Update README.md

Co-authored-by: Simon Alibert <75076266+aliberts@users.noreply.github.com>

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

@@ -14,20 +14,14 @@ env:
 jobs:
   latest-cpu:
     name: CPU
-    runs-on: ubuntu-latest
+    runs-on:
+      group: aws-general-8-plus
     steps:
-      - name: Cleanup disk
+      - name: Install Git LFS
         run: |
-          sudo df -h
-          # sudo ls -l /usr/local/lib/
-          # sudo ls -l /usr/share/
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /usr/share/dotnet
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo df -h
+          sudo apt-get update
+          sudo apt-get install git-lfs
+          git lfs install
 
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
@@ -55,20 +49,15 @@ jobs:
 
   latest-cuda:
     name: GPU
-    runs-on: ubuntu-latest
+    runs-on:
+      group: aws-general-8-plus
     steps:
-      - name: Cleanup disk
+      - name: Install Git LFS
         run: |
-          sudo df -h
-          # sudo ls -l /usr/local/lib/
-          # sudo ls -l /usr/share/
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /usr/share/dotnet
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo df -h
+          sudo apt-get update
+          sudo apt-get install git-lfs
+          git lfs install
+
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
 
@@ -95,20 +84,9 @@ jobs:
 
   latest-cuda-dev:
     name: GPU Dev
-    runs-on: ubuntu-latest
+    runs-on:
+      group: aws-general-8-plus
     steps:
-      - name: Cleanup disk
-        run: |
-          sudo df -h
-          # sudo ls -l /usr/local/lib/
-          # sudo ls -l /usr/share/
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /usr/share/dotnet
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo df -h
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
 

.github/workflows/nightly-tests.yml

@@ -16,7 +16,8 @@ jobs:
     name: CPU
     strategy:
       fail-fast: false
-    runs-on: ubuntu-latest
+    runs-on:
+      group: aws-general-8-plus
     container:
       image: huggingface/lerobot-cpu:latest
       options: --shm-size "16gb"
@@ -43,7 +44,8 @@ jobs:
     name: GPU
     strategy:
       fail-fast: false
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g6-4xlarge-plus
     env:
       CUDA_VISIBLE_DEVICES: "0"
       TEST_TYPE: "single_gpu"

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

@@ -42,26 +42,14 @@ jobs:
   build_modified_dockerfiles:
     name: Build modified Docker images
     needs: get_changed_files
-    runs-on: ubuntu-latest
+    runs-on:
+      group: aws-general-8-plus
     if: ${{ needs.get_changed_files.outputs.matrix }} != ''
     strategy:
       fail-fast: false
       matrix:
         docker-file: ${{ fromJson(needs.get_changed_files.outputs.matrix) }}
     steps:
-      - name: Cleanup disk
-        run: |
-          sudo df -h
-          # sudo ls -l /usr/local/lib/
-          # sudo ls -l /usr/share/
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo rm -rf /usr/local/lib/android
-          sudo rm -rf /usr/share/dotnet
-          sudo du -sh /usr/local/lib/
-          sudo du -sh /usr/share/
-          sudo df -h
-
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
 

.gitignore

@@ -121,6 +121,7 @@ celerybeat.pid
 # Environments
 .env
 .venv
+env/
 venv/
 env.bak/
 venv.bak/

README.md

@@ -22,8 +22,21 @@
 
 </div>
 
+<h2 align="center">
+    <p><a href="https://github.com/huggingface/lerobot/blob/main/examples/7_get_started_with_real_robot.md">Hot new tutorial: Getting started with real-world robots</a></p>
+</h2>
+
+<div align="center">
+    <img src="media/tutorial/koch_v1_1_leader_follower.webp?raw=true" alt="Koch v1.1 leader and follower arms" title="Koch v1.1 leader and follower arms" width="50%">
+    <p>We just dropped an in-depth tutorial on how to build your own robot!</p>
+    <p>Teach it new skills by showing it a few moves with just a laptop.</p>
+    <p>Then watch your homemade robot act autonomously 🤯</p>
+    <p>For more info, see <a href="https://x.com/RemiCadene/status/1825455895561859185">our thread on X</a> or <a href="https://github.com/huggingface/lerobot/blob/main/examples/7_get_started_with_real_robot.md">our tutorial page</a>.</p>
+</div>
+
+
 <h3 align="center">
-    <p>State-of-the-art Machine Learning for real-world robotics</p>
+    <p>State-of-the-art AI for real-world robotics</p>
 </h3>
 
 ---
@@ -65,17 +78,19 @@
 
 Download our source code:
 ```bash
-git clone https://github.com/huggingface/lerobot.git && cd lerobot
+git clone https://github.com/huggingface/lerobot.git
+cd lerobot
 ```
 
 Create a virtual environment with Python 3.10 and activate it, e.g. with [`miniconda`](https://docs.anaconda.com/free/miniconda/index.html):
 ```bash
-conda create -y -n lerobot python=3.10 && conda activate lerobot
+conda create -y -n lerobot python=3.10
+conda activate lerobot
 ```
 
 Install 🤗 LeRobot:
 ```bash
-pip install .
+pip install -e .
 ```
 
 > **NOTE:** Depending on your platform, If you encounter any build errors during this step
@@ -89,7 +104,7 @@ For simulations, 🤗 LeRobot comes with gymnasium environments that can be inst
 
 For instance, to install 🤗 LeRobot with aloha and pusht, use:
 ```bash
-pip install ".[aloha, pusht]"
+pip install -e ".[aloha, pusht]"
 ```
 
 To use [Weights and Biases](https://docs.wandb.ai/quickstart) for experiment tracking, log in with
@@ -114,10 +129,12 @@ wandb login
 |   |   ├── datasets       # various datasets of human demonstrations: aloha, pusht, xarm
 |   |   ├── envs           # various sim environments: aloha, pusht, xarm
 |   |   ├── policies       # various policies: act, diffusion, tdmpc
+|   |   ├── robot_devices  # various real devices: dynamixel motors, opencv cameras, koch robots
 |   |   └── utils          # various utilities
 |   └── scripts          # contains functions to execute via command line
 |       ├── eval.py                 # load policy and evaluate it on an environment
 |       ├── train.py                # train a policy via imitation learning and/or reinforcement learning
+|       ├── control_robot.py        # teleoperate a real robot, record data, run a policy
 |       ├── push_dataset_to_hub.py  # convert your dataset into LeRobot dataset format and upload it to the Hugging Face hub
 |       └── visualize_dataset.py    # load a dataset and render its demonstrations
 ├── outputs               # contains results of scripts execution: logs, videos, model checkpoints

docker/lerobot-cpu/Dockerfile

@@ -9,6 +9,7 @@ ARG DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential cmake \
     libglib2.0-0 libgl1-mesa-glx libegl1-mesa ffmpeg \
+    speech-dispatcher \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 
 # Create virtual environment

docker/lerobot-gpu-dev/Dockerfile

@@ -13,6 +13,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     sed gawk grep curl wget zip unzip \
     tcpdump sysstat screen tmux \
     libglib2.0-0 libgl1-mesa-glx libegl1-mesa \
+    speech-dispatcher \
     python${PYTHON_VERSION} python${PYTHON_VERSION}-venv \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 

docker/lerobot-gpu/Dockerfile

@@ -9,6 +9,7 @@ ARG DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential cmake \
     libglib2.0-0 libgl1-mesa-glx libegl1-mesa ffmpeg \
+    speech-dispatcher \
     python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 

examples/2_evaluate_pretrained_policy.py

@@ -18,8 +18,6 @@ from lerobot.common.policies.diffusion.modeling_diffusion import DiffusionPolicy
 output_directory = Path("outputs/eval/example_pusht_diffusion")
 output_directory.mkdir(parents=True, exist_ok=True)
 
-device = torch.device("cuda")
-
 # Download the diffusion policy for pusht environment
 pretrained_policy_path = Path(snapshot_download("lerobot/diffusion_pusht"))
 # OR uncomment the following to evaluate a policy from the local outputs/train folder.
@@ -27,6 +25,17 @@ pretrained_policy_path = Path(snapshot_download("lerobot/diffusion_pusht"))
 
 policy = DiffusionPolicy.from_pretrained(pretrained_policy_path)
 policy.eval()
+
+# Check if GPU is available
+if torch.cuda.is_available():
+    device = torch.device("cuda")
+    print("GPU is available. Device set to:", device)
+else:
+    device = torch.device("cpu")
+    print(f"GPU is not available. Device set to: {device}. Inference will be slower than on GPU.")
+    # Decrease the number of reverse-diffusion steps (trades off a bit of quality for 10x speed)
+    policy.diffusion.num_inference_steps = 10
+
 policy.to(device)
 
 # Initialize evaluation environment to render two observation types:

lerobot/common/datasets/utils.py

@@ -23,11 +23,19 @@ from typing import Dict
 import datasets
 import torch
 from datasets import load_dataset, load_from_disk
-from huggingface_hub import HfApi, hf_hub_download, snapshot_download
+from huggingface_hub import DatasetCard, HfApi, hf_hub_download, snapshot_download
 from PIL import Image as PILImage
 from safetensors.torch import load_file
 from torchvision import transforms
 
+DATASET_CARD_TEMPLATE = """
+---
+# Metadata will go there
+---
+This dataset was created using [🤗 LeRobot](https://github.com/huggingface/lerobot).
+
+"""
+
 
 def flatten_dict(d, parent_key="", sep="/"):
     """Flatten a nested dictionary structure by collapsing nested keys into one key with a separator.
@@ -385,3 +393,29 @@ def cycle(iterable):
             yield next(iterator)
         except StopIteration:
             iterator = iter(iterable)
+
+
+def create_branch(repo_id, *, branch: str, repo_type: str | None = None):
+    """Create a branch on a existing Hugging Face repo. Delete the branch if it already
+    exists before creating it.
+    """
+    api = HfApi()
+
+    branches = api.list_repo_refs(repo_id, repo_type=repo_type).branches
+    refs = [branch.ref for branch in branches]
+    ref = f"refs/heads/{branch}"
+    if ref in refs:
+        api.delete_branch(repo_id, repo_type=repo_type, branch=branch)
+
+    api.create_branch(repo_id, repo_type=repo_type, branch=branch)
+
+
+def create_lerobot_dataset_card(tags: list | None = None, text: str | None = None) -> DatasetCard:
+    card = DatasetCard(DATASET_CARD_TEMPLATE)
+    card.data.task_categories = ["robotics"]
+    card.data.tags = ["LeRobot"]
+    if tags is not None:
+        card.data.tags += tags
+    if text is not None:
+        card.text += text
+    return card

lerobot/common/datasets/video_utils.py

@@ -210,6 +210,12 @@ def encode_video_frames(
     # redirect stdin to subprocess.DEVNULL to prevent reading random keyboard inputs from terminal
     subprocess.run(ffmpeg_cmd, check=True, stdin=subprocess.DEVNULL)
 
+    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)}`"
+        )
+
 
 @dataclass
 class VideoFrame:

lerobot/common/policies/act/modeling_act.py

@@ -38,7 +38,13 @@ from lerobot.common.policies.act.configuration_act import ACTConfig
 from lerobot.common.policies.normalize import Normalize, Unnormalize
 
 
-class ACTPolicy(nn.Module, PyTorchModelHubMixin):
+class ACTPolicy(
+    nn.Module,
+    PyTorchModelHubMixin,
+    library_name="lerobot",
+    repo_url="https://github.com/huggingface/lerobot",
+    tags=["robotics", "act"],
+):
     """
     Action Chunking Transformer Policy as per Learning Fine-Grained Bimanual Manipulation with Low-Cost
     Hardware (paper: https://arxiv.org/abs/2304.13705, code: https://github.com/tonyzhaozh/act)

lerobot/common/policies/diffusion/modeling_diffusion.py

@@ -43,7 +43,13 @@ from lerobot.common.policies.utils import (
 )
 
 
-class DiffusionPolicy(nn.Module, PyTorchModelHubMixin):
+class DiffusionPolicy(
+    nn.Module,
+    PyTorchModelHubMixin,
+    library_name="lerobot",
+    repo_url="https://github.com/huggingface/lerobot",
+    tags=["robotics", "diffusion-policy"],
+):
     """
     Diffusion Policy as per "Diffusion Policy: Visuomotor Policy Learning via Action Diffusion"
     (paper: https://arxiv.org/abs/2303.04137, code: https://github.com/real-stanford/diffusion_policy).

lerobot/common/policies/tdmpc/modeling_tdmpc.py

@@ -41,7 +41,13 @@ from lerobot.common.policies.tdmpc.configuration_tdmpc import TDMPCConfig
 from lerobot.common.policies.utils import get_device_from_parameters, populate_queues
 
 
-class TDMPCPolicy(nn.Module, PyTorchModelHubMixin):
+class TDMPCPolicy(
+    nn.Module,
+    PyTorchModelHubMixin,
+    library_name="lerobot",
+    repo_url="https://github.com/huggingface/lerobot",
+    tags=["robotics", "tdmpc"],
+):
     """Implementation of TD-MPC learning + inference.
 
     Please note several warnings for this policy.

lerobot/common/policies/vqbet/modeling_vqbet.py

@@ -38,7 +38,13 @@ from lerobot.common.policies.vqbet.vqbet_utils import GPT, ResidualVQ
 # ruff: noqa: N806
 
 
-class VQBeTPolicy(nn.Module, PyTorchModelHubMixin):
+class VQBeTPolicy(
+    nn.Module,
+    PyTorchModelHubMixin,
+    library_name="lerobot",
+    repo_url="https://github.com/huggingface/lerobot",
+    tags=["robotics", "vqbet"],
+):
     """
     VQ-BeT Policy as per "Behavior Generation with Latent Actions"
     """
@@ -289,7 +295,7 @@ class VQBeTModel(nn.Module):
 
         # To input state and observation features into GPT layers, we first project the features to fit the shape of input size of GPT.
         self.state_projector = MLP(
-            config.output_shapes["action"][0], hidden_channels=[self.config.gpt_input_dim]
+            config.input_shapes["observation.state"][0], hidden_channels=[self.config.gpt_input_dim]
         )
         self.rgb_feature_projector = MLP(
             self.rgb_encoder.feature_dim, hidden_channels=[self.config.gpt_input_dim]

lerobot/common/robot_devices/cameras/opencv.py

@@ -5,6 +5,7 @@ This file contains utilities for recording frames from cameras. For more info lo
 import argparse
 import concurrent.futures
 import math
+import platform
 import shutil
 import threading
 import time
@@ -33,8 +34,22 @@ MAX_OPENCV_INDEX = 60
 
 
 def find_camera_indices(raise_when_empty=False, max_index_search_range=MAX_OPENCV_INDEX):
+    if platform.system() == "Linux":
+        # Linux uses camera ports
+        print("Linux detected. Finding available camera indices through scanning '/dev/video*' ports")
+        possible_camera_ids = []
+        for port in Path("/dev").glob("video*"):
+            camera_idx = int(str(port).replace("/dev/video", ""))
+            possible_camera_ids.append(camera_idx)
+    else:
+        print(
+            "Mac or Windows detected. Finding available camera indices through "
+            f"scanning all indices from 0 to {MAX_OPENCV_INDEX}"
+        )
+        possible_camera_ids = range(max_index_search_range)
+
     camera_ids = []
-    for camera_idx in range(max_index_search_range):
+    for camera_idx in possible_camera_ids:
         camera = cv2.VideoCapture(camera_idx)
         is_open = camera.isOpened()
         camera.release()
@@ -45,29 +60,12 @@ def find_camera_indices(raise_when_empty=False, max_index_search_range=MAX_OPENC
 
     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."
+            "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 find_camera_ports(raise_when_empty=False):
-    camera_ports = []
-    for port in Path("/dev").glob("video*"):
-        port = str(port)
-        camera = cv2.VideoCapture(port)
-        is_open = camera.isOpened()
-        camera.release()
-
-        if is_open:
-            print(f"Camera found at port {port}")
-            camera_ports.append(port)
-
-    if raise_when_empty and len(camera_ports) == 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_ports
 
 def save_image(img_array, camera_index, frame_index, images_dir):
     img = Image.fromarray(img_array)
@@ -77,16 +75,10 @@ def save_image(img_array, camera_index, frame_index, images_dir):
 
 
 def save_images_from_cameras(
-    images_dir: Path, camera_ids: list[int] | list[str] | None=None, fps=None, width=None, height=None, record_time_s=2
+    images_dir: Path, camera_ids: list[int] | None = None, fps=None, width=None, height=None, record_time_s=2
 ):
-    
     if camera_ids is None:
-        print("Finding available camera indices through '/dev/video*' ports")
-        camera_ids = find_camera_ports()
-
-        if len(camera_ids) == 0:
-            print(f"Finding available camera indices through scanning all indices from 0 to {MAX_OPENCV_INDEX}")
-            camera_ids = find_camera_indices()
+        camera_ids = find_camera_indices()
 
     print("Connecting cameras")
     cameras = []
@@ -94,13 +86,12 @@ def save_images_from_cameras(
         camera = OpenCVCamera(cam_idx, fps=fps, width=width, height=height)
         camera.connect()
         print(
-            f"OpenCVCamera({camera.camera_index}, fps={camera.fps}, width={camera.width}, height={camera.height}, color_mode={camera.color_mode})"
+            f"OpenCVCamera({camera.camera_index}, fps={camera.fps}, width={camera.width}, "
+            f"height={camera.height}, color_mode={camera.color_mode})"
         )
         cameras.append(camera)
 
-    images_dir = Path(
-        images_dir
-    )
+    images_dir = Path(images_dir)
     if images_dir.exists():
         shutil.rmtree(
             images_dir,
@@ -122,7 +113,7 @@ def save_images_from_cameras(
                 executor.submit(
                     save_image,
                     image,
-                    camera.camera_index_to_int(),
+                    camera.camera_index,
                     frame_index,
                     images_dir,
                 )
@@ -183,7 +174,7 @@ class OpenCVCamera:
     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.
 
