fix: sharing predicted chunk with user

Co-authored-by: makolon <smakolon385@gmail.com>
Co-authored-by: Steven Palma <imstevenpmwork@ieee.org>

.pre-commit-config.yaml

@@ -36,8 +36,8 @@ repos:
       - id: end-of-file-fixer
       - id: trailing-whitespace
 
-  - repo: https://github.com/crate-ci/typos
-    rev: v1
+  - repo: https://github.com/adhtruong/mirrors-typos
+    rev: v1.31.1
     hooks:
       - id: typos
         args: [--force-exclude]
@@ -48,7 +48,7 @@ repos:
     -   id: pyupgrade
 
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.11.4
+    rev: v0.11.5
     hooks:
       - id: ruff
         args: [--fix]
@@ -57,7 +57,7 @@ repos:
 
   ##### Security #####
   - repo: https://github.com/gitleaks/gitleaks
-    rev: v8.24.2
+    rev: v8.24.3
     hooks:
       - id: gitleaks
 

README.md

@@ -98,18 +98,25 @@ conda create -y -n lerobot python=3.10
 conda activate lerobot
 ```
 
-When using `miniconda`, if you don't have `ffmpeg` in your environment:
+When using `miniconda`, install `ffmpeg` in your environment:
 ```bash
 conda install ffmpeg -c conda-forge
 ```
 
+> **NOTE:** This usually installs `ffmpeg 7.X` for your platform compiled with the `libsvtav1` encoder. If `libsvtav1` is not supported (check supported encoders with `ffmpeg -encoders`), you can:
+>  - _[On any platform]_ Explicitly install `ffmpeg 7.X` using:
+>  ```bash
+>  conda install ffmpeg=7.1.1 -c conda-forge
+>  ```
+>  - _[On Linux only]_ Install [ffmpeg build dependencies](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#GettheDependencies) and [compile ffmpeg from source with libsvtav1](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#libsvtav1), and make sure you use the corresponding ffmpeg binary to your install with `which ffmpeg`.
+
 Install 🤗 LeRobot:
 ```bash
-pip install --no-binary=av -e .
+pip install -e .
 ```
 
 > **NOTE:** If you encounter build errors, you may need to install additional dependencies (`cmake`, `build-essential`, and `ffmpeg libs`). On Linux, run:
-`sudo apt-get install cmake build-essential python-dev pkg-config libavformat-dev libavcodec-dev libavdevice-dev libavutil-dev libswscale-dev libswresample-dev libavfilter-dev pkg-config`. For other systems, see: [Compiling PyAV](https://pyav.org/docs/develop/overview/installation.html#bring-your-own-ffmpeg)
+`sudo apt-get install cmake build-essential python3-dev pkg-config libavformat-dev libavcodec-dev libavdevice-dev libavutil-dev libswscale-dev libswresample-dev libavfilter-dev pkg-config`. For other systems, see: [Compiling PyAV](https://pyav.org/docs/develop/overview/installation.html#bring-your-own-ffmpeg)
 
 For simulations, 🤗 LeRobot comes with gymnasium environments that can be installed as extras:
 - [aloha](https://github.com/huggingface/gym-aloha)
@@ -118,7 +125,7 @@ For simulations, 🤗 LeRobot comes with gymnasium environments that can be inst
 
 For instance, to install 🤗 LeRobot with aloha and pusht, use:
 ```bash
-pip install --no-binary=av -e ".[aloha, pusht]"
+pip install -e ".[aloha, pusht]"
 ```
 
 To use [Weights and Biases](https://docs.wandb.ai/quickstart) for experiment tracking, log in with

benchmarks/video/capture_camera_feed.py

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

docker/lerobot-gpu-dev/Dockerfile

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

examples/10_use_so100.md

@@ -57,9 +57,15 @@ conda activate lerobot
 git clone https://github.com/huggingface/lerobot.git ~/lerobot
 ```
 
-#### 5. Install LeRobot with dependencies for the feetech motors:
+#### 5. Install ffmpeg in your environment:
+When using `miniconda`, install `ffmpeg` in your environment:
 ```bash
-cd ~/lerobot && pip install --no-binary=av -e ".[feetech]"
+conda install ffmpeg -c conda-forge
+```
+
+#### 6. Install LeRobot with dependencies for the feetech motors:
+```bash
+cd ~/lerobot && pip install -e ".[feetech]"
 ```
 
