()
-
-    return hf_dataset, episode_data_index, info

lerobot/common/datasets/push_dataset_to_hub/pusht_zarr_format.py

@@ -1,275 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Process zarr files formatted like in: https://github.com/real-stanford/diffusion_policy"""
-
-import shutil
-from pathlib import Path
-
-import numpy as np
-import torch
-import tqdm
-import zarr
-from datasets import Dataset, Features, Image, Sequence, Value
-from PIL import Image as PILImage
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub.utils import (
-    calculate_episode_data_index,
-    concatenate_episodes,
-    get_default_encoding,
-    save_images_concurrently,
-)
-from lerobot.common.datasets.utils import (
-    hf_transform_to_torch,
-)
-from lerobot.common.datasets.video_utils import VideoFrame, encode_video_frames
-
-
-def check_format(raw_dir):
-    zarr_path = raw_dir / "pusht_cchi_v7_replay.zarr"
-    zarr_data = zarr.open(zarr_path, mode="r")
-
-    required_datasets = {
-        "data/action",
-        "data/img",
-        "data/keypoint",
-        "data/n_contacts",
-        "data/state",
-        "meta/episode_ends",
-    }
-    for dataset in required_datasets:
-        assert dataset in zarr_data
-    nb_frames = zarr_data["data/img"].shape[0]
-
-    required_datasets.remove("meta/episode_ends")
-
-    assert all(nb_frames == zarr_data[dataset].shape[0] for dataset in required_datasets)
-
-
-def load_from_raw(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int,
-    video: bool,
-    episodes: list[int] | None = None,
-    keypoints_instead_of_image: bool = False,
-    encoding: dict | None = None,
-):
-    try:
-        import pymunk
-        from gym_pusht.envs.pusht import PushTEnv, pymunk_to_shapely
-
-        from lerobot.common.datasets.push_dataset_to_hub._diffusion_policy_replay_buffer import (
-            ReplayBuffer as DiffusionPolicyReplayBuffer,
-        )
-    except ModuleNotFoundError as e:
-        print("`gym_pusht` is not installed. Please install it with `pip install 'lerobot[gym_pusht]'`")
-        raise e
-    # as define in gmy-pusht env: https://github.com/huggingface/gym-pusht/blob/e0684ff988d223808c0a9dcfaba9dc4991791370/gym_pusht/envs/pusht.py#L174
-    success_threshold = 0.95  # 95% coverage,
-
-    zarr_path = raw_dir / "pusht_cchi_v7_replay.zarr"
-    zarr_data = DiffusionPolicyReplayBuffer.copy_from_path(zarr_path)
-
-    episode_ids = torch.from_numpy(zarr_data.get_episode_idxs())
-    assert len(
-        {zarr_data[key].shape[0] for key in zarr_data.keys()}  # noqa: SIM118
-    ), "Some data type dont have the same number of total frames."
-
-    # TODO(rcadene): verify that goal pose is expected to be fixed
-    goal_pos_angle = np.array([256, 256, np.pi / 4])  # x, y, theta (in radians)
-    goal_body = PushTEnv.get_goal_pose_body(goal_pos_angle)
-
-    imgs = torch.from_numpy(zarr_data["img"])  # b h w c
-    states = torch.from_numpy(zarr_data["state"])
-    actions = torch.from_numpy(zarr_data["action"])
-
-    # load data indices from which each episode starts and ends
-    from_ids, to_ids = [], []
-    from_idx = 0
-    for to_idx in zarr_data.meta["episode_ends"]:
-        from_ids.append(from_idx)
-        to_ids.append(to_idx)
-        from_idx = to_idx
-
-    num_episodes = len(from_ids)
-
-    ep_dicts = []
-    ep_ids = episodes if episodes else range(num_episodes)
-    for ep_idx, selected_ep_idx in tqdm.tqdm(enumerate(ep_ids)):
-        from_idx = from_ids[selected_ep_idx]
-        to_idx = to_ids[selected_ep_idx]
-        num_frames = to_idx - from_idx
-
-        # sanity check
-        assert (episode_ids[from_idx:to_idx] == ep_idx).all()
-
-        # get image
-        if not keypoints_instead_of_image:
-            image = imgs[from_idx:to_idx]
-            assert image.min() >= 0.0
-            assert image.max() <= 255.0
-            image = image.type(torch.uint8)
-
-        # get state
-        state = states[from_idx:to_idx]
-        agent_pos = state[:, :2]
-        block_pos = state[:, 2:4]
-        block_angle = state[:, 4]
-
-        # get reward, success, done, and (maybe) keypoints
-        reward = torch.zeros(num_frames)
-        success = torch.zeros(num_frames, dtype=torch.bool)
-        if keypoints_instead_of_image:
-            keypoints = torch.zeros(num_frames, 16)  # 8 keypoints each with 2 coords
-        done = torch.zeros(num_frames, dtype=torch.bool)
-        for i in range(num_frames):
-            space = pymunk.Space()
-            space.gravity = 0, 0
-            space.damping = 0
-
-            # Add walls.
-            walls = [
-                PushTEnv.add_segment(space, (5, 506), (5, 5), 2),
-                PushTEnv.add_segment(space, (5, 5), (506, 5), 2),
-                PushTEnv.add_segment(space, (506, 5), (506, 506), 2),
-                PushTEnv.add_segment(space, (5, 506), (506, 506), 2),
-            ]
-            space.add(*walls)
-
-            block_body, block_shapes = PushTEnv.add_tee(space, block_pos[i].tolist(), block_angle[i].item())
-            goal_geom = pymunk_to_shapely(goal_body, block_body.shapes)