-    Example of usage of the class:
+    Example of usage:
     ```python
     camera = OpenCVCamera(camera_index=0)
     camera.connect()
@@ -205,7 +196,7 @@ class OpenCVCamera:
     ```
     """
 
-    def __init__(self, camera_index: int | str, config: OpenCVCameraConfig | None = None, **kwargs):
+    def __init__(self, camera_index: int, config: OpenCVCameraConfig | None = None, **kwargs):
         if config is None:
             config = OpenCVCameraConfig()
         # Overwrite config arguments using kwargs
@@ -223,11 +214,6 @@ class OpenCVCamera:
         self.stop_event = None
         self.color_image = None
         self.logs = {}
-    
-    def camera_index_to_int(self):
-        if isinstance(self.camera_index, int):
-            return self.camera_index
-        return int(self.camera_index.replace("/dev/video", ""))
 
     def connect(self):
         if self.is_connected:
@@ -235,7 +221,13 @@ class OpenCVCamera:
 
         # First create a temporary camera trying to access `camera_index`,
         # and verify it is a valid camera by calling `isOpened`.
-        tmp_camera = cv2.VideoCapture(self.camera_index)
+
+        if platform.system() == "Linux":
+            # Linux uses ports for connecting to cameras
+            tmp_camera = cv2.VideoCapture(f"/dev/video{self.camera_index}")
+        else:
+            tmp_camera = cv2.VideoCapture(self.camera_index)
+
         is_camera_open = tmp_camera.isOpened()
         # Release camera to make it accessible for `find_camera_indices`
         del tmp_camera
@@ -244,12 +236,11 @@ class OpenCVCamera:
         # valid cameras.
         if not is_camera_open:
             # Verify that the provided `camera_index` is valid before printing the traceback
-            available_cam_ids = find_camera_ports()
-            if len(available_cam_ids) == 0:
-                available_cam_ids = find_camera_indices()
+            available_cam_ids = find_camera_indices()
             if self.camera_index not in available_cam_ids:
                 raise ValueError(
-                    f"`camera_index` is expected to be one of these available cameras {available_cam_ids}, but {self.camera_index} is provided instead."
+                    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`."
                 )
 
             raise OSError(f"Can't access camera {self.camera_index}.")
@@ -257,7 +248,10 @@ class OpenCVCamera:
         # 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(self.camera_index)
+        if platform.system() == "Linux":
+            self.camera = cv2.VideoCapture(f"/dev/video{self.camera_index}")
+        else:
+            self.camera = cv2.VideoCapture(self.camera_index)
 
         if self.fps is not None:
             self.camera.set(cv2.CAP_PROP_FPS, self.fps)
@@ -390,7 +384,7 @@ if __name__ == "__main__":
     )
     parser.add_argument(
         "--camera-ids",
-        type=str,
+        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.",

lerobot/common/robot_devices/motors/dynamixel.py

@@ -133,24 +133,18 @@ NUM_READ_RETRY = 10
 NUM_WRITE_RETRY = 10
 
 
-def convert_indices_to_baudrates(values: np.ndarray | list[int], models: list[str]):
-    assert len(values) == len(models)
-    for i in range(len(values)):
-        model = models[i]
-        index = values[i]
-        values[i] = MODEL_BAUDRATE_TABLE[model][index]
-    return values
+def convert_degrees_to_steps(degrees: float | np.ndarray, models: str | list[str]):
+    """This function convert the degree range to the step range for indicating motors rotation.
+    It assums a motor achieves a full rotation by going from -180 degree position to +180.
+    The motor resolution (e.g. 4096) corresponds to the number of steps needed to achieve a full rotation.
+    """
+    if isinstance(degrees, float):
+        degrees = np.array(degrees)
 
-
-def convert_baudrates_to_indices(values: np.ndarray | list[int], models: list[str]):
-    assert len(values) == len(models)
-    for i in range(len(values)):
-        model = models[i]
-        brate = values[i]
-        table_values = list(MODEL_BAUDRATE_TABLE[model].values())
-        table_keys = list(MODEL_BAUDRATE_TABLE[model].keys())
-        values[i] = table_keys[table_values.index(brate)]
-    return values
+    resolutions = [MODEL_RESOLUTION[model] for model in models]
+    steps = degrees / 180 * np.array(resolutions) / 2
+    steps = steps.astype(int)
+    return steps
 
 
 def convert_to_bytes(value, bytes):
@@ -286,13 +280,12 @@ class DynamixelMotorsBus:
     >>> The port of this DynamixelMotorsBus is /dev/tty.usbmodem575E0031751.
     >>> Reconnect the usb cable.
     ```
-    To find the motor indices, use [DynamixelWizzard2](https://emanual.robotis.com/docs/en/software/dynamixel/dynamixel_wizard2).
 
     Example of usage for 1 motor connected to the bus:
     ```python
     motor_name = "gripper"
     motor_index = 6
-    motor_model = "xl330-m077"
+    motor_model = "xl330-m288"
 
     motors_bus = DynamixelMotorsBus(
         port="/dev/tty.usbmodem575E0031751",
@@ -300,7 +293,11 @@ class DynamixelMotorsBus:
     )
     motors_bus.connect()
 
-    motors_bus.teleop_step()
+    position = motors_bus.read("Present_Position")
+
+    # move from a few motor steps as an example
+    few_steps = 30
+    motors_bus.write("Goal_Position", position + few_steps)
 
     # when done, consider disconnecting
     motors_bus.disconnect()
@@ -357,17 +354,17 @@ class DynamixelMotorsBus:
 
         self.port_handler.setPacketTimeoutMillis(TIMEOUT_MS)
 
-        # Set expected baud rate for the bus
+        # Set expected baudrate for the bus
         self.set_bus_baudrate(BAUDRATE)
 
         if not self.are_motors_configured():
-            print(
-                r"/!\ First, verify that all the cables are connected the proper way. If you detect an issue, before making any modification, unplug the power cord to not damage the motors. Rewire correctly. Then plug the power again and relaunch the script."
+            input(
+                "\n/!\\ A configuration issue has been detected with your motors: \n"
+                "If it's the first time that you use these motors, press enter to configure your motors... but before "
+                "verify that all the cables are connected the proper way. If you find an issue, before making a modification, "
+                "kill the python process, unplug the power cord to not damage the motors, rewire correctly, then plug the power "
+                "again and relaunch the script.\n"
             )
-            print(
-                r"/!\ Secondly, if the cables connection look correct and it is the first time that you use these motors, follow these manual steps to configure them."
-            )
-            input("Press Enter to configure your motors...")
             print()
             self.configure_motors()
 
@@ -379,6 +376,8 @@ class DynamixelMotorsBus:
         self.is_connected = True
 
     def are_motors_configured(self):
+        # Only check the motor indices and not baudrate, since if the motor baudrates are incorrect,
+        # a ConnectionError will be raised anyway.
         try:
             return (self.motor_indices == self.read("ID")).all()
         except ConnectionError as e:
@@ -387,6 +386,7 @@ class DynamixelMotorsBus:
 
     def configure_motors(self):
         # TODO(rcadene): This script assumes motors follow the X_SERIES baudrates
+        # TODO(rcadene): Refactor this function with intermediate high-level functions
 
         print("Scanning all baudrates and motor indices")
         all_baudrates = set(X_SERIES_BAUDRATE_TABLE.values())
@@ -406,12 +406,13 @@ class DynamixelMotorsBus:
         # Connect successively one motor to the chain and write a unique random index for each
         for i in range(len(self.motors)):
             self.disconnect()
-            print("1. Unplug the power cord")
-            print(
-                f"2. Plug/unplug minimal number of cables to only have the first {i+1} motor(s) ({self.motor_names[:i+1]}) connected."
+            input(
+                "1. Unplug the power cord\n"
+                "2. Plug/unplug minimal number of cables to only have the first "
+                f"{i+1} motor(s) ({self.motor_names[:i+1]}) connected.\n"
+                "3. Re-plug the power cord\n"
+                "Press Enter to continue..."
             )
-            print("3. Re-plug the power cord.")
-            input("Press Enter to continue...")
             print()
             self.reconnect()
 
@@ -534,15 +535,18 @@ class DynamixelMotorsBus:
         self.calibration = calibration
 
     def apply_calibration(self, values: np.ndarray | list, motor_names: list[str] | None):
-        """Convert from unsigned int32 joint position range [0, 2**32[ to the universal float32 centered degree range [-180.0, 180.0[
+        """Convert from unsigned int32 joint position range [0, 2**32[ to the universal float32 nominal degree range ]-180.0, 180.0[ with
+        a "zero position" at 0 degree.
+
+        Note: We say "nominal degree range" since the motors can take values outside this range. For instance, 190 degrees, if the motor
+        rotate more than a half a turn from the zero position. However, most motors can't rotate more than 180 degrees and will stay in this range.
 
         Joints values are original in [0, 2**32[ (unsigned int32). Each motor are expected to complete a full rotation
         when given a goal position that is + or - their resolution. For instance, dynamixel xl330-m077 have a resolution of 4096, and
         at any position in their original range, let's say the position 56734, they complete a full rotation clockwise by moving to 60830,
-        or anticlockwise by moving to 42638. The position in the original range is arbitrary and might change a lot between each motor.
+        or anticlockwise by moving to 52638. The position in the original range is arbitrary and might change a lot between each motor.
         To harmonize between motors of the same model, different robots, or even models of different brands, we propose to work
-        in the centered degree range [-180, 180[. This function first applies the pre-computed calibration to convert
-        from [0, 2**32[ to [-2048, 2048[, then divide by 2048.
+        in the centered nominal degree range ]-180, 180[.
         """
         if motor_names is None:
             motor_names = self.motor_names
@@ -558,10 +562,10 @@ class DynamixelMotorsBus:
             if drive_mode:
                 values[i] *= -1
 
-            # Convert from range [-2**31, 2**31[ to centered resolution range [-resolution, resolution[ (e.g. [-2048, 2048[)
+            # Convert from range [-2**31, 2**31[ to nominal range ]-resolution, resolution[ (e.g. ]-2048, 2048[)
             values[i] += homing_offset
 
-        # Convert from range [-resolution, resolution[ to the universal float32 centered degree range [-180, 180[
+        # Convert from range ]-resolution, resolution[ to the universal float32 centered degree range ]-180, 180[
         values = values.astype(np.float32)
         for i, name in enumerate(motor_names):
             _, model = self.motors[name]
@@ -571,19 +575,19 @@ class DynamixelMotorsBus:
         return values
 
     def revert_calibration(self, values: np.ndarray | list, motor_names: list[str] | None):
+        """Inverse of `apply_calibration`."""
         if motor_names is None:
             motor_names = self.motor_names
 
-        # Convert from the universal float32 centered degree range [-180, 180[ to centered resolution range [-resolution, resolution[
+        # Convert from the universal float32 centered degree range ]-180, 180[ to resolution range ]-resolution, resolution[
         for i, name in enumerate(motor_names):
             _, model = self.motors[name]
             resolution = self.model_resolution[model]
-
             values[i] = values[i] / 180 * (resolution // 2)
 
         values = np.round(values).astype(np.int32)
 
-        # Convert from range [-resolution, resolution[ to centered signed int32 range [-2**31, 2**31[
+        # Convert from nominal range ]-resolution, resolution[ to centered signed int32 range [-2**31, 2**31[
         for i, name in enumerate(motor_names):
             homing_offset, drive_mode = self.calibration[name]
             values[i] -= homing_offset
@@ -681,6 +685,18 @@ class DynamixelMotorsBus:
         if data_name in CALIBRATION_REQUIRED and self.calibration is not None:
             values = self.apply_calibration(values, motor_names)
 
+            # We expect our motors to stay in a nominal range of [-180, 180] degrees
+            # which corresponds to a half turn rotation.
+            # However, some motors can turn a bit more, hence we extend the nominal range to [-270, 270]
+            # which is less than a full 360 degree rotation.
+            if not np.all((values > -270) & (values < 270)):
+                raise ValueError(
+                    f"Wrong motor position range detected. "
+                    f"Expected to be in [-270, +270] but in [{values.min()}, {values.max()}]. "
+                    "This might be due to a cable connection issue creating an artificial 360 degrees jump in motor values. "
+                    "You need to recalibrate by running: `python lerobot/scripts/control_robot.py calibrate`"
+                )
+
         # log the number of seconds it took to read the data from the motors
         delta_ts_name = get_log_name("delta_timestamp_s", "read", data_name, motor_names)
         self.logs[delta_ts_name] = time.perf_counter() - start_time

lerobot/common/robot_devices/robots/koch.py

@@ -8,125 +8,43 @@ import torch
 
 from lerobot.common.robot_devices.cameras.utils import Camera
 from lerobot.common.robot_devices.motors.dynamixel import (
-    DriveMode,
-    DynamixelMotorsBus,
     OperatingMode,
     TorqueMode,
+    convert_degrees_to_steps,
 )
 from lerobot.common.robot_devices.motors.utils import MotorsBus
 from lerobot.common.robot_devices.utils import RobotDeviceAlreadyConnectedError, RobotDeviceNotConnectedError
 