 Great :hugs:! You are now done installing LeRobot and we can begin assembling the SO100 arms :robot:.
@@ -491,6 +497,9 @@ python lerobot/scripts/control_robot.py \
 
 #### a. Teleop with displaying cameras
 Follow [this guide to setup your cameras](https://github.com/huggingface/lerobot/blob/main/examples/7_get_started_with_real_robot.md#c-add-your-cameras-with-opencvcamera). Then you will be able to display the cameras on your computer while you are teleoperating by running the following code. This is useful to prepare your setup before recording your first dataset.
+
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`.
+
 ```bash
 python lerobot/scripts/control_robot.py \
   --robot.type=so100 \

examples/11_use_lekiwi.md

@@ -67,9 +67,15 @@ conda activate lerobot
 git clone https://github.com/huggingface/lerobot.git ~/lerobot
 ```
 
-#### 5. Install LeRobot with dependencies for the feetech motors:
+#### 5. Install ffmpeg in your environment:
+When using `miniconda`, install `ffmpeg` in your environment:
 ```bash
-cd ~/lerobot && pip install --no-binary=av -e ".[feetech]"
+conda install ffmpeg -c conda-forge
+```
+
+#### 6. Install LeRobot with dependencies for the feetech motors:
+```bash
+cd ~/lerobot && pip install -e ".[feetech]"
 ```
 
 ## C. Install LeRobot on laptop
@@ -108,9 +114,15 @@ conda activate lerobot
 git clone https://github.com/huggingface/lerobot.git ~/lerobot
 ```
 
-#### 5. Install LeRobot with dependencies for the feetech motors:
+#### 5. Install ffmpeg in your environment:
+When using `miniconda`, install `ffmpeg` in your environment:
 ```bash
-cd ~/lerobot && pip install --no-binary=av -e ".[feetech]"
+conda install ffmpeg -c conda-forge
+```
+
+#### 6. Install LeRobot with dependencies for the feetech motors:
+```bash
+cd ~/lerobot && pip install -e ".[feetech]"
 ```
 
 Great :hugs:! You are now done installing LeRobot and we can begin assembling the SO100 arms and Mobile base :robot:.
@@ -412,6 +424,8 @@ python lerobot/scripts/control_robot.py \
   --control.fps=30
 ```
 
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`. For the `--control.type=remote_robot` you will also need to set `--control.viewer_ip` and `--control.viewer_port`
+
 You should see on your laptop something like this: ```[INFO] Connected to remote robot at tcp://172.17.133.91:5555 and video stream at tcp://172.17.133.91:5556.``` Now you can move the leader arm and use the keyboard (w,a,s,d) to drive forward, left, backwards, right. And use (z,x) to turn left or turn right. You can use (r,f) to increase and decrease the speed of the mobile robot. There are three speed modes, see the table below:
 | Speed Mode | Linear Speed (m/s) | Rotation Speed (deg/s) |
 | ---------- | ------------------ | ---------------------- |

examples/11_use_moss.md

@@ -31,9 +31,15 @@ conda create -y -n lerobot python=3.10 && conda activate lerobot
 git clone https://github.com/huggingface/lerobot.git ~/lerobot
 ```
 
-5. Install LeRobot with dependencies for the feetech motors:
+5. Install ffmpeg in your environment:
+When using `miniconda`, install `ffmpeg` in your environment:
 ```bash
-cd ~/lerobot && pip install --no-binary=av -e ".[feetech]"
+conda install ffmpeg -c conda-forge
+```
+
+6. Install LeRobot with dependencies for the feetech motors:
+```bash
+cd ~/lerobot && pip install -e ".[feetech]"
 ```
 
 ## Configure the motors
@@ -212,6 +218,9 @@ python lerobot/scripts/control_robot.py \
 
 **Teleop with displaying cameras**
 Follow [this guide to setup your cameras](https://github.com/huggingface/lerobot/blob/main/examples/7_get_started_with_real_robot.md#c-add-your-cameras-with-opencvcamera). Then you will be able to display the cameras on your computer while you are teleoperating by running the following code. This is useful to prepare your setup before recording your first dataset.
+
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`.
+
 ```bash
 python lerobot/scripts/control_robot.py \
   --robot.type=moss \

examples/2_evaluate_pretrained_policy.py

@@ -18,7 +18,7 @@ training outputs directory. In the latter case, you might want to run examples/3
 