-            block_geom = pymunk_to_shapely(block_body, block_body.shapes)
-            intersection_area = goal_geom.intersection(block_geom).area
-            goal_area = goal_geom.area
-            coverage = intersection_area / goal_area
-            reward[i] = np.clip(coverage / success_threshold, 0, 1)
-            success[i] = coverage > success_threshold
-            if keypoints_instead_of_image:
-                keypoints[i] = torch.from_numpy(PushTEnv.get_keypoints(block_shapes).flatten())
-
-        # last step of demonstration is considered done
-        done[-1] = True
-
-        ep_dict = {}
-
-        if not keypoints_instead_of_image:
-            imgs_array = [x.numpy() for x in image]
-            img_key = "observation.image"
-            if video:
-                # save png images in temporary directory
-                tmp_imgs_dir = videos_dir / "tmp_images"
-                save_images_concurrently(imgs_array, tmp_imgs_dir)
-
-                # encode images to a mp4 video
-                fname = f"{img_key}_episode_{ep_idx:06d}.mp4"
-                video_path = videos_dir / fname
-                encode_video_frames(tmp_imgs_dir, video_path, fps, **(encoding or {}))
-
-                # clean temporary images directory
-                shutil.rmtree(tmp_imgs_dir)
-
-                # store the reference to the video frame
-                ep_dict[img_key] = [
-                    {"path": f"videos/{fname}", "timestamp": i / fps} for i in range(num_frames)
-                ]
-            else:
-                ep_dict[img_key] = [PILImage.fromarray(x) for x in imgs_array]
-
-        ep_dict["observation.state"] = agent_pos
-        if keypoints_instead_of_image:
-            ep_dict["observation.environment_state"] = keypoints
-        ep_dict["action"] = actions[from_idx:to_idx]
-        ep_dict["episode_index"] = torch.tensor([ep_idx] * num_frames, dtype=torch.int64)
-        ep_dict["frame_index"] = torch.arange(0, num_frames, 1)
-        ep_dict["timestamp"] = torch.arange(0, num_frames, 1) / fps
-        # ep_dict["next.observation.image"] = image[1:],
-        # ep_dict["next.observation.state"] = agent_pos[1:],
-        # TODO(rcadene)] = verify that reward and done are aligned with image and agent_pos
-        ep_dict["next.reward"] = torch.cat([reward[1:], reward[[-1]]])
-        ep_dict["next.done"] = torch.cat([done[1:], done[[-1]]])
-        ep_dict["next.success"] = torch.cat([success[1:], success[[-1]]])
-        ep_dicts.append(ep_dict)
-    data_dict = concatenate_episodes(ep_dicts)
-
-    total_frames = data_dict["frame_index"].shape[0]
-    data_dict["index"] = torch.arange(0, total_frames, 1)
-    return data_dict
-
-
-def to_hf_dataset(data_dict, video, keypoints_instead_of_image: bool = False):
-    features = {}
-
-    if not keypoints_instead_of_image:
-        if video:
-            features["observation.image"] = VideoFrame()
-        else:
-            features["observation.image"] = Image()
-
-    features["observation.state"] = Sequence(
-        length=data_dict["observation.state"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    if keypoints_instead_of_image:
-        features["observation.environment_state"] = Sequence(
-            length=data_dict["observation.environment_state"].shape[1],
-            feature=Value(dtype="float32", id=None),
-        )
-    features["action"] = Sequence(
-        length=data_dict["action"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["episode_index"] = Value(dtype="int64", id=None)
-    features["frame_index"] = Value(dtype="int64", id=None)
-    features["timestamp"] = Value(dtype="float32", id=None)
-    features["next.reward"] = Value(dtype="float32", id=None)
-    features["next.done"] = Value(dtype="bool", id=None)
-    features["next.success"] = Value(dtype="bool", id=None)
-    features["index"] = Value(dtype="int64", id=None)
-
-    hf_dataset = Dataset.from_dict(data_dict, features=Features(features))
-    hf_dataset.set_transform(hf_transform_to_torch)
-    return hf_dataset
-
-
-def from_raw_to_lerobot_format(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int | None = None,
-    video: bool = True,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    # Manually change this to True to use keypoints of the T instead of an image observation (but don't merge
-    # with True). Also make sure to use video = 0 in the `push_dataset_to_hub.py` script.
-    keypoints_instead_of_image = False
-
-    # sanity check
-    check_format(raw_dir)
-
-    if fps is None:
-        fps = 10
-
-    data_dict = load_from_raw(raw_dir, videos_dir, fps, video, episodes, keypoints_instead_of_image, encoding)
-    hf_dataset = to_hf_dataset(data_dict, video, keypoints_instead_of_image)
-    episode_data_index = calculate_episode_data_index(hf_dataset)
-    info = {
-        "codebase_version": CODEBASE_VERSION,
-        "fps": fps,
-        "video": video if not keypoints_instead_of_image else 0,
-    }
-    if video:
-        info["encoding"] = get_default_encoding()
-
-    return hf_dataset, episode_data_index, info