-URL_HORIZONTAL_POSITION = {
-    "follower": "https://raw.githubusercontent.com/huggingface/lerobot/main/media/koch/follower_horizontal.png",
-    "leader": "https://raw.githubusercontent.com/huggingface/lerobot/main/media/koch/leader_horizontal.png",
-}
-URL_90_DEGREE_POSITION = {
-    "follower": "https://raw.githubusercontent.com/huggingface/lerobot/main/media/koch/follower_90_degree.png",
-    "leader": "https://raw.githubusercontent.com/huggingface/lerobot/main/media/koch/leader_90_degree.png",
-}
-
 ########################################################################
 # Calibration logic
 ########################################################################
 
-# In range ]-2048, 2048[
-TARGET_HORIZONTAL_POSITION = np.array([0, -1024, 1024, 0, -1024, 0])
-TARGET_90_DEGREE_POSITION = np.array([1024, 0, 0, 1024, 0, -1024])
+URL_TEMPLATE = (
+    "https://raw.githubusercontent.com/huggingface/lerobot/main/media/{robot}/{arm}_{position}.webp"
+)
 
-# In range ]-180, 180[
-GRIPPER_OPEN = np.array([-35.156])
+# In nominal degree range ]-180, +180[
+ZERO_POSITION_DEGREE = 0
+ROTATED_POSITION_DEGREE = 90
+GRIPPER_OPEN_DEGREE = 35.156
 
 
-def apply_homing_offset(values: np.array, homing_offset: np.array) -> np.array:
-    for i in range(len(values)):
-        if values[i] is not None:
-            values[i] += homing_offset[i]
-    return values
+def assert_drive_mode(drive_mode):
+    # `drive_mode` is in [0,1] with 0 means original rotation direction for the motor, and 1 means inverted.
+    if not np.all(np.isin(drive_mode, [0, 1])):
+        raise ValueError(f"`drive_mode` contains values other than 0 or 1: ({drive_mode})")
 
 
-def apply_drive_mode(values: np.array, drive_mode: np.array) -> np.array:
-    for i in range(len(values)):
-        if values[i] is not None and drive_mode[i]:
-            values[i] = -values[i]
-    return values
+def apply_drive_mode(position, drive_mode):
+    assert_drive_mode(drive_mode)
+    # Convert `drive_mode` from [0, 1] with 0 indicates original rotation direction and 1 inverted,
+    # to [-1, 1] with 1 indicates original rotation direction and -1 inverted.
+    signed_drive_mode = -(drive_mode * 2 - 1)
+    position *= signed_drive_mode
+    return position
 
 
-def apply_calibration(values: np.array, homing_offset: np.array, drive_mode: np.array) -> np.array:
-    values = apply_drive_mode(values, drive_mode)
-    values = apply_homing_offset(values, homing_offset)
-    return values
-
-
-def revert_calibration(values: np.array, homing_offset: np.array, drive_mode: np.array) -> np.array:
-    """
-    Transform working position into real position for the robot.
-    """
-    values = apply_homing_offset(
-        values,
-        np.array([-homing_offset if homing_offset is not None else None for homing_offset in homing_offset]),
-    )
-    values = apply_drive_mode(values, drive_mode)
-    return values
-
-
-def revert_appropriate_positions(positions: np.array, drive_mode: list[bool]) -> np.array:
-    for i, revert in enumerate(drive_mode):
-        if not revert and positions[i] is not None:
-            positions[i] = -positions[i]
-    return positions
-
-
-def compute_corrections(positions: np.array, drive_mode: list[bool], target_position: np.array) -> np.array:
-    correction = revert_appropriate_positions(positions, drive_mode)
-
-    for i in range(len(positions)):
-        if correction[i] is not None:
-            if drive_mode[i]:
-                correction[i] -= target_position[i]
-            else:
-                correction[i] += target_position[i]
-
-    return correction
-
-
-def compute_nearest_rounded_positions(positions: np.array) -> np.array:
-    return np.array(
-        [
-            round(positions[i] / 1024) * 1024 if positions[i] is not None else None
-            for i in range(len(positions))
-        ]
-    )
-
-
-def compute_homing_offset(
-    arm: DynamixelMotorsBus, drive_mode: list[bool], target_position: np.array
-) -> np.array:
-    # Get the present positions of the servos
-    present_positions = apply_calibration(
-        arm.read("Present_Position"), np.array([0, 0, 0, 0, 0, 0]), drive_mode
-    )
-
-    nearest_positions = compute_nearest_rounded_positions(present_positions)
-    correction = compute_corrections(nearest_positions, drive_mode, target_position)
-    return correction
-
-
-def compute_drive_mode(arm: DynamixelMotorsBus, offset: np.array):
-    # Get current positions
-    present_positions = apply_calibration(
-        arm.read("Present_Position"), offset, np.array([False, False, False, False, False, False])
-    )
-
-    nearest_positions = compute_nearest_rounded_positions(present_positions)
-
-    # construct 'drive_mode' list comparing nearest_positions and TARGET_90_DEGREE_POSITION
-    drive_mode = []
-    for i in range(len(nearest_positions)):
-        drive_mode.append(nearest_positions[i] != TARGET_90_DEGREE_POSITION[i])
-    return drive_mode
-
-
-def reset_arm(arm: MotorsBus):
+def reset_torque_mode(arm: MotorsBus):
     # To be configured, all servos must be in "torque disable" mode
     arm.write("Torque_Enable", TorqueMode.DISABLED.value)
 
@@ -135,55 +53,95 @@ def reset_arm(arm: MotorsBus):
     # you could end up with a servo with a position 0 or 4095 at a crucial point See [
     # https://emanual.robotis.com/docs/en/dxl/x/x_series/#operating-mode11]
     all_motors_except_gripper = [name for name in arm.motor_names if name != "gripper"]
-    arm.write("Operating_Mode", OperatingMode.EXTENDED_POSITION.value, all_motors_except_gripper)
+    if len(all_motors_except_gripper) > 0:
+        arm.write("Operating_Mode", OperatingMode.EXTENDED_POSITION.value, all_motors_except_gripper)
 
-    # TODO(rcadene): why?
-    # Use 'position control current based' for gripper
+    # Use 'position control current based' for gripper to be limited by the limit of the current.
+    # For the follower gripper, it means it can grasp an object without forcing too much even tho,
+    # it's goal position is a complete grasp (both gripper fingers are ordered to join and reach a touch).
+    # For the leader gripper, it means we can use it as a physical trigger, since we can force with our finger
+    # to make it move, and it will move back to its original target position when we release the force.
     arm.write("Operating_Mode", OperatingMode.CURRENT_CONTROLLED_POSITION.value, "gripper")
 
-    # Make sure the native calibration (homing offset abd drive mode) is disabled, since we use our own calibration layer to be more generic
-    arm.write("Homing_Offset", 0)
-    arm.write("Drive_Mode", DriveMode.NON_INVERTED.value)
-
 
 def run_arm_calibration(arm: MotorsBus, name: str, arm_type: str):
-    """Example of usage:
+    """This function ensures that a neural network trained on data collected on a given robot
+    can work on another robot. For instance before calibration, setting a same goal position
+    for each motor of two different robots will get two very different positions. But after calibration,
+    the two robots will move to the same position.To this end, this function computes the homing offset
+    and the drive mode for each motor of a given robot.
+
+    Homing offset is used to shift the motor position to a ]-2048, +2048[ nominal range (when the motor uses 2048 steps
+    to complete a half a turn). This range is set around an arbitrary "zero position" corresponding to all motor positions
+    being 0. During the calibration process, you will need to manually move the robot to this "zero position".
+
+    Drive mode is used to invert the rotation direction of the motor. This is useful when some motors have been assembled
+    in the opposite orientation for some robots. During the calibration process, you will need to manually move the robot
+    to the "rotated position".
+
+    After calibration, the homing offsets and drive modes are stored in a cache.
+
+    Example of usage:
     ```python
     run_arm_calibration(arm, "left", "follower")
     ```
     """
-    reset_arm(arm)
+    reset_torque_mode(arm)
 
-    # TODO(rcadene): document what position 1 mean
-    print(
-        f"Please move the '{name} {arm_type}' arm to the horizontal position (gripper fully closed, see {URL_HORIZONTAL_POSITION[arm_type]})"
-    )
+    print(f"\nRunning calibration of {name} {arm_type}...")
+
+    print("\nMove arm to zero position")
+    print("See: " + URL_TEMPLATE.format(robot="koch", arm=arm_type, position="zero"))
     input("Press Enter to continue...")
 
-    horizontal_homing_offset = compute_homing_offset(
-        arm, [False, False, False, False, False, False], TARGET_HORIZONTAL_POSITION
-    )
+    # We arbitrarely choosed our zero target position to be a straight horizontal position with gripper upwards and closed.
+    # It is easy to identify and all motors are in a "quarter turn" position. Once calibration is done, this position will
+    # corresponds to every motor angle being 0. If you set all 0 as Goal Position, the arm will move in this position.
+    zero_position = convert_degrees_to_steps(ZERO_POSITION_DEGREE, arm.motor_models)
 
-    # TODO(rcadene): document what position 2 mean
-    print(
-        f"Please move the '{name} {arm_type}' arm to the 90 degree position (gripper fully open, see {URL_90_DEGREE_POSITION[arm_type]})"
-    )
+    def _compute_nearest_rounded_position(position, models):
+        # TODO(rcadene): Rework this function since some motors cant physically rotate a quarter turn
+        # (e.g. the gripper of Aloha arms can only rotate ~50 degree)
+        quarter_turn_degree = 90
+        quarter_turn = convert_degrees_to_steps(quarter_turn_degree, models)
+        nearest_pos = np.round(position.astype(float) / quarter_turn) * quarter_turn
+        return nearest_pos.astype(position.dtype)
+
+    # Compute homing offset so that `present_position + homing_offset ~= target_position`.
+    position = arm.read("Present_Position")
+    position = _compute_nearest_rounded_position(position, arm.motor_models)
+    homing_offset = zero_position - position
+
+    print("\nMove arm to rotated target position")
+    print("See: " + URL_TEMPLATE.format(robot="koch", arm=arm_type, position="rotated"))
     input("Press Enter to continue...")
 
-    drive_mode = compute_drive_mode(arm, horizontal_homing_offset)
-    homing_offset = compute_homing_offset(arm, drive_mode, TARGET_90_DEGREE_POSITION)
+    # The rotated target position corresponds to a rotation of a quarter turn from the zero position.
+    # This allows to identify the rotation direction of each motor.
+    # For instance, if the motor rotates 90 degree, and its value is -90 after applying the homing offset, then we know its rotation direction
+    # is inverted. However, for the calibration being successful, we need everyone to follow the same target position.
+    # Sometimes, there is only one possible rotation direction. For instance, if the gripper is closed, there is only one direction which
+    # corresponds to opening the gripper. When the rotation direction is ambiguous, we arbitrarely rotate clockwise from the point of view
+    # of the previous motor in the kinetic chain.
+    rotated_position = convert_degrees_to_steps(ROTATED_POSITION_DEGREE, arm.motor_models)
 
-    # Invert offset for all drive_mode servos
-    for i in range(len(drive_mode)):
-        if drive_mode[i]:
-            homing_offset[i] = -homing_offset[i]
+    # Find drive mode by rotating each motor by a quarter of a turn.
+    # Drive mode indicates if the motor rotation direction should be inverted (=1) or not (=0).
+    position = arm.read("Present_Position")
+    position += homing_offset
+    position = _compute_nearest_rounded_position(position, arm.motor_models)
+    drive_mode = (position != rotated_position).astype(np.int32)
 
-    print("Calibration is done!")
+    # Re-compute homing offset to take into account drive mode
+    position = arm.read("Present_Position")
+    position = apply_drive_mode(position, drive_mode)
+    position = _compute_nearest_rounded_position(position, arm.motor_models)
+    homing_offset = rotated_position - position
 
-    print("=====================================")
-    print("      HOMING_OFFSET: ", " ".join([str(i) for i in homing_offset]))
-    print("      DRIVE_MODE: ", " ".join([str(i) for i in drive_mode]))
-    print("=====================================")
+    print("\nMove arm to rest position")
+    print("See: " + URL_TEMPLATE.format(robot="koch", arm=arm_type, position="rest"))
+    input("Press Enter to continue...")
+    print()
 
     return homing_offset, drive_mode
 
@@ -210,7 +168,12 @@ class KochRobotConfig:
 
 class KochRobot:
     # TODO(rcadene): Implement force feedback
-    """Tau Robotics: https://tau-robotics.com
+    """This class allows to control any Koch robot of various number of motors.
+
+    A few versions are available:
+    - [Koch v1.0](https://github.com/AlexanderKoch-Koch/low_cost_robot), with and without the wrist-to-elbow expansion, which was developed
+    by Alexander Koch from [Tau Robotics](https://tau-robotics.com): [Github for sourcing and assembly](
+    - [Koch v1.1])https://github.com/jess-moss/koch-v1-1), which was developed by Jess Moss.
 
     Example of highest frequency teleoperation without camera:
     ```python
@@ -264,12 +227,12 @@ class KochRobot:
     Example of highest frequency data collection with cameras:
     ```python
     # Defines how to communicate with 2 cameras connected to the computer.
-    # Here, the webcam of the mackbookpro and the iphone (connected in USB to the macbookpro)
+    # Here, the webcam of the laptop and the phone (connected in USB to the laptop)
     # can be reached respectively using the camera indices 0 and 1. These indices can be
     # arbitrary. See the documentation of `OpenCVCamera` to find your own camera indices.
     cameras = {
-        "macbookpro": OpenCVCamera(camera_index=0, fps=30, width=640, height=480),
-        "iphone": OpenCVCamera(camera_index=1, fps=30, width=640, height=480),
+        "laptop": OpenCVCamera(camera_index=0, fps=30, width=640, height=480),
+        "phone": OpenCVCamera(camera_index=1, fps=30, width=640, height=480),
     }
 
     # Assumes leader and follower arms have been instantiated already (see first example)
@@ -342,16 +305,18 @@ class KochRobot:
         if self.calibration_path.exists():
             # Reset all arms before setting calibration
             for name in self.follower_arms:
-                reset_arm(self.follower_arms[name])
+                reset_torque_mode(self.follower_arms[name])
             for name in self.leader_arms:
-                reset_arm(self.leader_arms[name])
+                reset_torque_mode(self.leader_arms[name])
 
             with open(self.calibration_path, "rb") as f:
                 calibration = pickle.load(f)
         else:
+            print(f"Missing calibration file '{self.calibration_path}'. Starting calibration precedure.")
             # Run calibration process which begins by reseting all arms
             calibration = self.run_calibration()
 
+            print(f"Calibration is done! Saving calibration file '{self.calibration_path}'")
             self.calibration_path.parent.mkdir(parents=True, exist_ok=True)
             with open(self.calibration_path, "wb") as f:
                 pickle.dump(calibration, f)
@@ -362,18 +327,6 @@ class KochRobot:
         for name in self.leader_arms:
             self.leader_arms[name].set_calibration(calibration[f"leader_{name}"])
 
-        for name in self.leader_arms:
-            values = self.leader_arms[name].read("Present_Position")
-            if (values < -180).any() or (values >= 180).any():
-                raise ValueError(
-                    f"At least one of the motor of the {name} leader arm has a joint value outside of its centered degree range of ]-180, 180[."
-                    'This "jump of range" can be caused by a hardware issue, or you might have unexpectedly completed a full rotation of the motor '
-                    "during manipulation or transportation of your robot. "
-                    f"The values and motors: {values} {self.leader_arms[name].motor_names}.\n"
-                    "Rotate the arm to fit the range ]-180, 180[ and relaunch the script, or recalibrate all motors by setting a different "
-                    "`calibration_path` during the instatiation of your robot."
-                )
-
         # Set better PID values to close the gap between recored states and actions
         # TODO(rcadene): Implement an automatic procedure to set optimial PID values for each motor
         for name in self.follower_arms:
@@ -383,13 +336,14 @@ class KochRobot:
 