 It requires the installation of the 'gym_pusht' simulation environment. Install it by running:
 ```bash
-pip install --no-binary=av -e ".[pusht]"`
+pip install -e ".[pusht]"
 ```
 """
 

examples/4_train_policy_with_script.md

@@ -4,7 +4,7 @@ This tutorial will explain the training script, how to use it, and particularly
 
 ## The training script
 
-LeRobot offers a training script at [`lerobot/scripts/train.py`](../../lerobot/scripts/train.py). At a high level it does the following:
+LeRobot offers a training script at [`lerobot/scripts/train.py`](../lerobot/scripts/train.py). At a high level it does the following:
 
 - Initialize/load a configuration for the following steps using.
 - Instantiates a dataset.
@@ -21,7 +21,7 @@ In the training script, the main function `train` expects a `TrainPipelineConfig
 def train(cfg: TrainPipelineConfig):
 ```
 
-You can inspect the `TrainPipelineConfig` defined in [`lerobot/configs/train.py`](../../lerobot/configs/train.py) (which is heavily commented and meant to be a reference to understand any option)
+You can inspect the `TrainPipelineConfig` defined in [`lerobot/configs/train.py`](../lerobot/configs/train.py) (which is heavily commented and meant to be a reference to understand any option)
 
 When running the script, inputs for the command line are parsed thanks to the `@parser.wrap()` decorator and an instance of this class is automatically generated. Under the hood, this is done with [Draccus](https://github.com/dlwh/draccus) which is a tool dedicated for this purpose. If you're familiar with Hydra, Draccus can similarly load configurations from config files (.json, .yaml) and also override their values through command line inputs. Unlike Hydra, these configurations are pre-defined in the code through dataclasses rather than being defined entirely in config files. This allows for more rigorous serialization/deserialization, typing, and to manipulate configuration as objects directly in the code and not as dictionaries or namespaces (which enables nice features in an IDE such as autocomplete, jump-to-def, etc.)
 
@@ -50,7 +50,7 @@ By default, every field takes its default value specified in the dataclass. If a
 
 ## Specifying values from the CLI
 
-Let's say that we want to train [Diffusion Policy](../../lerobot/common/policies/diffusion) on the [pusht](https://huggingface.co/datasets/lerobot/pusht) dataset, using the [gym_pusht](https://github.com/huggingface/gym-pusht) environment for evaluation. The command to do so would look like this:
+Let's say that we want to train [Diffusion Policy](../lerobot/common/policies/diffusion) on the [pusht](https://huggingface.co/datasets/lerobot/pusht) dataset, using the [gym_pusht](https://github.com/huggingface/gym-pusht) environment for evaluation. The command to do so would look like this:
 ```bash
 python lerobot/scripts/train.py \
     --dataset.repo_id=lerobot/pusht \
@@ -60,10 +60,10 @@ python lerobot/scripts/train.py \
 
 Let's break this down:
 - To specify the dataset, we just need to specify its `repo_id` on the hub which is the only required argument in the `DatasetConfig`. The rest of the fields have default values and in this case we are fine with those so we can just add the option `--dataset.repo_id=lerobot/pusht`.
-- To specify the policy, we can just select diffusion policy using `--policy` appended with `.type`. Here, `.type` is a special argument which allows us to select config classes inheriting from `draccus.ChoiceRegistry` and that have been decorated with the `register_subclass()` method. To have a better explanation of this feature, have a look at this [Draccus demo](https://github.com/dlwh/draccus?tab=readme-ov-file#more-flexible-configuration-with-choice-types). In our code, we use this mechanism mainly to select policies, environments, robots, and some other components like optimizers. The policies available to select are located in [lerobot/common/policies](../../lerobot/common/policies)
-- Similarly, we select the environment with `--env.type=pusht`. The different environment configs are available in [`lerobot/common/envs/configs.py`](../../lerobot/common/envs/configs.py)
+- To specify the policy, we can just select diffusion policy using `--policy` appended with `.type`. Here, `.type` is a special argument which allows us to select config classes inheriting from `draccus.ChoiceRegistry` and that have been decorated with the `register_subclass()` method. To have a better explanation of this feature, have a look at this [Draccus demo](https://github.com/dlwh/draccus?tab=readme-ov-file#more-flexible-configuration-with-choice-types). In our code, we use this mechanism mainly to select policies, environments, robots, and some other components like optimizers. The policies available to select are located in [lerobot/common/policies](../lerobot/common/policies)
+- Similarly, we select the environment with `--env.type=pusht`. The different environment configs are available in [`lerobot/common/envs/configs.py`](../lerobot/common/envs/configs.py)
 