lerobot/common/datasets/push_dataset_to_hub/umi_zarr_format.py

@@ -1,234 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Process UMI (Universal Manipulation Interface) data stored in Zarr format like in: https://github.com/real-stanford/universal_manipulation_interface"""
-
-import logging
-import shutil
-from pathlib import Path
-
-import torch
-import tqdm
-import zarr
-from datasets import Dataset, Features, Image, Sequence, Value
-from PIL import Image as PILImage
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub._umi_imagecodecs_numcodecs import register_codecs
-from lerobot.common.datasets.push_dataset_to_hub.utils import (
-    calculate_episode_data_index,
-    concatenate_episodes,
-    get_default_encoding,
-    save_images_concurrently,
-)
-from lerobot.common.datasets.utils import (
-    hf_transform_to_torch,
-)
-from lerobot.common.datasets.video_utils import VideoFrame, encode_video_frames
-
-
-def check_format(raw_dir) -> bool:
-    zarr_path = raw_dir / "cup_in_the_wild.zarr"
-    zarr_data = zarr.open(zarr_path, mode="r")
-
-    required_datasets = {
-        "data/robot0_demo_end_pose",
-        "data/robot0_demo_start_pose",
-        "data/robot0_eef_pos",
-        "data/robot0_eef_rot_axis_angle",
-        "data/robot0_gripper_width",
-        "meta/episode_ends",
-        "data/camera0_rgb",
-    }
-    for dataset in required_datasets:
-        if dataset not in zarr_data:
-            return False
-
-    # mandatory to access zarr_data
-    register_codecs()
-    nb_frames = zarr_data["data/camera0_rgb"].shape[0]
-
-    required_datasets.remove("meta/episode_ends")
-    assert all(nb_frames == zarr_data[dataset].shape[0] for dataset in required_datasets)
-
-
-def load_from_raw(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int,
-    video: bool,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    zarr_path = raw_dir / "cup_in_the_wild.zarr"
-    zarr_data = zarr.open(zarr_path, mode="r")
-
-    # We process the image data separately because it is too large to fit in memory
-    end_pose = torch.from_numpy(zarr_data["data/robot0_demo_end_pose"][:])
-    start_pos = torch.from_numpy(zarr_data["data/robot0_demo_start_pose"][:])
-    eff_pos = torch.from_numpy(zarr_data["data/robot0_eef_pos"][:])
-    eff_rot_axis_angle = torch.from_numpy(zarr_data["data/robot0_eef_rot_axis_angle"][:])
-    gripper_width = torch.from_numpy(zarr_data["data/robot0_gripper_width"][:])
-
-    states_pos = torch.cat([eff_pos, eff_rot_axis_angle], dim=1)
-    states = torch.cat([states_pos, gripper_width], dim=1)
-
-    episode_ends = zarr_data["meta/episode_ends"][:]
-    num_episodes = episode_ends.shape[0]
-
-    # We convert it in torch tensor later because the jit function does not support torch tensors
-    episode_ends = torch.from_numpy(episode_ends)
-
-    # load data indices from which each episode starts and ends
-    from_ids, to_ids = [], []
-    from_idx = 0
-    for to_idx in episode_ends:
-        from_ids.append(from_idx)
-        to_ids.append(to_idx)
-        from_idx = to_idx
-
-    ep_dicts_dir = videos_dir / "ep_dicts"
-    ep_dicts_dir.mkdir(exist_ok=True, parents=True)
-    ep_dicts = []
-
-    ep_ids = episodes if episodes else range(num_episodes)
-    for ep_idx, selected_ep_idx in tqdm.tqdm(enumerate(ep_ids)):
-        ep_dict_path = ep_dicts_dir / f"{ep_idx}"
-        if not ep_dict_path.is_file():
-            from_idx = from_ids[selected_ep_idx]
-            to_idx = to_ids[selected_ep_idx]
-            num_frames = to_idx - from_idx
-
-            # TODO(rcadene): save temporary images of the episode?
-
-            state = states[from_idx:to_idx]
-
-            ep_dict = {}
-
-            # load 57MB of images in RAM (400x224x224x3 uint8)
-            imgs_array = zarr_data["data/camera0_rgb"][from_idx:to_idx]
-            img_key = "observation.image"
-            if video:
-                fname = f"{img_key}_episode_{ep_idx:06d}.mp4"
-                video_path = videos_dir / fname
-                if not video_path.is_file():
-                    # save png images in temporary directory
-                    tmp_imgs_dir = videos_dir / "tmp_images"
-                    save_images_concurrently(imgs_array, tmp_imgs_dir)
-
-                    # encode images to a mp4 video
-                    encode_video_frames(tmp_imgs_dir, video_path, fps, **(encoding or {}))
-
-                    # clean temporary images directory
-                    shutil.rmtree(tmp_imgs_dir)
-
-                # store the reference to the video frame
-                ep_dict[img_key] = [
-                    {"path": f"videos/{fname}", "timestamp": i / fps} for i in range(num_frames)
-                ]
-            else:
-                ep_dict[img_key] = [PILImage.fromarray(x) for x in imgs_array]
-
-            ep_dict["observation.state"] = state
-            ep_dict["episode_index"] = torch.tensor([ep_idx] * num_frames, dtype=torch.int64)
-            ep_dict["frame_index"] = torch.arange(0, num_frames, 1)
-            ep_dict["timestamp"] = torch.arange(0, num_frames, 1) / fps
-            ep_dict["episode_data_index_from"] = torch.tensor([from_idx] * num_frames)
-            ep_dict["episode_data_index_to"] = torch.tensor([from_idx + num_frames] * num_frames)
-            ep_dict["end_pose"] = end_pose[from_idx:to_idx]
-            ep_dict["start_pos"] = start_pos[from_idx:to_idx]
-            ep_dict["gripper_width"] = gripper_width[from_idx:to_idx]
-            torch.save(ep_dict, ep_dict_path)
-        else:
-            ep_dict = torch.load(ep_dict_path)
-
-        ep_dicts.append(ep_dict)
-
-    data_dict = concatenate_episodes(ep_dicts)
-
-    total_frames = data_dict["frame_index"].shape[0]
-    data_dict["index"] = torch.arange(0, total_frames, 1)
-    return data_dict
-
-
-def to_hf_dataset(data_dict, video):
-    features = {}
-
-    if video:
-        features["observation.image"] = VideoFrame()
-    else:
-        features["observation.image"] = Image()
-
-    features["observation.state"] = Sequence(
-        length=data_dict["observation.state"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["episode_index"] = Value(dtype="int64", id=None)
-    features["frame_index"] = Value(dtype="int64", id=None)
-    features["timestamp"] = Value(dtype="float32", id=None)
-    features["index"] = Value(dtype="int64", id=None)
-    features["episode_data_index_from"] = Value(dtype="int64", id=None)
-    features["episode_data_index_to"] = Value(dtype="int64", id=None)
-    # `start_pos` and `end_pos` respectively represent the positions of the end-effector
-    # at the beginning and the end of the episode.
-    # `gripper_width` indicates the distance between the grippers, and this value is included
-    # in the state vector, which comprises the concatenation of the end-effector position
-    # and gripper width.
-    features["end_pose"] = Sequence(
-        length=data_dict["end_pose"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["start_pos"] = Sequence(
-        length=data_dict["start_pos"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["gripper_width"] = Sequence(
-        length=data_dict["gripper_width"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-
-    hf_dataset = Dataset.from_dict(data_dict, features=Features(features))
-    hf_dataset.set_transform(hf_transform_to_torch)
-    return hf_dataset
-
-
-def from_raw_to_lerobot_format(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int | None = None,
-    video: bool = True,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    # sanity check
-    check_format(raw_dir)
-
-    if fps is None:
-        # For umi cup in the wild: https://arxiv.org/pdf/2402.10329#table.caption.16
-        fps = 10
-
-    if not video:
-        logging.warning(
-            "Generating UMI dataset without `video=True` creates ~150GB on disk and requires ~80GB in RAM."
-        )
-
-    data_dict = load_from_raw(raw_dir, videos_dir, fps, video, episodes, encoding)
-    hf_dataset = to_hf_dataset(data_dict, video)
-    episode_data_index = calculate_episode_data_index(hf_dataset)
-    info = {
-        "codebase_version": CODEBASE_VERSION,
-        "fps": fps,
-        "video": video,
-    }
-    if video:
-        info["encoding"] = get_default_encoding()
-
-    return hf_dataset, episode_data_index, info