         # Enable torque on all motors of the follower arms
         for name in self.follower_arms:
+            print(f"Activating torque on {name} follower arm.")
             self.follower_arms[name].write("Torque_Enable", 1)
 
         # Enable torque on the gripper of the leader arms, and move it to 45 degrees,
         # so that we can use it as a trigger to close the gripper of the follower arms.
         for name in self.leader_arms:
             self.leader_arms[name].write("Torque_Enable", 1, "gripper")
-            self.leader_arms[name].write("Goal_Position", GRIPPER_OPEN, "gripper")
+            self.leader_arms[name].write("Goal_Position", GRIPPER_OPEN_DEGREE, "gripper")
 
         # Connect the cameras
         for name in self.cameras:
@@ -424,12 +378,12 @@ class KochRobot:
                 "KochRobot is not connected. You need to run `robot.connect()`."
             )
 
-        # Prepare to assign the positions of the leader to the follower
+        # Prepare to assign the position of the leader to the follower
         leader_pos = {}
         for name in self.leader_arms:
-            now = time.perf_counter()
+            before_lread_t = time.perf_counter()
             leader_pos[name] = self.leader_arms[name].read("Present_Position")
-            self.logs[f"read_leader_{name}_pos_dt_s"] = time.perf_counter() - now
+            self.logs[f"read_leader_{name}_pos_dt_s"] = time.perf_counter() - before_lread_t
 
         follower_goal_pos = {}
         for name in self.leader_arms:
@@ -437,9 +391,9 @@ class KochRobot:
 
         # Send action
         for name in self.follower_arms:
-            now = time.perf_counter()
+            before_fwrite_t = time.perf_counter()
             self.follower_arms[name].write("Goal_Position", follower_goal_pos[name])
-            self.logs[f"write_follower_{name}_goal_pos_dt_s"] = time.perf_counter() - now
+            self.logs[f"write_follower_{name}_goal_pos_dt_s"] = time.perf_counter() - before_fwrite_t
 
         # Early exit when recording data is not requested
         if not record_data:
@@ -449,9 +403,9 @@ class KochRobot:
         # Read follower position
         follower_pos = {}
         for name in self.follower_arms:
-            now = time.perf_counter()
+            before_fread_t = time.perf_counter()
             follower_pos[name] = self.follower_arms[name].read("Present_Position")
-            self.logs[f"read_follower_{name}_pos_dt_s"] = time.perf_counter() - now
+            self.logs[f"read_follower_{name}_pos_dt_s"] = time.perf_counter() - before_fread_t
 
         # Create state by concatenating follower current position
         state = []
@@ -470,10 +424,10 @@ class KochRobot:
         # Capture images from cameras
         images = {}
         for name in self.cameras:
-            now = time.perf_counter()
+            before_camread_t = time.perf_counter()
             images[name] = self.cameras[name].async_read()
             self.logs[f"read_camera_{name}_dt_s"] = self.cameras[name].logs["delta_timestamp_s"]
-            self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - now
+            self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - before_camread_t
 
         # Populate output dictionnaries and format to pytorch
         obs_dict, action_dict = {}, {}
@@ -494,9 +448,9 @@ class KochRobot:
         # Read follower position
         follower_pos = {}
         for name in self.follower_arms:
-            now = time.perf_counter()
+            before_fread_t = time.perf_counter()
             follower_pos[name] = self.follower_arms[name].read("Present_Position")
-            self.logs[f"read_follower_{name}_pos_dt_s"] = time.perf_counter() - now
+            self.logs[f"read_follower_{name}_pos_dt_s"] = time.perf_counter() - before_fread_t
 
         # Create state by concatenating follower current position
         state = []
@@ -508,10 +462,10 @@ class KochRobot:
         # Capture images from cameras
         images = {}
         for name in self.cameras:
-            now = time.perf_counter()
+            before_camread_t = time.perf_counter()
             images[name] = self.cameras[name].async_read()
             self.logs[f"read_camera_{name}_dt_s"] = self.cameras[name].logs["delta_timestamp_s"]
-            self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - now
+            self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - before_camread_t
 
         # Populate output dictionnaries and format to pytorch
         obs_dict = {}

lerobot/common/utils/utils.py

@@ -158,6 +158,7 @@ def init_hydra_config(config_path: str, overrides: list[str] | None = None) -> D
         version_base="1.2",
     )
     cfg = hydra.compose(Path(config_path).stem, overrides)
+
     return cfg
 
 

lerobot/configs/default.yaml

@@ -107,13 +107,13 @@ training:
       min_max: [0.8, 1.2]
     saturation:
       weight: 1
-      min_max: [0.7, 1.3]
+      min_max: [0.5, 1.5]
     hue:
       weight: 1
-      min_max: [-0.02, 0.02]
+      min_max: [-0.05, 0.05]
     sharpness:
       weight: 1
-      min_max: [0.1, 1.9]
+      min_max: [0.8, 1.2]
 
 eval:
   n_episodes: 1

lerobot/configs/policy/act_koch_real_bigger.yaml

@@ -1,102 +0,0 @@
-# @package _global_
-
-# Use `act_koch_real.yaml` to train on real-world datasets collected on Alexander Koch's robots.
-# Compared to `act.yaml`, it contains 2 cameras (i.e. laptop, phone) instead of 1 camera (i.e. top).
-# Also, `training.eval_freq` is set to -1. This config is used to evaluate checkpoints at a certain frequency of training steps.
-# When it is set to -1, it deactivates evaluation. This is because real-world evaluation is done through our `control_robot.py` script.
-# Look at the documentation in header of `control_robot.py` for more information on how to collect data , train and evaluate a policy.
-#
-# Example of usage for training:
-# ```bash
-# python lerobot/scripts/train.py \
-#   policy=act_koch_real \
-#   env=koch_real
-# ```
-
-seed: 1000
-dataset_repo_id: lerobot/koch_pick_place_lego
-
-override_dataset_stats:
-  observation.images.laptop:
-    # stats from imagenet, since we use a pretrained vision model
-    mean: [[[0.485]], [[0.456]], [[0.406]]]  # (c,1,1)
-    std: [[[0.229]], [[0.224]], [[0.225]]]  # (c,1,1)
-  observation.images.phone:
-    # stats from imagenet, since we use a pretrained vision model
-    mean: [[[0.485]], [[0.456]], [[0.406]]]  # (c,1,1)
-    std: [[[0.229]], [[0.224]], [[0.225]]]  # (c,1,1)
-
-training:
-  offline_steps: 160000
-  online_steps: 0
-  eval_freq: -1
-  save_freq: 5000
-  log_freq: 100
-  save_checkpoint: true
-
-  batch_size: 16
-  lr: 1e-4
-  lr_backbone: 1e-4
-  weight_decay: 1e-4
-  grad_clip_norm: 20
-  online_steps_between_rollouts: 1
-
-  delta_timestamps:
-    action: "[i / ${fps} + 1 / ${fps} for i in range(${policy.chunk_size})]"
-
-eval:
-  n_episodes: 50
-  batch_size: 50
-
-# See `configuration_act.py` for more details.
-policy:
-  name: act
-
-  # Input / output structure.
-  n_obs_steps: 1
-  chunk_size: 100
-  n_action_steps: 100
-
-  input_shapes:
-    # TODO(rcadene, alexander-soare): add variables for height and width from the dataset/env?
-    observation.images.laptop: [3, 480, 640]
-    observation.images.phone: [3, 480, 640]
-    observation.state: ["${env.state_dim}"]
-  output_shapes:
-    action: ["${env.action_dim}"]
-
-  # Normalization / Unnormalization
-  input_normalization_modes:
-    observation.images.laptop: mean_std
-    observation.images.phone: mean_std
-    observation.state: mean_std
-  output_normalization_modes:
-    action: mean_std
-
-  # Architecture.
-  # Vision backbone.
-  vision_backbone: resnet18
-  pretrained_backbone_weights: ResNet18_Weights.IMAGENET1K_V1
-  replace_final_stride_with_dilation: false
-  # Transformer layers.
-  pre_norm: false
-  dim_model: 512
-  n_heads: 8
-  dim_feedforward: 3200
-  feedforward_activation: relu
-  n_encoder_layers: 4
-  # Note: Although the original ACT implementation has 7 for `n_decoder_layers`, there is a bug in the code
-  # that means only the first layer is used. Here we match the original implementation by setting this to 1.
-  # See this issue https://github.com/tonyzhaozh/act/issues/25#issue-2258740521.
-  n_decoder_layers: 4
-  # VAE.
-  use_vae: true
-  latent_dim: 32
-  n_vae_encoder_layers: 4
-
-  # Inference.
-  temporal_ensemble_momentum: null
-
-  # Training and loss computation.
-  dropout: 0.1
-  kl_weight: 10.0

lerobot/configs/robot/koch.yaml

@@ -1,4 +1,5 @@
 _target_: lerobot.common.robot_devices.robots.koch.KochRobot
+calibration_path: .cache/calibration/koch.pkl
 leader_arms:
   main:
     _target_: lerobot.common.robot_devices.motors.dynamixel.DynamixelMotorsBus

lerobot/scripts/control_robot.py

@@ -1,12 +1,22 @@
 """
+Utilities to control a robot.
+
+Useful to record a dataset, replay a recorded episode, run the policy on your robot
+and record an evaluation dataset, and to recalibrate your robot if needed.
+
 Examples of usage:
 
+- Recalibrate your robot:
+```bash
+python lerobot/scripts/control_robot.py calibrate
+```
+
 - Unlimited teleoperation at highest frequency (~200 Hz is expected), to exit with CTRL+C:
 ```bash
 python lerobot/scripts/control_robot.py teleoperate
 
 # Remove the cameras from the robot definition. They are not used in 'teleoperate' anyway.
-python lerobot/scripts/control_robot.py teleoperate '~cameras'
+python lerobot/scripts/control_robot.py teleoperate --robot-overrides '~cameras'
 ```
 
 - Unlimited teleoperation at a limited frequency of 30 Hz, to simulate data recording frequency:
@@ -96,6 +106,7 @@ import os
 import platform
 import shutil
 import time
+import traceback
 from contextlib import nullcontext
 from functools import cache
 from pathlib import Path
@@ -103,7 +114,6 @@ from pathlib import Path
 import cv2
 import torch
 import tqdm
-from huggingface_hub import create_branch
 from omegaconf import DictConfig
 from PIL import Image
 from termcolor import colored
@@ -113,20 +123,45 @@ from lerobot.common.datasets.compute_stats import compute_stats
 from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION, LeRobotDataset
 from lerobot.common.datasets.push_dataset_to_hub.aloha_hdf5_format import to_hf_dataset
 from lerobot.common.datasets.push_dataset_to_hub.utils import concatenate_episodes, get_default_encoding
-from lerobot.common.datasets.utils import calculate_episode_data_index
+from lerobot.common.datasets.utils import calculate_episode_data_index, create_branch
 from lerobot.common.datasets.video_utils import encode_video_frames
 from lerobot.common.policies.factory import make_policy
 from lerobot.common.robot_devices.robots.factory import make_robot
 from lerobot.common.robot_devices.robots.utils import Robot
 from lerobot.common.utils.utils import get_safe_torch_device, init_hydra_config, init_logging, set_global_seed
 from lerobot.scripts.eval import get_pretrained_policy_path
-from lerobot.scripts.push_dataset_to_hub import push_meta_data_to_hub, push_videos_to_hub, save_meta_data
+from lerobot.scripts.push_dataset_to_hub import (
+    push_dataset_card_to_hub,
+    push_meta_data_to_hub,
+    push_videos_to_hub,
+    save_meta_data,
+)
 
 ########################################################################################
 # Utilities
 ########################################################################################
 
 
+def say(text, blocking=False):
+    # Check if mac, linux, or windows.
+    if platform.system() == "Darwin":
+        cmd = f'say "{text}"'
+    elif platform.system() == "Linux":
+        cmd = f'spd-say "{text}"'
+    elif platform.system() == "Windows":
+        cmd = (
+            'PowerShell -Command "Add-Type -AssemblyName System.Speech; '
+            f"(New-Object System.Speech.Synthesis.SpeechSynthesizer).Speak('{text}')\""
+        )
+
+    if not blocking and platform.system() in ["Darwin", "Linux"]:
+        # TODO(rcadene): Make it work for Windows
+        # Use the ampersand to run command in the background
+        cmd += " &"
+
+    os.system(cmd)
+
+
 def save_image(img_tensor, key, frame_index, episode_index, videos_dir):
     img = Image.fromarray(img_tensor.numpy())
     path = videos_dir / f"{key}_episode_{episode_index:06d}" / f"frame_{frame_index:06d}.png"
@@ -171,11 +206,11 @@ def log_control_info(robot, dt_s, episode_index=None, frame_index=None, fps=None
     for name in robot.follower_arms:
         key = f"write_follower_{name}_goal_pos_dt_s"
         if key in robot.logs:
-            log_dt("dtRfoll", robot.logs[key])
+            log_dt("dtWfoll", robot.logs[key])
 
         key = f"read_follower_{name}_pos_dt_s"
         if key in robot.logs:
-            log_dt("dtWfoll", robot.logs[key])
+            log_dt("dtRfoll", robot.logs[key])
 
     for name in robot.cameras:
         key = f"read_camera_{name}_dt_s"
@@ -192,11 +227,21 @@ def log_control_info(robot, dt_s, episode_index=None, frame_index=None, fps=None
 
 @cache
 def is_headless():
-    if platform.system() == "Linux":
-        display = os.environ.get("DISPLAY")
-        if display is None or display == "":
-            return True
-    return False
+    """Detects if python is running without a monitor."""
+    try:
+        import pynput  # noqa
+
+        return False
+    except Exception:
+        print(
+            "Error trying to import pynput. Switching to headless mode. "
+            "As a result, the video stream from the cameras won't be shown, "
+            "and you won't be able to change the control flow with keyboards. "
+            "For more info, see traceback below.\n"
+        )
+        traceback.print_exc()
+        print()
+        return True
 
 
 ########################################################################################
@@ -204,24 +249,37 @@ def is_headless():
 ########################################################################################
 
 
+def calibrate(robot: Robot):
+    if robot.calibration_path.exists():
+        print(f"Removing '{robot.calibration_path}'")
+        robot.calibration_path.unlink()
+
+    if robot.is_connected:
+        robot.disconnect()
+
+    # Calling `connect` automatically runs calibration
+    # when the calibration file is missing
+    robot.connect()
+
+
 def teleoperate(robot: Robot, fps: int | None = None, teleop_time_s: float | None = None):
     # TODO(rcadene): Add option to record logs
     if not robot.is_connected:
         robot.connect()
 
-    start_time = time.perf_counter()
+    start_teleop_t = time.perf_counter()
     while True:
-        now = time.perf_counter()
+        start_loop_t = time.perf_counter()
         robot.teleop_step()
 
         if fps is not None:
-            dt_s = time.perf_counter() - now
+            dt_s = time.perf_counter() - start_loop_t
             busy_wait(1 / fps - dt_s)
 
-        dt_s = time.perf_counter() - now
+        dt_s = time.perf_counter() - start_loop_t
         log_control_info(robot, dt_s, fps=fps)
 
-        if teleop_time_s is not None and time.perf_counter() - start_time > teleop_time_s:
+        if teleop_time_s is not None and time.perf_counter() - start_teleop_t > teleop_time_s:
             break
 