-Let's see another example. Let's say you've been training [ACT](../../lerobot/common/policies/act) on [lerobot/aloha_sim_insertion_human](https://huggingface.co/datasets/lerobot/aloha_sim_insertion_human) using the [gym-aloha](https://github.com/huggingface/gym-aloha) environment for evaluation with:
+Let's see another example. Let's say you've been training [ACT](../lerobot/common/policies/act) on [lerobot/aloha_sim_insertion_human](https://huggingface.co/datasets/lerobot/aloha_sim_insertion_human) using the [gym-aloha](https://github.com/huggingface/gym-aloha) environment for evaluation with:
 ```bash
 python lerobot/scripts/train.py \
     --policy.type=act \
@@ -74,7 +74,7 @@ python lerobot/scripts/train.py \
 > Notice we added `--output_dir` to explicitly tell where to write outputs from this run (checkpoints, training state, configs etc.). This is not mandatory and if you don't specify it, a default directory will be created from the current date and time, env.type and policy.type. This will typically look like `outputs/train/2025-01-24/16-10-05_aloha_act`.
 
 We now want to train a different policy for aloha on another task. We'll change the dataset and use [lerobot/aloha_sim_transfer_cube_human](https://huggingface.co/datasets/lerobot/aloha_sim_transfer_cube_human) instead. Of course, we also need to change the task of the environment as well to match this other task.
-Looking at the [`AlohaEnv`](../../lerobot/common/envs/configs.py) config, the task is `"AlohaInsertion-v0"` by default, which corresponds to the task we trained on in the command above. The [gym-aloha](https://github.com/huggingface/gym-aloha?tab=readme-ov-file#description) environment also has the `AlohaTransferCube-v0` task which corresponds to this other task we want to train on. Putting this together, we can train this new policy on this different task using:
+Looking at the [`AlohaEnv`](../lerobot/common/envs/configs.py) config, the task is `"AlohaInsertion-v0"` by default, which corresponds to the task we trained on in the command above. The [gym-aloha](https://github.com/huggingface/gym-aloha?tab=readme-ov-file#description) environment also has the `AlohaTransferCube-v0` task which corresponds to this other task we want to train on. Putting this together, we can train this new policy on this different task using:
 ```bash
 python lerobot/scripts/train.py \
     --policy.type=act \

examples/7_get_started_with_real_robot.md

@@ -33,7 +33,7 @@ First, install the additional dependencies required for robots built with dynami
 
 Using `pip`:
 ```bash
-pip install --no-binary=av -e ".[dynamixel]"
+pip install -e ".[dynamixel]"
 ```
 
 Using `poetry`:
@@ -55,6 +55,9 @@ Finally, connect both arms to your computer via USB. Note that the USB doesn't p
 Now you are ready to configure your motors for the first time, as detailed in the sections below. In the upcoming sections, you'll learn about our classes and functions by running some python code in an interactive session, or by copy-pasting it in a python file.
 
 If you have already configured your motors the first time, you can streamline the process by directly running the teleoperate script (which is detailed further in the tutorial):
+
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`.
+
 ```bash
 python lerobot/scripts/control_robot.py \
   --robot.type=koch \
@@ -827,11 +830,6 @@ It contains:
 - `dtRphone:33.84 (29.5hz)` which is the delta time of capturing an image from the phone camera in the thread running asynchronously.
 
 Troubleshooting:
-- On Linux, if you encounter any issue during video encoding with `ffmpeg: unknown encoder libsvtav1`, you can:
-  - install with conda-forge by running `conda install -c conda-forge ffmpeg` (it should be compiled with `libsvtav1`),
-  - or, install [Homebrew](https://brew.sh) and run `brew install ffmpeg` (it should be compiled with `libsvtav1`),
-  - or, install [ffmpeg build dependencies](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#GettheDependencies) and [compile ffmpeg from source with libsvtav1](https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#libsvtav1),
-  - and, make sure you use the corresponding ffmpeg binary to your install with `which ffmpeg`.
 - On Linux, if the left and right arrow keys and escape key don't have any effect during data recording, make sure you've set the `$DISPLAY` environment variable. See [pynput limitations](https://pynput.readthedocs.io/en/latest/limitations.html#linux).
 
 At the end of data recording, your dataset will be uploaded on your Hugging Face page (e.g. https://huggingface.co/datasets/cadene/koch_test) that you can obtain by running:

examples/8_use_stretch.md

@@ -43,14 +43,19 @@ conda create -y -n lerobot python=3.10 && conda activate lerobot
 git clone https://github.com/huggingface/lerobot.git ~/lerobot
 ```
 