lerobot/common/datasets/push_dataset_to_hub/xarm_pkl_format.py

@@ -1,200 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Process pickle files formatted like in: https://github.com/fyhMer/fowm"""
-
-import pickle
-import shutil
-from pathlib import Path
-
-import einops
-import torch
-import tqdm
-from datasets import Dataset, Features, Image, Sequence, Value
-from PIL import Image as PILImage
-
-from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION
-from lerobot.common.datasets.push_dataset_to_hub.utils import (
-    calculate_episode_data_index,
-    concatenate_episodes,
-    get_default_encoding,
-    save_images_concurrently,
-)
-from lerobot.common.datasets.utils import (
-    hf_transform_to_torch,
-)
-from lerobot.common.datasets.video_utils import VideoFrame, encode_video_frames
-
-
-def check_format(raw_dir):
-    keys = {"actions", "rewards", "dones"}
-    nested_keys = {"observations": {"rgb", "state"}, "next_observations": {"rgb", "state"}}
-
-    xarm_files = list(raw_dir.glob("*.pkl"))
-    assert len(xarm_files) > 0
-
-    with open(xarm_files[0], "rb") as f:
-        dataset_dict = pickle.load(f)
-
-    assert isinstance(dataset_dict, dict)
-    assert all(k in dataset_dict for k in keys)
-
-    # Check for consistent lengths in nested keys
-    expected_len = len(dataset_dict["actions"])
-    assert all(len(dataset_dict[key]) == expected_len for key in keys if key in dataset_dict)
-
-    for key, subkeys in nested_keys.items():
-        nested_dict = dataset_dict.get(key, {})
-        assert all(len(nested_dict[subkey]) == expected_len for subkey in subkeys if subkey in nested_dict)
-
-
-def load_from_raw(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int,
-    video: bool,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    pkl_path = raw_dir / "buffer.pkl"
-
-    with open(pkl_path, "rb") as f:
-        pkl_data = pickle.load(f)
-
-    # load data indices from which each episode starts and ends
-    from_ids, to_ids = [], []
-    from_idx, to_idx = 0, 0
-    for done in pkl_data["dones"]:
-        to_idx += 1
-        if not done:
-            continue
-        from_ids.append(from_idx)
-        to_ids.append(to_idx)
-        from_idx = to_idx
-
-    num_episodes = len(from_ids)
-
-    ep_dicts = []
-    ep_ids = episodes if episodes else range(num_episodes)
-    for ep_idx, selected_ep_idx in tqdm.tqdm(enumerate(ep_ids)):
-        from_idx = from_ids[selected_ep_idx]
-        to_idx = to_ids[selected_ep_idx]
-        num_frames = to_idx - from_idx
-
-        image = torch.tensor(pkl_data["observations"]["rgb"][from_idx:to_idx])
-        image = einops.rearrange(image, "b c h w -> b h w c")
-        state = torch.tensor(pkl_data["observations"]["state"][from_idx:to_idx])
-        action = torch.tensor(pkl_data["actions"][from_idx:to_idx])
-        # TODO(rcadene): we have a missing last frame which is the observation when the env is done
-        # it is critical to have this frame for tdmpc to predict a "done observation/state"
-        # next_image = torch.tensor(pkl_data["next_observations"]["rgb"][from_idx:to_idx])
-        # next_state = torch.tensor(pkl_data["next_observations"]["state"][from_idx:to_idx])
-        next_reward = torch.tensor(pkl_data["rewards"][from_idx:to_idx])
-        next_done = torch.tensor(pkl_data["dones"][from_idx:to_idx])
-
-        ep_dict = {}
-
-        imgs_array = [x.numpy() for x in image]
-        img_key = "observation.image"
-        if video:
-            # save png images in temporary directory
-            tmp_imgs_dir = videos_dir / "tmp_images"
-            save_images_concurrently(imgs_array, tmp_imgs_dir)
-
-            # encode images to a mp4 video
-            fname = f"{img_key}_episode_{ep_idx:06d}.mp4"
-            video_path = videos_dir / fname
-            encode_video_frames(tmp_imgs_dir, video_path, fps, **(encoding or {}))
-
-            # clean temporary images directory
-            shutil.rmtree(tmp_imgs_dir)
-
-            # store the reference to the video frame
-            ep_dict[img_key] = [{"path": f"videos/{fname}", "timestamp": i / fps} for i in range(num_frames)]
-        else:
-            ep_dict[img_key] = [PILImage.fromarray(x) for x in imgs_array]
-
-        ep_dict["observation.state"] = state
-        ep_dict["action"] = action
-        ep_dict["episode_index"] = torch.tensor([ep_idx] * num_frames, dtype=torch.int64)
-        ep_dict["frame_index"] = torch.arange(0, num_frames, 1)
-        ep_dict["timestamp"] = torch.arange(0, num_frames, 1) / fps
-        # ep_dict["next.observation.image"] = next_image
-        # ep_dict["next.observation.state"] = next_state
-        ep_dict["next.reward"] = next_reward
-        ep_dict["next.done"] = next_done
-        ep_dicts.append(ep_dict)
-
-    data_dict = concatenate_episodes(ep_dicts)
-
-    total_frames = data_dict["frame_index"].shape[0]
-    data_dict["index"] = torch.arange(0, total_frames, 1)
-    return data_dict
-
-
-def to_hf_dataset(data_dict, video):
-    features = {}
-
-    if video:
-        features["observation.image"] = VideoFrame()
-    else:
-        features["observation.image"] = Image()
-
-    features["observation.state"] = Sequence(
-        length=data_dict["observation.state"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["action"] = Sequence(
-        length=data_dict["action"].shape[1], feature=Value(dtype="float32", id=None)
-    )
-    features["episode_index"] = Value(dtype="int64", id=None)
-    features["frame_index"] = Value(dtype="int64", id=None)
-    features["timestamp"] = Value(dtype="float32", id=None)
-    features["next.reward"] = Value(dtype="float32", id=None)
-    features["next.done"] = Value(dtype="bool", id=None)
-    features["index"] = Value(dtype="int64", id=None)
-    # TODO(rcadene): add success
-    # features["next.success"] = Value(dtype='bool', id=None)
-
-    hf_dataset = Dataset.from_dict(data_dict, features=Features(features))
-    hf_dataset.set_transform(hf_transform_to_torch)
-    return hf_dataset
-
-
-def from_raw_to_lerobot_format(
-    raw_dir: Path,
-    videos_dir: Path,
-    fps: int | None = None,
-    video: bool = True,
-    episodes: list[int] | None = None,
-    encoding: dict | None = None,
-):
-    # sanity check
-    check_format(raw_dir)
-
-    if fps is None:
-        fps = 15
-
-    data_dict = load_from_raw(raw_dir, videos_dir, fps, video, episodes, encoding)
-    hf_dataset = to_hf_dataset(data_dict, video)
-    episode_data_index = calculate_episode_data_index(hf_dataset)
-    info = {
-        "codebase_version": CODEBASE_VERSION,
-        "fps": fps,
-        "video": video,
-    }
-    if video:
-        info["encoding"] = get_default_encoding()
-
-    return hf_dataset, episode_data_index, info