 
@@ -239,10 +297,18 @@ def record(
     video=True,
     run_compute_stats=True,
     push_to_hub=True,
+    tags=None,
     num_image_writers=8,
     force_override=False,
 ):
     # TODO(rcadene): Add option to record logs
+    # TODO(rcadene): Clean this function via decomposition in higher level functions
+
+    _, dataset_name = repo_id.split("/")
+    if dataset_name.startswith("eval_") and policy is None:
+        raise ValueError(
+            f"Your dataset name begins by 'eval_' ({dataset_name}) but no policy is provided ({policy})."
+        )
 
     if not video:
         raise NotImplementedError()
@@ -271,7 +337,7 @@ def record(
 
     if is_headless():
         logging.info(
-            "Headless environment detected. Display cameras on screen and keyboard inputs will not be available."
+            "Headless environment detected. On-screen cameras display and keyboard inputs will not be available."
         )
 
     # Allow to exit early while recording an episode or resetting the environment,
@@ -323,15 +389,15 @@ def record(
     # Execute a few seconds without recording data, to give times
     # to the robot devices to connect and start synchronizing.
     timestamp = 0
-    start_time = time.perf_counter()
+    start_warmup_t = time.perf_counter()
     is_warmup_print = False
     while timestamp < warmup_time_s:
         if not is_warmup_print:
             logging.info("Warming up (no data recording)")
-            os.system('say "Warmup" &')
+            say("Warming up")
             is_warmup_print = True
 
-        now = time.perf_counter()
+        start_loop_t = time.perf_counter()
 
         if policy is None:
             observation, action = robot.teleop_step(record_data=True)
@@ -344,13 +410,13 @@ def record(
                 cv2.imshow(key, cv2.cvtColor(observation[key].numpy(), cv2.COLOR_RGB2BGR))
             cv2.waitKey(1)
 
-        dt_s = time.perf_counter() - now
+        dt_s = time.perf_counter() - start_loop_t
         busy_wait(1 / fps - dt_s)
 
-        dt_s = time.perf_counter() - now
+        dt_s = time.perf_counter() - start_loop_t
         log_control_info(robot, dt_s, fps=fps)
 
-        timestamp = time.perf_counter() - start_time
+        timestamp = time.perf_counter() - start_warmup_t
 
     # Save images using threads to reach high fps (30 and more)
     # Using `with` to exist smoothly if an execption is raised.
@@ -360,13 +426,13 @@ def record(
         # Start recording all episodes
         while episode_index < num_episodes:
             logging.info(f"Recording episode {episode_index}")
-            os.system(f'say "Recording episode {episode_index}" &')
+            say(f"Recording episode {episode_index}")
             ep_dict = {}
             frame_index = 0
             timestamp = 0
-            start_time = time.perf_counter()
+            start_episode_t = time.perf_counter()
             while timestamp < episode_time_s:
-                now = time.perf_counter()
+                start_loop_t = time.perf_counter()
 
                 if policy is None:
                     observation, action = robot.teleop_step(record_data=True)
@@ -407,17 +473,19 @@ def record(
                                 observation[name] = observation[name].type(torch.float32) / 255
                                 observation[name] = observation[name].permute(2, 0, 1).contiguous()
                             observation[name] = observation[name].unsqueeze(0)
+                            observation[name] = observation[name].to(device)
 
-                        if device.type == "mps":
-                            for name in observation:
-                                observation[name] = observation[name].to(device)
-
+                        # Compute the next action with the policy
+                        # based on the current observation
                         action = policy.select_action(observation)
 
-                        # remove batch dimension
+                        # Remove batch dimension
                         action = action.squeeze(0)
+
+                        # Move to cpu, if not already the case
                         action = action.to("cpu")
 
+                    # Order the robot to move
                     robot.send_action(action)
                     action = {"action": action}
 
@@ -428,14 +496,13 @@ def record(
 
                 frame_index += 1
 
-                dt_s = time.perf_counter() - now
+                dt_s = time.perf_counter() - start_loop_t
                 busy_wait(1 / fps - dt_s)
 
-                dt_s = time.perf_counter() - now
+                dt_s = time.perf_counter() - start_loop_t
                 log_control_info(robot, dt_s, fps=fps)
 
-                timestamp = time.perf_counter() - start_time
-
+                timestamp = time.perf_counter() - start_episode_t
                 if exit_early:
                     exit_early = False
                     break
@@ -443,10 +510,10 @@ def record(
             if not stop_recording:
                 # Start resetting env while the executor are finishing
                 logging.info("Reset the environment")
-                os.system('say "Reset the environment" &')
+                say("Reset the environment")
 
             timestamp = 0
-            start_time = time.perf_counter()
+            start_vencod_t = time.perf_counter()
 
             # During env reset we save the data and encode the videos
             num_frames = frame_index
@@ -492,7 +559,7 @@ def record(
             with tqdm.tqdm(total=reset_time_s, desc="Waiting") as pbar:
                 while timestamp < reset_time_s and not is_last_episode:
                     time.sleep(1)
-                    timestamp = time.perf_counter() - start_time
+                    timestamp = time.perf_counter() - start_vencod_t
                     pbar.update(1)
                     if exit_early:
                         exit_early = False
@@ -507,7 +574,7 @@ def record(
 
             if is_last_episode:
                 logging.info("Done recording")
-                os.system('say "Done recording"')
+                say("Done recording", blocking=True)
                 if not is_headless():
                     listener.stop()
 
@@ -525,7 +592,7 @@ def record(
     num_episodes = episode_index
 
     logging.info("Encoding videos")
-    os.system('say "Encoding videos" &')
+    say("Encoding videos")
     # Use ffmpeg to convert frames stored as png into mp4 videos
     for episode_index in tqdm.tqdm(range(num_episodes)):
         for key in image_keys:
@@ -533,6 +600,7 @@ def record(
             fname = f"{key}_episode_{episode_index:06d}.mp4"
             video_path = local_dir / "videos" / fname
             if video_path.exists():
+                # Skip if video is already encoded. Could be the case when resuming data recording.
                 continue
             # note: `encode_video_frames` is a blocking call. Making it asynchronous shouldn't speedup encoding,
             # since video encoding with ffmpeg is already using multithreading.
@@ -569,12 +637,12 @@ def record(
     )
     if run_compute_stats:
         logging.info("Computing dataset statistics")
-        os.system('say "Computing dataset statistics" &')
+        say("Computing dataset statistics")
         stats = compute_stats(lerobot_dataset)
         lerobot_dataset.stats = stats
     else:
         stats = {}
-        logging.info("Skipping computation of the dataset statistrics")
+        logging.info("Skipping computation of the dataset statistics")
 
     hf_dataset = hf_dataset.with_format(None)  # to remove transforms that cant be saved
     hf_dataset.save_to_disk(str(local_dir / "train"))
@@ -585,13 +653,13 @@ def record(
     if push_to_hub:
         hf_dataset.push_to_hub(repo_id, revision="main")
         push_meta_data_to_hub(repo_id, meta_data_dir, revision="main")
+        push_dataset_card_to_hub(repo_id, revision="main", tags=tags)
         if video:
             push_videos_to_hub(repo_id, videos_dir, revision="main")
         create_branch(repo_id, repo_type="dataset", branch=CODEBASE_VERSION)
 
     logging.info("Exiting")
-    os.system('say "Exiting" &')
-
+    say("Exiting")
     return lerobot_dataset
 
 
@@ -610,18 +678,17 @@ def replay(robot: Robot, episode: int, fps: int | None = None, root="data", repo
         robot.connect()
 
     logging.info("Replaying episode")
-    os.system('say "Replaying episode"')
-
+    say("Replaying episode", blocking=True)
     for idx in range(from_idx, to_idx):
-        now = time.perf_counter()
+        start_episode_t = time.perf_counter()
 
         action = items[idx]["action"]
         robot.send_action(action)
 
-        dt_s = time.perf_counter() - now
+        dt_s = time.perf_counter() - start_episode_t
         busy_wait(1 / fps - dt_s)
 
-        dt_s = time.perf_counter() - now
+        dt_s = time.perf_counter() - start_episode_t
         log_control_info(robot, dt_s, fps=fps)
 
 
@@ -644,6 +711,8 @@ if __name__ == "__main__":
         help="Any key=value arguments to override config values (use dots for.nested=overrides)",
     )
 
+    parser_calib = subparsers.add_parser("calibrate", parents=[base_parser])
+
     parser_teleop = subparsers.add_parser("teleoperate", parents=[base_parser])
     parser_teleop.add_argument(
         "--fps", type=none_or_int, default=None, help="Frames per second (set to None to disable)"
@@ -668,19 +737,19 @@ if __name__ == "__main__":
     parser_record.add_argument(
         "--warmup-time-s",
         type=int,
-        default=2,
+        default=10,
         help="Number of seconds before starting data collection. It allows the robot devices to warmup and synchronize.",
     )
     parser_record.add_argument(
         "--episode-time-s",
         type=int,
-        default=10,
+        default=60,
         help="Number of seconds for data recording for each episode.",
     )
     parser_record.add_argument(
         "--reset-time-s",
         type=int,
-        default=5,
+        default=60,
         help="Number of seconds for resetting the environment after each episode.",
     )
     parser_record.add_argument("--num-episodes", type=int, default=50, help="Number of episodes to record.")
@@ -696,6 +765,12 @@ if __name__ == "__main__":
         default=1,
         help="Upload dataset to Hugging Face hub.",
     )
+    parser_record.add_argument(
+        "--tags",
+        type=str,
+        nargs="*",
+        help="Add tags to your dataset on the hub.",
+    )
     parser_record.add_argument(
         "--num-image-writers",
         type=int,
@@ -757,7 +832,10 @@ if __name__ == "__main__":
     robot_cfg = init_hydra_config(robot_path, robot_overrides)
     robot = make_robot(robot_cfg)
 
-    if control_mode == "teleoperate":
+    if control_mode == "calibrate":
+        calibrate(robot, **kwargs)
+
+    elif control_mode == "teleoperate":
         teleoperate(robot, **kwargs)
 
     elif control_mode == "record":
@@ -777,3 +855,8 @@ if __name__ == "__main__":
 
     elif control_mode == "replay":
         replay(robot, **kwargs)
+
+    if robot.is_connected:
+        # Disconnect manually to avoid a "Core dump" during process
+        # termination due to camera threads not properly exiting.
+        robot.disconnect()

lerobot/scripts/push_dataset_to_hub.py

@@ -56,7 +56,7 @@ from safetensors.torch import save_file
 from lerobot.common.datasets.compute_stats import compute_stats
 from lerobot.common.datasets.lerobot_dataset import CODEBASE_VERSION, LeRobotDataset
 from lerobot.common.datasets.push_dataset_to_hub.utils import check_repo_id
-from lerobot.common.datasets.utils import flatten_dict
+from lerobot.common.datasets.utils import create_branch, create_lerobot_dataset_card, flatten_dict
 
 
 def get_from_raw_to_lerobot_format_fn(raw_format: str):
@@ -114,6 +114,14 @@ def push_meta_data_to_hub(repo_id: str, meta_data_dir: str | Path, revision: str
     )
 
 
+def push_dataset_card_to_hub(
+    repo_id: str, revision: str | None, tags: list | None = None, text: str | None = None
+):
+    """Creates and pushes a LeRobotDataset Card with appropriate tags to easily find it on the hub."""
+    card = create_lerobot_dataset_card(tags=tags, text=text)
+    card.push_to_hub(repo_id=repo_id, repo_type="dataset", revision=revision)
+
+
 def push_videos_to_hub(repo_id: str, videos_dir: str | Path, revision: str | None):
     """Expect mp4 files to be all stored in a single "videos" directory.
     On the hugging face repositery, they will be uploaded in a "videos" directory at the root.
@@ -213,10 +221,10 @@ def push_dataset_to_hub(
     if push_to_hub:
         hf_dataset.push_to_hub(repo_id, revision="main")
         push_meta_data_to_hub(repo_id, meta_data_dir, revision="main")
+        push_dataset_card_to_hub(repo_id, revision="main")
         if video:
             push_videos_to_hub(repo_id, videos_dir, revision="main")
-        api = HfApi()
-        api.create_branch(repo_id, repo_type="dataset", branch=CODEBASE_VERSION)
+        create_branch(repo_id, repo_type="dataset", branch=CODEBASE_VERSION)
 
     if tests_data_dir:
         # get the first episode

lerobot/scripts/visualize_dataset.py

@@ -108,8 +108,8 @@ def visualize_dataset(
     web_port: int = 9090,
     ws_port: int = 9087,
     save: bool = False,
-    output_dir: Path | None = None,
     root: Path | None = None,
+    output_dir: Path | None = None,
 ) -> Path | None:
     if save:
         assert (
@@ -209,6 +209,18 @@ def main():
         required=True,
         help="Episode to visualize.",
     )
+    parser.add_argument(
+        "--root",
+        type=Path,
+        default=None,
+        help="Root directory for a dataset stored locally (e.g. `--root data`). By default, the dataset will be loaded from hugging face cache folder, or downloaded from the hub if available.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=Path,
+        default=None,
+        help="Directory path to write a .rrd file when `--save 1` is set.",
+    )
     parser.add_argument(
         "--batch-size",
         type=int,
@@ -254,17 +266,6 @@ def main():
             "Visualize the data by running `rerun path/to/file.rrd` on your local machine."
         ),
     )
-    parser.add_argument(
-        "--output-dir",
-        type=str,
-        help="Directory path to write a .rrd file when `--save 1` is set.",
-    )
-
-    parser.add_argument(
-        "--root",
-        type=str,
-        help="Root directory for a dataset stored on a local machine.",
-    )
 
     args = parser.parse_args()
     visualize_dataset(**vars(args))