-6. Install LeRobot with stretch dependencies:
+6. When using `miniconda`, install `ffmpeg` in your environment:
 ```bash
-cd ~/lerobot && pip install --no-binary=av -e ".[stretch]"
+conda install ffmpeg -c conda-forge
+```
+
+7. Install LeRobot with stretch dependencies:
+```bash
+cd ~/lerobot && pip install -e ".[stretch]"
 ```
 
 > **Note:** If you get this message, you can ignore it: `ERROR: pip's dependency resolver does not currently take into account all the packages that are installed.`
 
-7. Run a [system check](https://docs.hello-robot.com/0.3/getting_started/stretch_hardware_overview/#system-check) to make sure your robot is ready:
+8. Run a [system check](https://docs.hello-robot.com/0.3/getting_started/stretch_hardware_overview/#system-check) to make sure your robot is ready:
 ```bash
 stretch_system_check.py
 ```
@@ -97,6 +102,8 @@ This is equivalent to running `stretch_robot_home.py`
 Before trying teleoperation, you need activate the gamepad controller by pressing the middle button. For more info, see Stretch's [doc](https://docs.hello-robot.com/0.3/getting_started/hello_robot/#gamepad-teleoperation).
 
 Now try out teleoperation (see above documentation to learn about the gamepad controls):
+
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`.
 ```bash
 python lerobot/scripts/control_robot.py \
     --robot.type=stretch \

examples/9_use_aloha.md

@@ -30,9 +30,14 @@ conda create -y -n lerobot python=3.10 && conda activate lerobot
 git clone https://github.com/huggingface/lerobot.git ~/lerobot
 ```
 
-5. Install LeRobot with dependencies for the Aloha motors (dynamixel) and cameras (intelrealsense):
+5. When using `miniconda`, install `ffmpeg` in your environment:
 ```bash
-cd ~/lerobot && pip install --no-binary=av -e ".[dynamixel, intelrealsense]"
+conda install ffmpeg -c conda-forge
+```
+
+6. Install LeRobot with dependencies for the Aloha motors (dynamixel) and cameras (intelrealsense):
+```bash
+cd ~/lerobot && pip install -e ".[dynamixel, intelrealsense]"
 ```
 
 ## Teleoperate
@@ -43,6 +48,9 @@ Teleoperation consists in manually operating the leader arms to move the followe
 2. Our code assumes that your robot has been assembled following Trossen Robotics instructions. This allows us to skip calibration, as we use the pre-defined calibration files in `.cache/calibration/aloha_default`. If you replace a motor, make sure you follow the exact instructions from Trossen Robotics.
 