lerobot/common/datasets/transforms.py

@@ -128,7 +128,7 @@ class SharpnessJitter(Transform):
             raise TypeError(f"{sharpness=} should be a single number or a sequence with length 2.")
 
         if not 0.0 <= sharpness[0] <= sharpness[1]:
-            raise ValueError(f"sharpnesss values should be between (0., inf), but got {sharpness}.")
+            raise ValueError(f"sharpness values should be between (0., inf), but got {sharpness}.")
 
         return float(sharpness[0]), float(sharpness[1])
 

lerobot/common/datasets/utils.py

@@ -40,7 +40,7 @@ from lerobot.common.datasets.backward_compatibility import (
     BackwardCompatibilityError,
     ForwardCompatibilityError,
 )
-from lerobot.common.robots.utils import Robot
+from lerobot.common.robots import Robot
 from lerobot.common.utils.utils import is_valid_numpy_dtype_string
 from lerobot.configs.types import DictLike, FeatureType, PolicyFeature
 
@@ -240,7 +240,7 @@ def load_episodes_stats(local_dir: Path) -> dict:
 def backward_compatible_episodes_stats(
     stats: dict[str, dict[str, np.ndarray]], episodes: list[int]
 ) -> dict[str, dict[str, np.ndarray]]:
-    return {ep_idx: stats for ep_idx in episodes}
+    return dict.fromkeys(episodes, stats)
 
 
 def load_image_as_numpy(
@@ -387,6 +387,52 @@ def get_hf_features_from_features(features: dict) -> datasets.Features:
     return datasets.Features(hf_features)
 
 
+def _validate_feature_names(features: dict[str, dict]) -> None:
+    invalid_features = {name: ft for name, ft in features.items() if "/" in name}
+    if invalid_features:
+        raise ValueError(f"Feature names should not contain '/'. Found '/' in '{invalid_features}'.")
+
+
+def hw_to_dataset_features(
+    hw_features: dict[str, type | tuple], prefix: str, use_video: bool = True
+) -> dict[str, dict]:
+    features = {}
+    joint_fts = {key: ftype for key, ftype in hw_features.items() if ftype is float}
+    cam_fts = {key: shape for key, shape in hw_features.items() if isinstance(shape, tuple)}
+
+    if joint_fts:
+        features[f"{prefix}.joints"] = {
+            "dtype": "float32",
+            "shape": (len(joint_fts),),
+            "names": list(joint_fts),
+        }
+
+    for key, shape in cam_fts.items():
+        features[f"{prefix}.cameras.{key}"] = {
+            "dtype": "video" if use_video else "image",
+            "shape": shape,
+            "names": ["height", "width", "channels"],
+        }
+
+    _validate_feature_names(features)
+    return features
+
+
+def build_dataset_frame(
+    ds_features: dict[str, dict], values: dict[str, Any], prefix: str
+) -> dict[str, np.ndarray]:
+    frame = {}
+    for key, ft in ds_features.items():
+        if key in DEFAULT_FEATURES or not key.startswith(prefix):
+            continue
+        elif ft["dtype"] == "float32" and len(ft["shape"]) == 1:
+            frame[key] = np.array([values[name] for name in ft["names"]], dtype=np.float32)
+        elif ft["dtype"] in ["image", "video"]:
+            frame[key] = values[key.removeprefix(f"{prefix}.cameras.")]
+
+    return frame
+
+
 def get_features_from_robot(robot: Robot, use_videos: bool = True) -> dict:
     camera_ft = {}
     if robot.cameras:
@@ -431,9 +477,9 @@ def dataset_to_policy_features(features: dict[str, dict]) -> dict[str, PolicyFea
 def create_empty_dataset_info(
     codebase_version: str,
     fps: int,
-    robot_type: str,
     features: dict,
     use_videos: bool,
+    robot_type: str | None = None,
 ) -> dict:
     return {
         "codebase_version": codebase_version,
@@ -699,16 +745,12 @@ class IterableNamespace(SimpleNamespace):
 
 
 def validate_frame(frame: dict, features: dict):
-    optional_features = {"timestamp"}
-    expected_features = (set(features) - set(DEFAULT_FEATURES.keys())) | {"task"}
-    actual_features = set(frame.keys())
+    expected_features = set(features) - set(DEFAULT_FEATURES)
+    actual_features = set(frame)
 
-    error_message = validate_features_presence(actual_features, expected_features, optional_features)
+    error_message = validate_features_presence(actual_features, expected_features)
 
-    if "task" in frame:
-        error_message += validate_feature_string("task", frame["task"])
-
-    common_features = actual_features & (expected_features | optional_features)
+    common_features = actual_features & expected_features
     for name in common_features - {"task"}:
         error_message += validate_feature_dtype_and_shape(name, features[name], frame[name])
 
@@ -716,12 +758,10 @@ def validate_frame(frame: dict, features: dict):
         raise ValueError(error_message)
 
 
-def validate_features_presence(
-    actual_features: set[str], expected_features: set[str], optional_features: set[str]
-):
+def validate_features_presence(actual_features: set[str], expected_features: set[str]):
     error_message = ""
     missing_features = expected_features - actual_features
-    extra_features = actual_features - (expected_features | optional_features)
+    extra_features = actual_features - expected_features
 
     if missing_features or extra_features:
         error_message += "Feature mismatch in `frame` dictionary:\n"

lerobot/common/datasets/v2/convert_dataset_v1_to_v2.py

@@ -481,7 +481,7 @@ def convert_dataset(
 
     # Tasks
     if single_task:
-        tasks_by_episodes = {ep_idx: single_task for ep_idx in episode_indices}
+        tasks_by_episodes = dict.fromkeys(episode_indices, single_task)
         dataset, tasks = add_task_index_by_episodes(dataset, tasks_by_episodes)
         tasks_by_episodes = {ep_idx: [task] for ep_idx, task in tasks_by_episodes.items()}
     elif tasks_path:

lerobot/common/datasets/video_utils.py

@@ -13,15 +13,15 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-import json
+import glob
+import importlib
 import logging
-import subprocess
 import warnings
-from collections import OrderedDict
 from dataclasses import dataclass, field
 from pathlib import Path
 from typing import Any, ClassVar
 