lerobot/scripts/visualize_dataset_html.py

@@ -0,0 +1,300 @@
+#!/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.
+""" Visualize data of **all** frames of any episode of a dataset of type LeRobotDataset.
+
+Note: The last frame of the episode doesnt always correspond to a final state.
+That's because our datasets are composed of transition from state to state up to
+the antepenultimate state associated to the ultimate action to arrive in the final state.
+However, there might not be a transition from a final state to another state.
+
+Note: This script aims to visualize the data used to train the neural networks.
+~What you see is what you get~. When visualizing image modality, it is often expected to observe
+lossly compression artifacts since these images have been decoded from compressed mp4 videos to
+save disk space. The compression factor applied has been tuned to not affect success rate.
+
+Example of usage:
+
+- Visualize data stored on a local machine:
+```bash
+local$ python lerobot/scripts/visualize_dataset_html.py \
+    --repo-id lerobot/pusht
+
+local$ open http://localhost:9090
+```
+
+- Visualize data stored on a distant machine with a local viewer:
+```bash
+distant$ python lerobot/scripts/visualize_dataset_html.py \
+    --repo-id lerobot/pusht
+
+local$ ssh -L 9090:localhost:9090 distant  # create a ssh tunnel
+local$ open http://localhost:9090
+```
+
+- Select episodes to visualize:
+```bash
+python lerobot/scripts/visualize_dataset_html.py \
+    --repo-id lerobot/pusht \
+    --episodes 7 3 5 1 4
+```
+"""
+
+import argparse
+import logging
+import shutil
+from pathlib import Path
+
+import torch
+import tqdm
+from flask import Flask, redirect, render_template, url_for
+
+from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
+from lerobot.common.utils.utils import init_logging
+
+
+class EpisodeSampler(torch.utils.data.Sampler):
+    def __init__(self, dataset, episode_index):
+        from_idx = dataset.episode_data_index["from"][episode_index].item()
+        to_idx = dataset.episode_data_index["to"][episode_index].item()
+        self.frame_ids = range(from_idx, to_idx)
+
+    def __iter__(self):
+        return iter(self.frame_ids)
+
+    def __len__(self):
+        return len(self.frame_ids)
+
+
+def run_server(
+    dataset: LeRobotDataset,
+    episodes: list[int],
+    host: str,
+    port: str,
+    static_folder: Path,
+    template_folder: Path,
+):
+    app = Flask(__name__, static_folder=static_folder.resolve(), template_folder=template_folder.resolve())
+    app.config["SEND_FILE_MAX_AGE_DEFAULT"] = 0  # specifying not to cache
+
+    @app.route("/")
+    def index():
+        # home page redirects to the first episode page
+        [dataset_namespace, dataset_name] = dataset.repo_id.split("/")
+        first_episode_id = episodes[0]
+        return redirect(
+            url_for(
+                "show_episode",
+                dataset_namespace=dataset_namespace,
+                dataset_name=dataset_name,
+                episode_id=first_episode_id,
+            )
+        )
+
+    @app.route("/<string:dataset_namespace>/<string:dataset_name>/episode_<int:episode_id>")
+    def show_episode(dataset_namespace, dataset_name, episode_id):
+        dataset_info = {
+            "repo_id": dataset.repo_id,
+            "num_samples": dataset.num_samples,
+            "num_episodes": dataset.num_episodes,
+            "fps": dataset.fps,
+        }
+        video_paths = get_episode_video_paths(dataset, episode_id)
+        videos_info = [
+            {"url": url_for("static", filename=video_path), "filename": Path(video_path).name}
+            for video_path in video_paths
+        ]
+        ep_csv_url = url_for("static", filename=get_ep_csv_fname(episode_id))
+        return render_template(
+            "visualize_dataset_template.html",
+            episode_id=episode_id,
+            episodes=episodes,
+            dataset_info=dataset_info,
+            videos_info=videos_info,
+            ep_csv_url=ep_csv_url,
+            has_policy=False,
+        )
+
+    app.run(host=host, port=port)
+
+
+def get_ep_csv_fname(episode_id: int):
+    ep_csv_fname = f"episode_{episode_id}.csv"
+    return ep_csv_fname
+
+
+def write_episode_data_csv(output_dir, file_name, episode_index, dataset):
+    """Write a csv file containg timeseries data of an episode (e.g. state and action).
+    This file will be loaded by Dygraph javascript to plot data in real time."""
+    from_idx = dataset.episode_data_index["from"][episode_index]
+    to_idx = dataset.episode_data_index["to"][episode_index]
+
+    has_state = "observation.state" in dataset.hf_dataset.features
+    has_action = "action" in dataset.hf_dataset.features
+
+    # init header of csv with state and action names
+    header = ["timestamp"]
+    if has_state:
+        dim_state = len(dataset.hf_dataset["observation.state"][0])
+        header += [f"state_{i}" for i in range(dim_state)]
+    if has_action:
+        dim_action = len(dataset.hf_dataset["action"][0])
+        header += [f"action_{i}" for i in range(dim_action)]
+
+    columns = ["timestamp"]
+    if has_state:
+        columns += ["observation.state"]
+    if has_action:
+        columns += ["action"]
+
+    rows = []
+    data = dataset.hf_dataset.select_columns(columns)
+    for i in range(from_idx, to_idx):
+        row = [data[i]["timestamp"].item()]
+        if has_state:
+            row += data[i]["observation.state"].tolist()
+        if has_action:
+            row += data[i]["action"].tolist()
+        rows.append(row)
+
+    output_dir.mkdir(parents=True, exist_ok=True)
+    with open(output_dir / file_name, "w") as f:
+        f.write(",".join(header) + "\n")
+        for row in rows:
+            row_str = [str(col) for col in row]
+            f.write(",".join(row_str) + "\n")
+
+
+def get_episode_video_paths(dataset: LeRobotDataset, ep_index: int) -> list[str]:
+    # get first frame of episode (hack to get video_path of the episode)
+    first_frame_idx = dataset.episode_data_index["from"][ep_index].item()
+    return [
+        dataset.hf_dataset.select_columns(key)[first_frame_idx][key]["path"]
+        for key in dataset.video_frame_keys
+    ]
+
+
+def visualize_dataset_html(
+    repo_id: str,
+    root: Path | None = None,
+    episodes: list[int] = None,
+    output_dir: Path | None = None,
+    serve: bool = True,
+    host: str = "127.0.0.1",
+    port: int = 9090,
+    force_override: bool = False,
+) -> Path | None:
+    init_logging()
+
+    dataset = LeRobotDataset(repo_id, root=root)
+
+    if not dataset.video:
+        raise NotImplementedError(f"Image datasets ({dataset.video=}) are currently not supported.")
+
+    if output_dir is None:
+        output_dir = f"outputs/visualize_dataset_html/{repo_id}"
+
+    output_dir = Path(output_dir)
+    if output_dir.exists():
+        if force_override:
+            shutil.rmtree(output_dir)
+        else:
+            logging.info(f"Output directory already exists. Loading from it: '{output_dir}'")
+
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    # Create a simlink from the dataset video folder containg mp4 files to the output directory
+    # so that the http server can get access to the mp4 files.
+    static_dir = output_dir / "static"
+    static_dir.mkdir(parents=True, exist_ok=True)
+    ln_videos_dir = static_dir / "videos"
+    if not ln_videos_dir.exists():
+        ln_videos_dir.symlink_to(dataset.videos_dir.resolve())
+
+    template_dir = Path(__file__).resolve().parent.parent / "templates"
+
+    if episodes is None:
+        episodes = list(range(dataset.num_episodes))
+
+    logging.info("Writing CSV files")
+    for episode_index in tqdm.tqdm(episodes):
+        # write states and actions in a csv (it can be slow for big datasets)
+        ep_csv_fname = get_ep_csv_fname(episode_index)
+        # TODO(rcadene): speedup script by loading directly from dataset, pyarrow, parquet, safetensors?
+        write_episode_data_csv(static_dir, ep_csv_fname, episode_index, dataset)
+
+    if serve:
+        run_server(dataset, episodes, host, port, static_dir, template_dir)
+
+
+def main():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--repo-id",
+        type=str,
+        required=True,
+        help="Name of hugging face repositery containing a LeRobotDataset dataset (e.g. `lerobot/pusht` for https://huggingface.co/datasets/lerobot/pusht).",
+    )
+    parser.add_argument(
+        "--root",
+        type=Path,
+        default=None,
+        help="Root directory for a dataset stored locally (e.g. `--root data`). By default, the dataset will be loaded from hugging face cache folder, or downloaded from the hub if available.",
+    )
+    parser.add_argument(
+        "--episodes",
+        type=int,
+        nargs="*",
+        default=None,
+        help="Episode indices to visualize (e.g. `0 1 5 6` to load episodes of index 0, 1, 5 and 6). By default loads all episodes.",
+    )
+    parser.add_argument(
+        "--output-dir",
+        type=Path,
+        default=None,
+        help="Directory path to write html files and kickoff a web server. By default write them to 'outputs/visualize_dataset/REPO_ID'.",
+    )
+    parser.add_argument(
+        "--serve",
+        type=int,
+        default=1,
+        help="Launch web server.",
+    )
+    parser.add_argument(
+        "--host",
+        type=str,
+        default="127.0.0.1",
+        help="Web host used by the http server.",
+    )
+    parser.add_argument(
+        "--port",
+        type=int,
+        default=9090,
+        help="Web port used by the http server.",
+    )
+    parser.add_argument(
+        "--force-override",
+        type=int,
+        default=0,
+        help="Delete the output directory if it exists already.",
+    )
+
+    args = parser.parse_args()
+    visualize_dataset_html(**vars(args))
+
+
+if __name__ == "__main__":
+    main()

lerobot/templates/visualize_dataset_template.html

@@ -0,0 +1,360 @@
+<!DOCTYPE html>
+<html lang="en">
+
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <!-- # TODO(rcadene, mishig25): store the js files locally -->
+    <script src="https://cdnjs.cloudflare.com/ajax/libs/alpinejs/3.13.5/cdn.min.js" defer></script>
+    <script src="https://cdn.jsdelivr.net/npm/dygraphs@2.2.1/dist/dygraph.min.js" type="text/javascript"></script>
+    <script src="https://cdn.tailwindcss.com"></script>
+    <title>{{ dataset_info.repo_id }} episode {{ episode_id }}</title>
+</head>
+
+<!-- Use [Alpin.js](https://alpinejs.dev), a lightweight and easy to learn JS framework -->
+<!-- Use [tailwindcss](https://tailwindcss.com/), CSS classes for styling html -->
+<!-- Use [dygraphs](https://dygraphs.com/), a lightweight JS charting library -->
+<body class="flex h-screen max-h-screen bg-slate-950 text-gray-200" x-data="createAlpineData()" @keydown.window="(e) => {
+    // Use the space bar to play and pause, instead of default action (e.g. scrolling)
+    const { keyCode, key } = e;
+    if (keyCode === 32 || key === ' ') {
+        e.preventDefault();
+        $refs.btnPause.classList.contains('hidden') ? $refs.btnPlay.click() : $refs.btnPause.click();
+    }else if (key === 'ArrowDown' || key === 'ArrowUp'){
+        const nextEpisodeId = key === 'ArrowDown' ? {{ episode_id }} + 1 : {{ episode_id }} - 1;
+        const lowestEpisodeId = {{ episodes }}.at(0);
+        const highestEpisodeId = {{ episodes }}.at(-1);
+        if(nextEpisodeId >= lowestEpisodeId && nextEpisodeId <= highestEpisodeId){
+            window.location.href = `./episode_${nextEpisodeId}`;
+        }
+    }
+}">
+    <!-- Sidebar -->
+    <div x-ref="sidebar" class="w-60 bg-slate-900 p-5 break-words max-h-screen overflow-y-auto">
+        <h1 class="mb-4 text-xl font-semibold">{{ dataset_info.repo_id }}</h1>
+
+        <ul>
+            <li>
+                Number of samples/frames: {{ dataset_info.num_samples }}
+            </li>
+            <li>
+                Number of episodes: {{ dataset_info.num_episodes }}
+            </li>
+            <li>
+                Frames per second: {{ dataset_info.fps }}
+            </li>
+        </ul>
+
+        <p>Episodes:</p>
+        <ul class="ml-2">
+            {% for episode in episodes %}
+            <li class="font-mono text-sm mt-0.5">
+                <a href="episode_{{ episode }}" class="underline {% if episode_id == episode %}font-bold -ml-1{% endif %}">
+                    Episode {{ episode }}
+                </a>
+            </li>
+            {% endfor %}
+        </ul>
+
+    </div>
+
+    <!-- Toggle sidebar button -->
+    <button class="flex items-center opacity-50 hover:opacity-100 mx-1"
+        @click="() => ($refs.sidebar.classList.toggle('hidden'))" title="Toggle sidebar">
+        <div class="bg-slate-500 w-2 h-10 rounded-full"></div>
+    </button>
+
+    <!-- Content -->
+    <div class="flex-1 max-h-screen flex flex-col gap-4 overflow-y-auto">
+        <h1 class="text-xl font-bold mt-4 font-mono">
+            Episode {{ episode_id }}
+        </h1>
+
+        <!-- Videos -->
+        <div class="flex flex-wrap gap-1">
+            {% for video_info in videos_info %}
+            <div class="max-w-96">
+                <p class="text-sm text-gray-300 bg-gray-800 px-2 rounded-t-xl truncate">{{ video_info.filename }}</p>
+                <video autoplay muted loop type="video/mp4" class="min-w-64" @timeupdate="() => {
+                    if (video.duration) {
+                      const time = video.currentTime;
+                      const pc = (100 / video.duration) * time;
+                      $refs.slider.value = pc;
+                      dygraphTime = time;
+                      dygraphIndex = Math.floor(pc * dygraph.numRows() / 100);
+                      dygraph.setSelection(dygraphIndex, undefined, true, true);
+
+                      $refs.timer.textContent = formatTime(time) + ' / ' + formatTime(video.duration);
+
+                      updateTimeQuery(time.toFixed(2));
+                    }
+                }" @ended="() => {
+                    $refs.btnPlay.classList.remove('hidden');
+                    $refs.btnPause.classList.add('hidden');
+                }"
+                    @loadedmetadata="() => ($refs.timer.textContent = formatTime(0) + ' / ' + formatTime(video.duration))">
+                    <source src="{{ video_info.url }}">
+                    Your browser does not support the video tag.
+                </video>
+            </div>
+            {% endfor %}
+        </div>
+
+        <!-- Shortcuts info -->
+        <div class="text-sm hidden md:block">
+            Hotkeys: <span class="font-mono">Space</span> to pause/unpause, <span class="font-mono">Arrow Down</span> to go to next episode, <span class="font-mono">Arrow Up</span> to go to previous episode.
+        </div>
+
+        <!-- Controllers -->
+        <div class="flex gap-1 text-3xl items-center">
+            <button x-ref="btnPlay" class="-rotate-90 hidden" class="-rotate-90" title="Play. Toggle with Space" @click="() => {
+                videos.forEach(video => video.play());
+                $refs.btnPlay.classList.toggle('hidden');
+                $refs.btnPause.classList.toggle('hidden');
+            }">🔽</button>
+            <button x-ref="btnPause" title="Pause. Toggle with Space" @click="() => {
+                videos.forEach(video => video.pause());
+                $refs.btnPlay.classList.toggle('hidden');
+                $refs.btnPause.classList.toggle('hidden');
+            }">⏸️</button>
+            <button title="Jump backward 5 seconds"
+                @click="() => (videos.forEach(video => (video.currentTime -= 5)))">⏪</button>
+            <button title="Jump forward 5 seconds"
+                @click="() => (videos.forEach(video => (video.currentTime += 5)))">⏩</button>
+            <button title="Rewind from start"
+                @click="() => (videos.forEach(video => (video.currentTime = 0.0)))">↩️</button>
+            <input x-ref="slider" max="100" min="0" step="1" type="range" value="0" class="w-80 mx-2" @input="() => {
+                const sliderValue = $refs.slider.value;
+                $refs.btnPause.click();
+                videos.forEach(video => {
+                    const time = (video.duration * sliderValue) / 100;
+                    video.currentTime = time;
+                });
+            }" />
+            <div x-ref="timer" class="font-mono text-sm border border-slate-500 rounded-lg px-1 py-0.5 shrink-0">0:00 /
+                0:00
+            </div>
+        </div>
+
+        <!-- Graph -->
+        <div class="flex gap-2 mb-4 flex-wrap">
+            <div>
+                <div id="graph" @mouseleave="() => {
+                    dygraph.setSelection(dygraphIndex, undefined, true, true);
+                    dygraphTime = video.currentTime;
+                }">
+                </div>
+                <p x-ref="graphTimer" class="font-mono ml-14 mt-4"
+                    x-init="$watch('dygraphTime', value => ($refs.graphTimer.innerText = `Time: ${dygraphTime.toFixed(2)}s`))">
+                    Time: 0.00s
+                </p>
+            </div>
+
+            <table class="text-sm border-collapse border border-slate-700" x-show="currentFrameData">
+                <thead>
+                    <tr>
+                        <th></th>
+                        <template x-for="(_, colIndex) in Array.from({length: nColumns}, (_, index) => index)">
+                            <th class="border border-slate-700">
+                                <div class="flex gap-x-2 justify-between px-2">
+                                    <input type="checkbox" :checked="isColumnChecked(colIndex)"
+                                        @change="toggleColumn(colIndex)">
+                                    <p x-text="`${columnNames[colIndex]}`"></p>
+                                </div>
+                            </th>
+                        </template>
+                    </tr>
+                </thead>
+                <tbody>
+                    <template x-for="(row, rowIndex) in rows">
+                        <tr class="odd:bg-gray-800 even:bg-gray-900">
+                            <td class="border border-slate-700">
+                                <div class="flex gap-x-2 w-24 font-semibold px-1">
+                                    <input type="checkbox" :checked="isRowChecked(rowIndex)"
+                                        @change="toggleRow(rowIndex)">
+                                    <p x-text="`Motor ${rowIndex}`"></p>
+                                </div>
+                            </td>
+                            <template x-for="(cell, colIndex) in row">
+                                <td x-show="cell" class="border border-slate-700">
+                                    <div class="flex gap-x-2 w-24 justify-between px-2">
+                                        <input type="checkbox" x-model="cell.checked" @change="updateTableValues()">
+                                        <span x-text="`${cell.value.toFixed(2)}`"
+                                            :style="`color: ${cell.color}`"></span>
+                                    </div>
+                                </td>
+                            </template>
+                        </tr>
+                    </template>
+                </tbody>
+            </table>
+
+            <div id="labels" class="hidden">
+            </div>
+        </div>
+    </div>
+
+    <script>
+        function createAlpineData() {
+            return {
+                // state
+                dygraph: null,
+                currentFrameData: null,
+                columnNames: ["state", "action", "pred action"],
+                nColumns: {% if has_policy %}3{% else %}2{% endif %},
+                checked: [],
+                dygraphTime: 0.0,
+                dygraphIndex: 0,
+                videos: null,
+                video: null,
+                colors: null,
+
+                // alpine initialization
+                init() {
+                    this.videos = document.querySelectorAll('video');
+                    this.video = this.videos[0];
+                    this.dygraph = new Dygraph(document.getElementById("graph"), '{{ ep_csv_url }}', {
+                        pixelsPerPoint: 0.01,
+                        legend: 'always',
+                        labelsDiv: document.getElementById('labels'),
+                        labelsKMB: true,
+                        strokeWidth: 1.5,
+                        pointClickCallback: (event, point) => {
+                            this.dygraphTime = point.xval;
+                            this.updateTableValues(this.dygraphTime);
+                        },
+                        highlightCallback: (event, x, points, row, seriesName) => {
+                            this.dygraphTime = x;
+                            this.updateTableValues(this.dygraphTime);
+                        },
+                        drawCallback: (dygraph, is_initial) => {
+                            if (is_initial) {
+                                // dygraph initialization
+                                this.dygraph.setSelection(this.dygraphIndex, undefined, true, true);
+                                this.colors = this.dygraph.getColors();
+                                this.checked = Array(this.colors.length).fill(true);
+
+                                const seriesNames = this.dygraph.getLabels().slice(1);
+                                const colors = [];
+                                const LIGHTNESS = [30, 65, 85]; // state_lightness, action_lightness, pred_action_lightness
+                                let lightnessIdx = 0;
+                                const chunkSize = Math.ceil(seriesNames.length / this.nColumns);
+                                for (let i = 0; i < seriesNames.length; i += chunkSize) {
+                                    const lightness = LIGHTNESS[lightnessIdx];
+                                    for (let hue = 0; hue < 360; hue += parseInt(360/chunkSize)) {
+                                        const color = `hsl(${hue}, 100%, ${lightness}%)`;
+                                        colors.push(color);
+                                    }
+                                    lightnessIdx += 1;
+                                }
+                                this.dygraph.updateOptions({ colors });
+                                this.colors = colors;
+
+                                this.updateTableValues();
+
+                                let url = new URL(window.location.href);
+                                let params = new URLSearchParams(url.search);
+                                let time = params.get("t");
+                                if(time){
+                                    time = parseFloat(time);
+                                    this.videos.forEach(video => (video.currentTime = time));
+                                }
+                            }
+                        },
+                    });
+                },
+
+                //#region Table Data
+
+                // turn dygraph's 1D data (at a given time t) to 2D data that whose columns names are defined in this.columnNames.
+                // 2d data view is used to create html table element.
+                get rows() {
+                    if (!this.currentFrameData) {
+                        return [];
+                    }
+                    const columnSize = Math.ceil(this.currentFrameData.length / this.nColumns);
+                    return Array.from({
+                        length: columnSize
+                    }, (_, rowIndex) => {
+                        const row = [
+                            this.currentFrameData[rowIndex] || null,
+                            this.currentFrameData[rowIndex + columnSize] || null,
+                        ];
+                        if (this.nColumns === 3) {
+                            row.push(this.currentFrameData[rowIndex + 2 * columnSize] || null)
+                        }
+                        return row;
+                    });
+                },
+                isRowChecked(rowIndex) {
+                    return this.rows[rowIndex].every(cell => cell && cell.checked);
+                },
+                isColumnChecked(colIndex) {
+                    return this.rows.every(row => row[colIndex] && row[colIndex].checked);
+                },
+                toggleRow(rowIndex) {
+                    const newState = !this.isRowChecked(rowIndex);
+                    this.rows[rowIndex].forEach(cell => {
+                        if (cell) cell.checked = newState;
+                    });
+                    this.updateTableValues();
+                },
+                toggleColumn(colIndex) {
+                    const newState = !this.isColumnChecked(colIndex);
+                    this.rows.forEach(row => {
+                        if (row[colIndex]) row[colIndex].checked = newState;
+                    });
+                    this.updateTableValues();
+                },
+
+                // given time t, update the values in the html table with "data[t]"
+                updateTableValues(time) {
+                    if (!this.colors) {
+                        return;
+                    }
+                    let pc = (100 / this.video.duration) * (time === undefined ? this.video.currentTime : time);
+                    if (isNaN(pc)) pc = 0;
+                    const index = Math.floor(pc * this.dygraph.numRows() / 100);
+                    // slice(1) to remove the timestamp point that we do not need
+                    const labels = this.dygraph.getLabels().slice(1);
+                    const values = this.dygraph.rawData_[index].slice(1);
+                    const checkedNew = this.currentFrameData ? this.currentFrameData.map(cell => cell.checked) : Array(
+                        this.colors.length).fill(true);
+                    this.currentFrameData = labels.map((label, idx) => ({
+                        label,
+                        value: values[idx],
+                        color: this.colors[idx],
+                        checked: checkedNew[idx],
+                    }));
+                    const shouldUpdateVisibility = !this.checked.every((value, index) => value === checkedNew[index]);
+                    if (shouldUpdateVisibility) {
+                        this.checked = checkedNew;
+                        this.dygraph.setVisibility(this.checked);
+                    }
+                },
+
+                //#endregion
+
+                updateTimeQuery(time) {
+                    let url = new URL(window.location.href);
+                    let params = new URLSearchParams(url.search);
+                    params.set("t", time);
+                    url.search = params.toString();
+                    window.history.replaceState({}, '', url.toString());
+                },
+
+
+                formatTime(time) {
+                    var hours = Math.floor(time / 3600);
+                    var minutes = Math.floor((time % 3600) / 60);
+                    var seconds = Math.floor(time % 60);
+                    return (hours > 0 ? hours + ':' : '') + (minutes < 10 ? '0' + minutes : minutes) + ':' + (seconds <
+                        10 ?
+                        '0' + seconds : seconds);
+                }
+            };
+        }
+    </script>
+</body>
+
+</html>

poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 1.8.3 and should not be changed by hand.
+# This file is automatically @generated by Poetry 1.8.2 and should not be changed by hand.
 
 [[package]]
 name = "absl-py"
@@ -192,6 +192,17 @@ charset-normalizer = ["charset-normalizer"]
 html5lib = ["html5lib"]
 lxml = ["lxml"]
 
+[[package]]
+name = "blinker"
+version = "1.8.2"
+description = "Fast, simple object-to-object and broadcast signaling"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "blinker-1.8.2-py3-none-any.whl", hash = "sha256:1779309f71bf239144b9399d06ae925637cf6634cf6bd131104184531bf67c01"},
+    {file = "blinker-1.8.2.tar.gz", hash = "sha256:8f77b09d3bf7c795e969e9486f39c2c5e9c39d4ee07424be2bc594ece9642d83"},
+]
+
 [[package]]
 name = "certifi"
 version = "2024.7.4"
@@ -584,17 +595,6 @@ files = [
     {file = "debugpy-1.8.2.zip", hash = "sha256:95378ed08ed2089221896b9b3a8d021e642c24edc8fef20e5d4342ca8be65c00"},
 ]
 
-[[package]]
-name = "decorator"
-version = "4.4.2"
-description = "Decorators for Humans"
-optional = false
-python-versions = ">=2.6, !=3.0.*, !=3.1.*"
-files = [
-    {file = "decorator-4.4.2-py2.py3-none-any.whl", hash = "sha256:41fa54c2a0cc4ba648be4fd43cff00aedf5b9465c9bf18d64325bc225f08f760"},
-    {file = "decorator-4.4.2.tar.gz", hash = "sha256:e3a62f0520172440ca0dcc823749319382e377f37f140a0b99ef45fecb84bfe7"},
-]
-
 [[package]]
 name = "deepdiff"
 version = "7.0.1"
@@ -795,6 +795,7 @@ files = [
     {file = "dora_rs-0.3.5-cp37-abi3-macosx_10_12_x86_64.whl", hash = "sha256:01f811d0c6722f74743c153a7be0144686daeafa968c473e60f6b6c5dc8f5bff"},
     {file = "dora_rs-0.3.5-cp37-abi3-macosx_11_0_arm64.whl", hash = "sha256:a36e97d31eeb66e6d5913130695d188ceee1248029961012a8b4f59fd3f58670"},
     {file = "dora_rs-0.3.5-cp37-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:25d620123a733661dc740ef2b456601ddbaa69ae2b50d8141daa3c684bda385c"},
+    {file = "dora_rs-0.3.5-cp37-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:a9fdc4e73578bebb1c8d0f8bea2243a5a9e179f08c74d98576123b59b75e5cac"},
     {file = "dora_rs-0.3.5-cp37-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e65830634c58158557f0ab90e5d1f492bcbc6b74587b05825ba4c20b634dc1bd"},
     {file = "dora_rs-0.3.5-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c01f9ab8f93295341aeab2d606d484d9cff9d05f57581e2180433ec8e0d38307"},
     {file = "dora_rs-0.3.5-cp37-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:5d6d46a49a34cd7e4f74496a1089b9a1b78282c219a28d98fe031a763e92d530"},
@@ -892,6 +893,28 @@ docs = ["furo (>=2023.9.10)", "sphinx (>=7.2.6)", "sphinx-autodoc-typehints (>=1
 testing = ["covdefaults (>=2.3)", "coverage (>=7.3.2)", "diff-cover (>=8.0.1)", "pytest (>=7.4.3)", "pytest-asyncio (>=0.21)", "pytest-cov (>=4.1)", "pytest-mock (>=3.12)", "pytest-timeout (>=2.2)", "virtualenv (>=20.26.2)"]
 typing = ["typing-extensions (>=4.8)"]
 
+[[package]]
+name = "flask"
+version = "3.0.3"
+description = "A simple framework for building complex web applications."
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "flask-3.0.3-py3-none-any.whl", hash = "sha256:34e815dfaa43340d1d15a5c3a02b8476004037eb4840b34910c6e21679d288f3"},
+    {file = "flask-3.0.3.tar.gz", hash = "sha256:ceb27b0af3823ea2737928a4d99d125a06175b8512c445cbd9a9ce200ef76842"},
+]
+
+[package.dependencies]
+blinker = ">=1.6.2"
+click = ">=8.1.3"
+itsdangerous = ">=2.1.2"
+Jinja2 = ">=3.1.2"
+Werkzeug = ">=3.0.0"
+
+[package.extras]
+async = ["asgiref (>=3.2)"]
+dotenv = ["python-dotenv"]
+
 [[package]]
 name = "frozenlist"
 version = "1.4.1"
@@ -1350,6 +1373,7 @@ files = [
 filelock = "*"
 fsspec = ">=2023.5.0"
 hf-transfer = {version = ">=0.1.4", optional = true, markers = "extra == \"hf-transfer\""}
+InquirerPy = {version = "0.3.4", optional = true, markers = "extra == \"cli\""}
 packaging = ">=20.9"
 pyyaml = ">=5.1"
 requests = "*"
@@ -1536,6 +1560,24 @@ files = [
     {file = "iniconfig-2.0.0.tar.gz", hash = "sha256:2d91e135bf72d31a410b17c16da610a82cb55f6b0477d1a902134b24a455b8b3"},
 ]
 
+[[package]]
+name = "inquirerpy"
+version = "0.3.4"
+description = "Python port of Inquirer.js (A collection of common interactive command-line user interfaces)"
+optional = false
+python-versions = ">=3.7,<4.0"
+files = [
+    {file = "InquirerPy-0.3.4-py3-none-any.whl", hash = "sha256:c65fdfbac1fa00e3ee4fb10679f4d3ed7a012abf4833910e63c295827fe2a7d4"},
+    {file = "InquirerPy-0.3.4.tar.gz", hash = "sha256:89d2ada0111f337483cb41ae31073108b2ec1e618a49d7110b0d7ade89fc197e"},
+]
+
+[package.dependencies]
+pfzy = ">=0.3.1,<0.4.0"
+prompt-toolkit = ">=3.0.1,<4.0.0"
+
+[package.extras]
+docs = ["Sphinx (>=4.1.2,<5.0.0)", "furo (>=2021.8.17-beta.43,<2022.0.0)", "myst-parser (>=0.15.1,<0.16.0)", "sphinx-autobuild (>=2021.3.14,<2022.0.0)", "sphinx-copybutton (>=0.4.0,<0.5.0)"]
+
 [[package]]
 name = "intel-openmp"
 version = "2021.4.0"
@@ -1550,6 +1592,17 @@ files = [
     {file = "intel_openmp-2021.4.0-py2.py3-none-win_amd64.whl", hash = "sha256:eef4c8bcc8acefd7f5cd3b9384dbf73d59e2c99fc56545712ded913f43c4a94f"},
 ]
 
+[[package]]
+name = "itsdangerous"
+version = "2.2.0"
+description = "Safely pass data to untrusted environments and back."
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "itsdangerous-2.2.0-py3-none-any.whl", hash = "sha256:c6242fc49e35958c8b15141343aa660db5fc54d4f13a1db01a3f5891b98700ef"},
+    {file = "itsdangerous-2.2.0.tar.gz", hash = "sha256:e0050c0b7da1eea53ffaf149c0cfbb5c6e2e2b69c4bef22c81fa6eb73e5f6173"},
+]
+
 [[package]]
 name = "jinja2"
 version = "3.1.4"
@@ -1741,9 +1794,13 @@ files = [
     {file = "lxml-5.2.2-cp36-cp36m-win_amd64.whl", hash = "sha256:edcfa83e03370032a489430215c1e7783128808fd3e2e0a3225deee278585196"},
     {file = "lxml-5.2.2-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:28bf95177400066596cdbcfc933312493799382879da504633d16cf60bba735b"},
     {file = "lxml-5.2.2-cp37-cp37m-manylinux_2_12_i686.manylinux2010_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:3a745cc98d504d5bd2c19b10c79c61c7c3df9222629f1b6210c0368177589fb8"},
+    {file = "lxml-5.2.2-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1b590b39ef90c6b22ec0be925b211298e810b4856909c8ca60d27ffbca6c12e6"},
     {file = "lxml-5.2.2-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b336b0416828022bfd5a2e3083e7f5ba54b96242159f83c7e3eebaec752f1716"},
+    {file = "lxml-5.2.2-cp37-cp37m-manylinux_2_28_aarch64.whl", hash = "sha256:c2faf60c583af0d135e853c86ac2735ce178f0e338a3c7f9ae8f622fd2eb788c"},
     {file = "lxml-5.2.2-cp37-cp37m-manylinux_2_28_x86_64.whl", hash = "sha256:4bc6cb140a7a0ad1f7bc37e018d0ed690b7b6520ade518285dc3171f7a117905"},
+    {file = "lxml-5.2.2-cp37-cp37m-musllinux_1_1_aarch64.whl", hash = "sha256:7ff762670cada8e05b32bf1e4dc50b140790909caa8303cfddc4d702b71ea184"},
     {file = "lxml-5.2.2-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:57f0a0bbc9868e10ebe874e9f129d2917750adf008fe7b9c1598c0fbbfdde6a6"},
+    {file = "lxml-5.2.2-cp37-cp37m-musllinux_1_2_aarch64.whl", hash = "sha256:a6d2092797b388342c1bc932077ad232f914351932353e2e8706851c870bca1f"},
     {file = "lxml-5.2.2-cp37-cp37m-musllinux_1_2_x86_64.whl", hash = "sha256:60499fe961b21264e17a471ec296dcbf4365fbea611bf9e303ab69db7159ce61"},
     {file = "lxml-5.2.2-cp37-cp37m-win32.whl", hash = "sha256:d9b342c76003c6b9336a80efcc766748a333573abf9350f4094ee46b006ec18f"},
     {file = "lxml-5.2.2-cp37-cp37m-win_amd64.whl", hash = "sha256:b16db2770517b8799c79aa80f4053cd6f8b716f21f8aca962725a9565ce3ee40"},
@@ -1901,30 +1958,6 @@ files = [
 intel-openmp = "==2021.*"
 tbb = "==2021.*"
 