 By running the following code, you can start your first **SAFE** teleoperation:
+
+> **NOTE:** To visualize the data, enable `--control.display_data=true`. This streams the data using `rerun`.
+
 ```bash
 python lerobot/scripts/control_robot.py \
   --robot.type=aloha \

lerobot/common/policies/act/modeling_act.py

@@ -142,6 +142,37 @@ class ACTPolicy(PreTrainedPolicy):
             self._action_queue.extend(actions.transpose(0, 1))
         return self._action_queue.popleft()
 
+    @torch.no_grad
+    def predict_chunk(self, batch: dict[str, Tensor]) -> Tensor:
+        """Predict a chunk of actions given environment observations.
+
+        This method returns the raw chunk of actions predicted by the model without
+        any queue management or action consumption logic.
+
+        Args:
+            batch: A dictionary of observation tensors.
+
+        Returns:
+            A tensor of shape (batch_size, chunk_size, action_dim) containing predicted actions.
+        """
+        self.eval()
+
+        batch = self.normalize_inputs(batch)
+        if self.config.image_features:
+            batch = dict(batch)  # shallow copy so that adding a key doesn't modify the original
+            batch["observation.images"] = [batch[key] for key in self.config.image_features]
+
+        # If we are using temporal ensembling
+        if self.config.temporal_ensemble_coeff is not None:
+            actions = self.model(batch)[0]  # (batch_size, chunk_size, action_dim)
+            actions = self.unnormalize_outputs({"action": actions})["action"]
+            return actions
+
+        # Standard action prediction
+        actions = self.model(batch)[0]
+        actions = self.unnormalize_outputs({"action": actions})["action"]
+        return actions
+
     def forward(self, batch: dict[str, Tensor]) -> tuple[Tensor, dict]:
         """Run the batch through the model and compute the loss for training or validation."""
         batch = self.normalize_inputs(batch)

lerobot/common/policies/pi0/modeling_pi0.py

@@ -24,7 +24,7 @@ Designed by Physical Intelligence. Ported from Jax by Hugging Face.
 
 Install pi0 extra dependencies:
 ```bash
-pip install --no-binary=av -e ".[pi0]"
+pip install -e ".[pi0]"
 ```
 
 Example of finetuning the pi0 pretrained model (`pi0_base` in `openpi`):

lerobot/common/robot_devices/cameras/intelrealsense.py

@@ -512,13 +512,13 @@ if __name__ == "__main__":
     )
     parser.add_argument(
         "--width",
-        type=str,
+        type=int,
         default=640,
         help="Set the width for all cameras. If not provided, use the default width of each camera.",
     )
     parser.add_argument(
         "--height",
-        type=str,
+        type=int,
         default=480,
         help="Set the height for all cameras. If not provided, use the default height of each camera.",
     )

lerobot/common/robot_devices/cameras/opencv.py

@@ -492,13 +492,13 @@ if __name__ == "__main__":
     )
     parser.add_argument(
         "--width",
-        type=str,
+        type=int,
         default=None,
         help="Set the width for all cameras. If not provided, use the default width of each camera.",
     )
     parser.add_argument(
         "--height",
-        type=str,
+        type=int,
         default=None,
         help="Set the height for all cameras. If not provided, use the default height of each camera.",
     )

lerobot/common/robot_devices/control_configs.py

@@ -41,7 +41,7 @@ class TeleoperateControlConfig(ControlConfig):
     fps: int | None = None
     teleop_time_s: float | None = None
     # Display all cameras on screen
-    display_cameras: bool = True
+    display_data: bool = False
 
 
 @ControlConfig.register_subclass("record")
@@ -82,7 +82,7 @@ class RecordControlConfig(ControlConfig):
     # Not enough threads might cause low camera fps.
     num_image_writer_threads_per_camera: int = 4
     # Display all cameras on screen
-    display_cameras: bool = True
+    display_data: bool = False
     # Use vocal synthesis to read events.
     play_sounds: bool = True
     # Resume recording on an existing dataset.
@@ -116,6 +116,11 @@ class ReplayControlConfig(ControlConfig):
 @dataclass
 class RemoteRobotConfig(ControlConfig):
     log_interval: int = 100
+    # Display all cameras on screen
+    display_data: bool = False
+    # Rerun configuration for remote robot (https://ref.rerun.io/docs/python/0.22.1/common/initialization_functions/#rerun.connect_tcp)
+    viewer_ip: str | None = None
+    viewer_port: str | None = None
 
 
 @dataclass

lerobot/common/robot_devices/control_utils.py

@@ -24,7 +24,7 @@ from contextlib import nullcontext
 from copy import copy
 from functools import cache
 
-import cv2
+import rerun as rr
 import torch
 from deepdiff import DeepDiff