+import av
 import pyarrow as pa
 import torch
 import torchvision
@@ -29,6 +29,46 @@ from datasets.features.features import register_feature
 from PIL import Image
 
 
+def get_safe_default_codec():
+    if importlib.util.find_spec("torchcodec"):
+        return "torchcodec"
+    else:
+        logging.warning(
+            "'torchcodec' is not available in your platform, falling back to 'pyav' as a default decoder"
+        )
+        return "pyav"
+
+
+def decode_video_frames(
+    video_path: Path | str,
+    timestamps: list[float],
+    tolerance_s: float,
+    backend: str | None = None,
+) -> torch.Tensor:
+    """
+    Decodes video frames using the specified backend.
+
+    Args:
+        video_path (Path): Path to the video file.
+        timestamps (list[float]): List of timestamps to extract frames.
+        tolerance_s (float): Allowed deviation in seconds for frame retrieval.
+        backend (str, optional): Backend to use for decoding. Defaults to "torchcodec" when available in the platform; otherwise, defaults to "pyav"..
+
+    Returns:
+        torch.Tensor: Decoded frames.
+
+    Currently supports torchcodec on cpu and pyav.
+    """
+    if backend is None:
+        backend = get_safe_default_codec()
+    if backend == "torchcodec":
+        return decode_video_frames_torchcodec(video_path, timestamps, tolerance_s)
+    elif backend in ["pyav", "video_reader"]:
+        return decode_video_frames_torchvision(video_path, timestamps, tolerance_s, backend)
+    else:
+        raise ValueError(f"Unsupported video backend: {backend}")
+
+
 def decode_video_frames_torchvision(
     video_path: Path | str,
     timestamps: list[float],
@@ -127,6 +167,81 @@ def decode_video_frames_torchvision(
     return closest_frames
 
 
+def decode_video_frames_torchcodec(
+    video_path: Path | str,
+    timestamps: list[float],
+    tolerance_s: float,
+    device: str = "cpu",
+    log_loaded_timestamps: bool = False,
+) -> torch.Tensor:
+    """Loads frames associated with the requested timestamps of a video using torchcodec.
+
+    Note: Setting device="cuda" outside the main process, e.g. in data loader workers, will lead to CUDA initialization errors.
+
+    Note: Video benefits from inter-frame compression. Instead of storing every frame individually,
+    the encoder stores a reference frame (or a key frame) and subsequent frames as differences relative to
+    that key frame. As a consequence, to access a requested frame, we need to load the preceding key frame,
+    and all subsequent frames until reaching the requested frame. The number of key frames in a video
+    can be adjusted during encoding to take into account decoding time and video size in bytes.
+    """
+
+    if importlib.util.find_spec("torchcodec"):
+        from torchcodec.decoders import VideoDecoder
+    else:
+        raise ImportError("torchcodec is required but not available.")
+
+    # initialize video decoder
+    decoder = VideoDecoder(video_path, device=device, seek_mode="approximate")
+    loaded_frames = []
+    loaded_ts = []
+    # get metadata for frame information
+    metadata = decoder.metadata
+    average_fps = metadata.average_fps
+
+    # convert timestamps to frame indices
+    frame_indices = [round(ts * average_fps) for ts in timestamps]
+
+    # retrieve frames based on indices
+    frames_batch = decoder.get_frames_at(indices=frame_indices)
+
+    for frame, pts in zip(frames_batch.data, frames_batch.pts_seconds, strict=False):
+        loaded_frames.append(frame)
+        loaded_ts.append(pts.item())
+        if log_loaded_timestamps:
+            logging.info(f"Frame loaded at timestamp={pts:.4f}")
+
+    query_ts = torch.tensor(timestamps)
+    loaded_ts = torch.tensor(loaded_ts)
+
+    # compute distances between each query timestamp and loaded timestamps
+    dist = torch.cdist(query_ts[:, None], loaded_ts[:, None], p=1)
+    min_, argmin_ = dist.min(1)
+
+    is_within_tol = min_ < tolerance_s
+    assert is_within_tol.all(), (
+        f"One or several query timestamps unexpectedly violate the tolerance ({min_[~is_within_tol]} > {tolerance_s=})."
+        "It means that the closest frame that can be loaded from the video is too far away in time."
+        "This might be due to synchronization issues with timestamps during data collection."
+        "To be safe, we advise to ignore this item during training."
+        f"\nqueried timestamps: {query_ts}"
+        f"\nloaded timestamps: {loaded_ts}"
+        f"\nvideo: {video_path}"
+    )
+
+    # get closest frames to the query timestamps
+    closest_frames = torch.stack([loaded_frames[idx] for idx in argmin_])
+    closest_ts = loaded_ts[argmin_]
+
+    if log_loaded_timestamps:
+        logging.info(f"{closest_ts=}")
+
+    # convert to float32 in [0,1] range (channel first)
+    closest_frames = closest_frames.type(torch.float32) / 255
+
+    assert len(timestamps) == len(closest_frames)
+    return closest_frames
+
+
 def encode_video_frames(
     imgs_dir: Path | str,
     video_path: Path | str,
@@ -136,50 +251,83 @@ def encode_video_frames(
     g: int | None = 2,
     crf: int | None = 30,
     fast_decode: int = 0,
-    log_level: str | None = "error",
+    log_level: int | None = av.logging.ERROR,
     overwrite: bool = False,
 ) -> None:
     """More info on ffmpeg arguments tuning on `benchmark/video/README.md`"""
-    video_path = Path(video_path)
-    video_path.parent.mkdir(parents=True, exist_ok=True)
+    # Check encoder availability
+    if vcodec not in ["h264", "hevc", "libsvtav1"]:
+        raise ValueError(f"Unsupported video codec: {vcodec}. Supported codecs are: h264, hevc, libsvtav1.")
 