-[[package]]
-name = "moviepy"
-version = "1.0.3"
-description = "Video editing with Python"
-optional = false
-python-versions = "*"
-files = [
-    {file = "moviepy-1.0.3.tar.gz", hash = "sha256:2884e35d1788077db3ff89e763c5ba7bfddbd7ae9108c9bc809e7ba58fa433f5"},
-]
-
-[package.dependencies]
-decorator = ">=4.0.2,<5.0"
-imageio = {version = ">=2.5,<3.0", markers = "python_version >= \"3.4\""}
-imageio_ffmpeg = {version = ">=0.2.0", markers = "python_version >= \"3.4\""}
-numpy = {version = ">=1.17.3", markers = "python_version > \"2.7\""}
-proglog = "<=1.0.0"
-requests = ">=2.8.1,<3.0"
-tqdm = ">=4.11.2,<5.0"
-
-[package.extras]
-doc = ["Sphinx (>=1.5.2,<2.0)", "numpydoc (>=0.6.0,<1.0)", "pygame (>=1.9.3,<2.0)", "sphinx_rtd_theme (>=0.1.10b0,<1.0)"]
-optional = ["matplotlib (>=2.0.0,<3.0)", "opencv-python (>=3.0,<4.0)", "scikit-image (>=0.13.0,<1.0)", "scikit-learn", "scipy (>=0.19.0,<1.5)", "youtube_dl"]
-test = ["coverage (<5.0)", "coveralls (>=1.1,<2.0)", "pytest (>=3.0.0,<4.0)", "pytest-cov (>=2.5.1,<3.0)", "requests (>=2.8.1,<3.0)"]
-
 [[package]]
 name = "mpmath"
 version = "1.3.0"
@@ -2550,6 +2583,20 @@ other = ["pillow (>=8.0.1)"]
 sisl = ["box2d-py (==2.3.5)", "pygame (==2.3.0)", "pymunk (==6.2.0)", "scipy (>=1.4.1)"]
 testing = ["AutoROM", "pre-commit", "pynput", "pytest", "pytest-cov", "pytest-markdown-docs", "pytest-xdist"]
 
+[[package]]
+name = "pfzy"
+version = "0.3.4"
+description = "Python port of the fzy fuzzy string matching algorithm"
+optional = false
+python-versions = ">=3.7,<4.0"
+files = [
+    {file = "pfzy-0.3.4-py3-none-any.whl", hash = "sha256:5f50d5b2b3207fa72e7ec0ef08372ef652685470974a107d0d4999fc5a903a96"},
+    {file = "pfzy-0.3.4.tar.gz", hash = "sha256:717ea765dd10b63618e7298b2d98efd819e0b30cd5905c9707223dceeb94b3f1"},
+]
+
+[package.extras]
+docs = ["Sphinx (>=4.1.2,<5.0.0)", "furo (>=2021.8.17-beta.43,<2022.0.0)", "myst-parser (>=0.15.1,<0.16.0)", "sphinx-autobuild (>=2021.3.14,<2022.0.0)", "sphinx-copybutton (>=0.4.0,<0.5.0)"]
+
 [[package]]
 name = "pillow"
 version = "10.4.0"
@@ -2697,18 +2744,18 @@ pyyaml = ">=5.1"
 virtualenv = ">=20.10.0"
 
 [[package]]
-name = "proglog"
-version = "0.1.10"
-description = "Log and progress bar manager for console, notebooks, web..."
+name = "prompt-toolkit"
+version = "3.0.47"
+description = "Library for building powerful interactive command lines in Python"
 optional = false
-python-versions = "*"
+python-versions = ">=3.7.0"
 files = [
-    {file = "proglog-0.1.10-py3-none-any.whl", hash = "sha256:19d5da037e8c813da480b741e3fa71fb1ac0a5b02bf21c41577c7f327485ec50"},
-    {file = "proglog-0.1.10.tar.gz", hash = "sha256:658c28c9c82e4caeb2f25f488fff9ceace22f8d69b15d0c1c86d64275e4ddab4"},
+    {file = "prompt_toolkit-3.0.47-py3-none-any.whl", hash = "sha256:0d7bfa67001d5e39d02c224b663abc33687405033a8c422d0d675a5a13361d10"},
+    {file = "prompt_toolkit-3.0.47.tar.gz", hash = "sha256:1e1b29cb58080b1e69f207c893a1a7bf16d127a5c30c9d17a25a5d77792e5360"},
 ]
 
 [package.dependencies]
-tqdm = "*"
+wcwidth = "*"
 
 [[package]]
 name = "protobuf"
@@ -3276,6 +3323,7 @@ files = [
     {file = "PyYAML-6.0.1-cp311-cp311-win_amd64.whl", hash = "sha256:bf07ee2fef7014951eeb99f56f39c9bb4af143d8aa3c21b1677805985307da34"},
     {file = "PyYAML-6.0.1-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:855fb52b0dc35af121542a76b9a84f8d1cd886ea97c84703eaa6d88e37a2ad28"},
     {file = "PyYAML-6.0.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:40df9b996c2b73138957fe23a16a4f0ba614f4c0efce1e9406a184b6d07fa3a9"},
+    {file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a08c6f0fe150303c1c6b71ebcd7213c2858041a7e01975da3a99aed1e7a378ef"},
     {file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6c22bec3fbe2524cde73d7ada88f6566758a8f7227bfbf93a408a9d86bcc12a0"},
     {file = "PyYAML-6.0.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:8d4e9c88387b0f5c7d5f281e55304de64cf7f9c0021a3525bd3b1c542da3b0e4"},
     {file = "PyYAML-6.0.1-cp312-cp312-win32.whl", hash = "sha256:d483d2cdf104e7c9fa60c544d92981f12ad66a457afae824d146093b8c294c54"},
@@ -3809,13 +3857,13 @@ test = ["pytest"]
 
 [[package]]
 name = "setuptools"
-version = "71.0.1"
+version = "71.0.0"
 description = "Easily download, build, install, upgrade, and uninstall Python packages"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "setuptools-71.0.1-py3-none-any.whl", hash = "sha256:1eb8ef012efae7f6acbc53ec0abde4bc6746c43087fd215ee09e1df48998711f"},
-    {file = "setuptools-71.0.1.tar.gz", hash = "sha256:c51d7fd29843aa18dad362d4b4ecd917022131425438251f4e3d766c964dd1ad"},
+    {file = "setuptools-71.0.0-py3-none-any.whl", hash = "sha256:f06fbe978a91819d250a30e0dc4ca79df713d909e24438a42d0ec300fc52247f"},
+    {file = "setuptools-71.0.0.tar.gz", hash = "sha256:98da3b8aca443b9848a209ae4165e2edede62633219afa493a58fbba57f72e2e"},
 ]
 
 [package.extras]
@@ -4215,6 +4263,34 @@ perf = ["orjson"]
 sweeps = ["sweeps (>=0.2.0)"]
 workspaces = ["wandb-workspaces"]
 
+[[package]]
+name = "wcwidth"
+version = "0.2.13"
+description = "Measures the displayed width of unicode strings in a terminal"
+optional = false
+python-versions = "*"
+files = [
+    {file = "wcwidth-0.2.13-py2.py3-none-any.whl", hash = "sha256:3da69048e4540d84af32131829ff948f1e022c1c6bdb8d6102117aac784f6859"},
+    {file = "wcwidth-0.2.13.tar.gz", hash = "sha256:72ea0c06399eb286d978fdedb6923a9eb47e1c486ce63e9b4e64fc18303972b5"},
+]
+
+[[package]]
+name = "werkzeug"
+version = "3.0.3"
+description = "The comprehensive WSGI web application library."
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "werkzeug-3.0.3-py3-none-any.whl", hash = "sha256:fc9645dc43e03e4d630d23143a04a7f947a9a3b5727cd535fdfe155a17cc48c8"},
+    {file = "werkzeug-3.0.3.tar.gz", hash = "sha256:097e5bfda9f0aba8da6b8545146def481d06aa7d3266e7448e2cccf67dd8bd18"},
+]
+
+[package.dependencies]
+MarkupSafe = ">=2.1.1"
+
+[package.extras]
+watchdog = ["watchdog (>=2.3)"]
+
 [[package]]
 name = "xxhash"
 version = "3.4.1"
@@ -4485,4 +4561,4 @@ xarm = ["gym-xarm"]
 [metadata]
 lock-version = "2.0"
 python-versions = ">=3.10,<3.13"
-content-hash = "dfe9c6a54e0382156e62e7bd2c7aab1be6372da76d30c61b06d27232276638cb"
+content-hash = "a340f2ed23db2f3c371c494cbc9a33392e122ed6713e6098277a87b3fb805f2b"

pyproject.toml

@@ -43,7 +43,7 @@ opencv-python = ">=4.9.0"
 diffusers = ">=0.27.2"
 torchvision = ">=0.17.1"
 h5py = ">=3.10.0"
-huggingface-hub = {extras = ["hf-transfer"], version = ">=0.23.0"}
+huggingface-hub = {extras = ["hf-transfer", "cli"], version = ">=0.23.0"}
 gymnasium = ">=0.29.1"
 cmake = ">=3.29.0.1"
 gym-dora = { git = "https://github.com/dora-rs/dora-lerobot.git", subdirectory = "gym_dora", optional = true }
@@ -57,13 +57,15 @@ pytest-cov = {version = ">=5.0.0", optional = true}
 datasets = ">=2.19.0"
 imagecodecs = { version = ">=2024.1.1", optional = true }
 pyav = ">=12.0.5"
-moviepy = ">=1.0.3"
 rerun-sdk = ">=0.15.1"
 deepdiff = ">=7.0.1"
-scikit-image = {version = ">=0.23.2", optional = true}
+flask = ">=3.0.3"
 pandas = {version = ">=2.2.2", optional = true}
+scikit-image = {version = ">=0.23.2", optional = true}
 dynamixel-sdk = {version = ">=3.7.31", optional = true}
 pynput = {version = ">=1.7.7", optional = true}
+# TODO(rcadene, salibert): 71.0.1 has a bug
+setuptools = {version = "!=71.0.1", optional = true}
 
 
 

tests/test_control_robot.py

@@ -3,17 +3,18 @@ from pathlib import Path
 from lerobot.common.policies.factory import make_policy
 from lerobot.common.robot_devices.robots.factory import make_robot
 from lerobot.common.utils.utils import init_hydra_config
-from lerobot.scripts.control_robot import record, replay, teleoperate
+from lerobot.scripts.control_robot import calibrate, record, replay, teleoperate
 from tests.utils import DEFAULT_CONFIG_PATH, DEVICE, KOCH_ROBOT_CONFIG_PATH, require_koch
 
 
-def make_robot_():
-    robot_cfg = init_hydra_config(KOCH_ROBOT_CONFIG_PATH)
+def make_robot_(overrides=None):
+    robot_cfg = init_hydra_config(KOCH_ROBOT_CONFIG_PATH, overrides)
     robot = make_robot(robot_cfg)
     return robot
 
 
 @require_koch
+# `require_koch` uses `request` to access `is_koch_available` fixture
 def test_teleoperate(request):
     robot = make_robot_()
     teleoperate(robot, teleop_time_s=1)
@@ -22,6 +23,22 @@ def test_teleoperate(request):
     del robot
 
 
+@require_koch
+def test_calibrate(request):
+    robot = make_robot_()
+    calibrate(robot)
+    del robot
+
+
+@require_koch
+def test_record_without_cameras(tmpdir, request):
+    root = Path(tmpdir)
+    repo_id = "lerobot/debug"
+
+    robot = make_robot_(overrides=["~cameras"])
+    record(robot, fps=30, root=root, repo_id=repo_id, warmup_time_s=1, episode_time_s=1, num_episodes=2)
+
+
 @require_koch
 def test_record_and_replay_and_policy(tmpdir, request):
     env_name = "koch_real"

tests/test_datasets.py

@@ -23,6 +23,7 @@ import einops
 import pytest
 import torch
 from datasets import Dataset
+from huggingface_hub import HfApi
 from safetensors.torch import load_file
 
 import lerobot
@@ -34,6 +35,7 @@ from lerobot.common.datasets.compute_stats import (
 from lerobot.common.datasets.factory import make_dataset
 from lerobot.common.datasets.lerobot_dataset import LeRobotDataset, MultiLeRobotDataset
 from lerobot.common.datasets.utils import (
+    create_branch,
     flatten_dict,
     hf_transform_to_torch,
     load_previous_and_future_frames,
@@ -385,3 +387,29 @@ def test_aggregate_stats():
         for agg_fn in ["mean", "min", "max"]:
             assert torch.allclose(stats[data_key][agg_fn], einops.reduce(data, "n -> 1", agg_fn))
         assert torch.allclose(stats[data_key]["std"], torch.std(data, correction=0))
+
+
+@pytest.mark.skip("Requires internet access")
+def test_create_branch():
+    api = HfApi()
+
+    repo_id = "cadene/test_create_branch"
+    repo_type = "dataset"
+    branch = "test"
+    ref = f"refs/heads/{branch}"
+
+    # Prepare a repo with a test branch
+    api.delete_repo(repo_id, repo_type=repo_type, missing_ok=True)
+    api.create_repo(repo_id, repo_type=repo_type)
+    create_branch(repo_id, repo_type=repo_type, branch=branch)
+
+    # Make sure the test branch exists
+    branches = api.list_repo_refs(repo_id, repo_type=repo_type).branches
+    refs = [branch.ref for branch in branches]
+    assert ref in refs
+
+    # Overwrite it
+    create_branch(repo_id, repo_type=repo_type, branch=branch)
+
+    # Clean
+    api.delete_repo(repo_id, repo_type=repo_type)

tests/test_visualize_dataset_html.py

@@ -0,0 +1,36 @@
+#!/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.
+
+from pathlib import Path
+
+import pytest
+
+from lerobot.scripts.visualize_dataset_html import visualize_dataset_html
+
+
+@pytest.mark.parametrize(
+    "repo_id",
+    ["lerobot/pusht"],
+)
+def test_visualize_dataset_html(tmpdir, repo_id):
+    tmpdir = Path(tmpdir)
+    visualize_dataset_html(
+        repo_id,
+        episodes=[0],
+        output_dir=tmpdir,
+        serve=False,
+    )
+    assert (tmpdir / "static" / "episode_0.csv").exists()