-    ffmpeg_args = OrderedDict(
-        [
-            ("-f", "image2"),
-            ("-r", str(fps)),
-            ("-i", str(imgs_dir / "frame_%06d.png")),
-            ("-vcodec", vcodec),
-            ("-pix_fmt", pix_fmt),
-        ]
+    video_path = Path(video_path)
+    imgs_dir = Path(imgs_dir)
+
+    video_path.parent.mkdir(parents=True, exist_ok=overwrite)
+
+    # Encoders/pixel formats incompatibility check
+    if (vcodec == "libsvtav1" or vcodec == "hevc") and pix_fmt == "yuv444p":
+        logging.warning(
+            f"Incompatible pixel format 'yuv444p' for codec {vcodec}, auto-selecting format 'yuv420p'"
+        )
+        pix_fmt = "yuv420p"
+
+    # Get input frames
+    template = "frame_" + ("[0-9]" * 6) + ".png"
+    input_list = sorted(
+        glob.glob(str(imgs_dir / template)), key=lambda x: int(x.split("_")[-1].split(".")[0])
     )
 
+    # Define video output frame size (assuming all input frames are the same size)
+    if len(input_list) == 0:
+        raise FileNotFoundError(f"No images found in {imgs_dir}.")
+    dummy_image = Image.open(input_list[0])
+    width, height = dummy_image.size
+
+    # Define video codec options
+    video_options = {}
+
     if g is not None:
-        ffmpeg_args["-g"] = str(g)
+        video_options["g"] = str(g)
 
     if crf is not None:
-        ffmpeg_args["-crf"] = str(crf)
+        video_options["crf"] = str(crf)
 
     if fast_decode:
-        key = "-svtav1-params" if vcodec == "libsvtav1" else "-tune"
+        key = "svtav1-params" if vcodec == "libsvtav1" else "tune"
         value = f"fast-decode={fast_decode}" if vcodec == "libsvtav1" else "fastdecode"
-        ffmpeg_args[key] = value
+        video_options[key] = value
 
+    # Set logging level
     if log_level is not None:
-        ffmpeg_args["-loglevel"] = str(log_level)
+        # "While less efficient, it is generally preferable to modify logging with Python’s logging"
+        logging.getLogger("libav").setLevel(log_level)
 
-    ffmpeg_args = [item for pair in ffmpeg_args.items() for item in pair]
-    if overwrite:
-        ffmpeg_args.append("-y")
+    # Create and open output file (overwrite by default)
+    with av.open(str(video_path), "w") as output:
+        output_stream = output.add_stream(vcodec, fps, options=video_options)
+        output_stream.pix_fmt = pix_fmt
+        output_stream.width = width
+        output_stream.height = height
 
-    ffmpeg_cmd = ["ffmpeg"] + ffmpeg_args + [str(video_path)]
-    # redirect stdin to subprocess.DEVNULL to prevent reading random keyboard inputs from terminal
-    subprocess.run(ffmpeg_cmd, check=True, stdin=subprocess.DEVNULL)
+        # Loop through input frames and encode them
+        for input_data in input_list:
+            input_image = Image.open(input_data).convert("RGB")
+            input_frame = av.VideoFrame.from_image(input_image)
+            packet = output_stream.encode(input_frame)
+            if packet:
+                output.mux(packet)
+
+        # Flush the encoder
+        packet = output_stream.encode()
+        if packet:
+            output.mux(packet)
+
+    # Reset logging level
+    if log_level is not None:
+        av.logging.restore_default_callback()
 
     if not video_path.exists():
-        raise OSError(
-            f"Video encoding did not work. File not found: {video_path}. "
-            f"Try running the command manually to debug: `{''.join(ffmpeg_cmd)}`"
-        )
+        raise OSError(f"Video encoding did not work. File not found: {video_path}.")
 
 
 @dataclass
@@ -215,78 +363,68 @@ with warnings.catch_warnings():
 
 
 def get_audio_info(video_path: Path | str) -> dict:
-    ffprobe_audio_cmd = [
-        "ffprobe",
-        "-v",
-        "error",
-        "-select_streams",
-        "a:0",
-        "-show_entries",
-        "stream=channels,codec_name,bit_rate,sample_rate,bit_depth,channel_layout,duration",
-        "-of",
-        "json",
-        str(video_path),
-    ]
-    result = subprocess.run(ffprobe_audio_cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
-    if result.returncode != 0:
-        raise RuntimeError(f"Error running ffprobe: {result.stderr}")
+    # Set logging level
+    logging.getLogger("libav").setLevel(av.logging.ERROR)
 
-    info = json.loads(result.stdout)
-    audio_stream_info = info["streams"][0] if info.get("streams") else None
-    if audio_stream_info is None:
-        return {"has_audio": False}
+    # Getting audio stream information
+    audio_info = {}
+    with av.open(str(video_path), "r") as audio_file:
+        try:
+            audio_stream = audio_file.streams.audio[0]
+        except IndexError:
+            # Reset logging level
+            av.logging.restore_default_callback()
+            return {"has_audio": False}
 
-    # Return the information, defaulting to None if no audio stream is present
-    return {
-        "has_audio": True,
-        "audio.channels": audio_stream_info.get("channels", None),
-        "audio.codec": audio_stream_info.get("codec_name", None),
-        "audio.bit_rate": int(audio_stream_info["bit_rate"]) if audio_stream_info.get("bit_rate") else None,
-        "audio.sample_rate": int(audio_stream_info["sample_rate"])
-        if audio_stream_info.get("sample_rate")
-        else None,
-        "audio.bit_depth": audio_stream_info.get("bit_depth", None),
-        "audio.channel_layout": audio_stream_info.get("channel_layout", None),
-    }
+        audio_info["audio.channels"] = audio_stream.channels
+        audio_info["audio.codec"] = audio_stream.codec.canonical_name
+        # In an ideal loseless case : bit depth x sample rate x channels = bit rate.
+        # In an actual compressed case, the bit rate is set according to the compression level : the lower the bit rate, the more compression is applied.
+        audio_info["audio.bit_rate"] = audio_stream.bit_rate
+        audio_info["audio.sample_rate"] = audio_stream.sample_rate  # Number of samples per second
+        # In an ideal loseless case : fixed number of bits per sample.
+        # In an actual compressed case : variable number of bits per sample (often reduced to match a given depth rate).
+        audio_info["audio.bit_depth"] = audio_stream.format.bits
+        audio_info["audio.channel_layout"] = audio_stream.layout.name
+        audio_info["has_audio"] = True
+
+    # Reset logging level
+    av.logging.restore_default_callback()
+
+    return audio_info
 
 
 def get_video_info(video_path: Path | str) -> dict:
-    ffprobe_video_cmd = [
-        "ffprobe",
-        "-v",
-        "error",
-        "-select_streams",
-        "v:0",
-        "-show_entries",
-        "stream=r_frame_rate,width,height,codec_name,nb_frames,duration,pix_fmt",
-        "-of",
-        "json",
-        str(video_path),
-    ]
-    result = subprocess.run(ffprobe_video_cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
-    if result.returncode != 0:
-        raise RuntimeError(f"Error running ffprobe: {result.stderr}")
+    # Set logging level
+    logging.getLogger("libav").setLevel(av.logging.ERROR)
 
-    info = json.loads(result.stdout)
-    video_stream_info = info["streams"][0]
+    # Getting video stream information
+    video_info = {}
+    with av.open(str(video_path), "r") as video_file:
+        try:
+            video_stream = video_file.streams.video[0]
+        except IndexError:
+            # Reset logging level
+            av.logging.restore_default_callback()
+            return {}
 
-    # Calculate fps from r_frame_rate
-    r_frame_rate = video_stream_info["r_frame_rate"]
-    num, denom = map(int, r_frame_rate.split("/"))
-    fps = num / denom
+        video_info["video.height"] = video_stream.height
+        video_info["video.width"] = video_stream.width
+        video_info["video.codec"] = video_stream.codec.canonical_name
+        video_info["video.pix_fmt"] = video_stream.pix_fmt
+        video_info["video.is_depth_map"] = False
 
-    pixel_channels = get_video_pixel_channels(video_stream_info["pix_fmt"])
+        # Calculate fps from r_frame_rate
+        video_info["video.fps"] = int(video_stream.base_rate)
 
-    video_info = {
-        "video.fps": fps,
-        "video.height": video_stream_info["height"],
-        "video.width": video_stream_info["width"],
-        "video.channels": pixel_channels,
-        "video.codec": video_stream_info["codec_name"],
-        "video.pix_fmt": video_stream_info["pix_fmt"],
-        "video.is_depth_map": False,
-        **get_audio_info(video_path),
-    }
+        pixel_channels = get_video_pixel_channels(video_stream.pix_fmt)
+        video_info["video.channels"] = pixel_channels
+
+    # Reset logging level
+    av.logging.restore_default_callback()
+
+    # Adding audio stream information
+    video_info.update(**get_audio_info(video_path))
 
     return video_info
 

lerobot/common/envs/configs.py

@@ -32,7 +32,8 @@ class EnvConfig(draccus.ChoiceRegistry, abc.ABC):
     def type(self) -> str:
         return self.get_choice_name(self.__class__)
 
-    @abc.abstractproperty
+    @property
+    @abc.abstractmethod
     def gym_kwargs(self) -> dict:
         raise NotImplementedError()
 

lerobot/common/envs/utils.py

@@ -13,7 +13,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import warnings
+from typing import Any
+
 import einops
+import gymnasium as gym
 import numpy as np
 import torch
 from torch import Tensor
@@ -86,3 +90,38 @@ def env_to_policy_features(env_cfg: EnvConfig) -> dict[str, PolicyFeature]:
         policy_features[policy_key] = feature
 
     return policy_features
+
+
+def are_all_envs_same_type(env: gym.vector.VectorEnv) -> bool:
+    first_type = type(env.envs[0])  # Get type of first env
+    return all(type(e) is first_type for e in env.envs)  # Fast type check
+
+
+def check_env_attributes_and_types(env: gym.vector.VectorEnv) -> None:
+    with warnings.catch_warnings():
+        warnings.simplefilter("once", UserWarning)  # Apply filter only in this function
+
+        if not (hasattr(env.envs[0], "task_description") and hasattr(env.envs[0], "task")):
+            warnings.warn(
+                "The environment does not have 'task_description' and 'task'. Some policies require these features.",
+                UserWarning,
+                stacklevel=2,
+            )
+        if not are_all_envs_same_type(env):
+            warnings.warn(
+                "The environments have different types. Make sure you infer the right task from each environment. Empty task will be passed instead.",
+                UserWarning,
+                stacklevel=2,
+            )
+
+
+def add_envs_task(env: gym.vector.VectorEnv, observation: dict[str, Any]) -> dict[str, Any]:
+    """Adds task feature to the observation dict with respect to the first environment attribute."""
+    if hasattr(env.envs[0], "task_description"):
+        observation["task"] = env.call("task_description")
+    elif hasattr(env.envs[0], "task"):
+        observation["task"] = env.call("task")
+    else:  #  For envs without language instructions, e.g. aloha transfer cube and etc.
+        num_envs = observation[list(observation.keys())[0]].shape[0]
+        observation["task"] = ["" for _ in range(num_envs)]
+    return observation

lerobot/common/errors.py

@@ -1,3 +1,18 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
 class DeviceNotConnectedError(ConnectionError):
     """Exception raised when the device is not connected."""
 
@@ -15,3 +30,14 @@ class DeviceAlreadyConnectedError(ConnectionError):
     ):
         self.message = message
         super().__init__(self.message)
+
+
+class InvalidActionError(ValueError):
+    """Exception raised when an action is already invalid."""
+
+    def __init__(
+        self,
+        message="The action is invalid. Check the value follows what it is expected from the action space.",
+    ):
+        self.message = message
+        super().__init__(self.message)

lerobot/common/motors/__init__.py

@@ -1,3 +1 @@
-from .motors_bus import MotorsBus
-
-__all__ = ["MotorsBus"]
+from .motors_bus import Motor, MotorCalibration, MotorNormMode, MotorsBus

lerobot/common/motors/dynamixel/__init__.py

@@ -1,4 +1,2 @@
-from .dynamixel import DynamixelMotorsBus, TorqueMode, set_operating_mode
-from .dynamixel_calibration import run_arm_calibration
-
-__all__ = ["DynamixelMotorsBus", "TorqueMode", "set_operating_mode", "run_arm_calibration"]
+from .dynamixel import DriveMode, DynamixelMotorsBus, OperatingMode, TorqueMode
+from .tables import *