 from termcolor import colored
@@ -174,13 +174,13 @@ def warmup_record(
     events,
     enable_teleoperation,
     warmup_time_s,
-    display_cameras,
+    display_data,
     fps,
 ):
     control_loop(
         robot=robot,
         control_time_s=warmup_time_s,
-        display_cameras=display_cameras,
+        display_data=display_data,
         events=events,
         fps=fps,
         teleoperate=enable_teleoperation,
@@ -192,7 +192,7 @@ def record_episode(
     dataset,
     events,
     episode_time_s,
-    display_cameras,
+    display_data,
     policy,
     fps,
     single_task,
@@ -200,7 +200,7 @@ def record_episode(
     control_loop(
         robot=robot,
         control_time_s=episode_time_s,
-        display_cameras=display_cameras,
+        display_data=display_data,
         dataset=dataset,
         events=events,
         policy=policy,
@@ -215,7 +215,7 @@ def control_loop(
     robot,
     control_time_s=None,
     teleoperate=False,
-    display_cameras=False,
+    display_data=False,
     dataset: LeRobotDataset | None = None,
     events=None,
     policy: PreTrainedPolicy = None,
@@ -264,11 +264,15 @@ def control_loop(
             frame = {**observation, **action, "task": single_task}
             dataset.add_frame(frame)
 
-        if display_cameras and not is_headless():
+        # TODO(Steven): This should be more general (for RemoteRobot instead of checking the name, but anyways it will change soon)
+        if (display_data and not is_headless()) or (display_data and robot.robot_type.startswith("lekiwi")):
+            for k, v in action.items():
+                for i, vv in enumerate(v):
+                    rr.log(f"sent_{k}_{i}", rr.Scalar(vv.numpy()))
+
             image_keys = [key for key in observation if "image" in key]
             for key in image_keys:
-                cv2.imshow(key, cv2.cvtColor(observation[key].numpy(), cv2.COLOR_RGB2BGR))
-            cv2.waitKey(1)
+                rr.log(key, rr.Image(observation[key].numpy()), static=True)
 
         if fps is not None:
             dt_s = time.perf_counter() - start_loop_t
@@ -297,15 +301,11 @@ def reset_environment(robot, events, reset_time_s, fps):
     )
 
 
-def stop_recording(robot, listener, display_cameras):
+def stop_recording(robot, listener, display_data):
     robot.disconnect()
 
-    if not is_headless():
-        if listener is not None:
-            listener.stop()
-
-        if display_cameras:
-            cv2.destroyAllWindows()
+    if not is_headless() and listener is not None:
+        listener.stop()
 
 
 def sanity_check_dataset_name(repo_id, policy_cfg):

lerobot/scripts/control_robot.py

@@ -135,15 +135,19 @@ python lerobot/scripts/control_robot.py \
 """
 
 import logging
+import os
 import time
 from dataclasses import asdict
 from pprint import pformat
 
+import rerun as rr
+
 # from safetensors.torch import load_file, save_file
 from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
 from lerobot.common.policies.factory import make_policy
 from lerobot.common.robot_devices.control_configs import (
     CalibrateControlConfig,
+    ControlConfig,
     ControlPipelineConfig,
     RecordControlConfig,
     RemoteRobotConfig,
@@ -153,6 +157,7 @@ from lerobot.common.robot_devices.control_configs import (
 from lerobot.common.robot_devices.control_utils import (
     control_loop,
     init_keyboard_listener,
+    is_headless,
     log_control_info,
     record_episode,
     reset_environment,
@@ -232,7 +237,7 @@ def teleoperate(robot: Robot, cfg: TeleoperateControlConfig):
         control_time_s=cfg.teleop_time_s,
         fps=cfg.fps,
         teleoperate=True,
-        display_cameras=cfg.display_cameras,
+        display_data=cfg.display_data,
     )
 
 
@@ -280,7 +285,7 @@ def record(
     # 3. place the cameras windows on screen
     enable_teleoperation = policy is None
     log_say("Warmup record", cfg.play_sounds)
-    warmup_record(robot, events, enable_teleoperation, cfg.warmup_time_s, cfg.display_cameras, cfg.fps)
+    warmup_record(robot, events, enable_teleoperation, cfg.warmup_time_s, cfg.display_data, cfg.fps)
 
     if has_method(robot, "teleop_safety_stop"):
         robot.teleop_safety_stop()
@@ -296,7 +301,7 @@ def record(
             dataset=dataset,
             events=events,
             episode_time_s=cfg.episode_time_s,
-            display_cameras=cfg.display_cameras,
+            display_data=cfg.display_data,
             policy=policy,
             fps=cfg.fps,
             single_task=cfg.single_task,
@@ -326,7 +331,7 @@ def record(
             break
 
     log_say("Stop recording", cfg.play_sounds, blocking=True)
-    stop_recording(robot, listener, cfg.display_cameras)
+    stop_recording(robot, listener, cfg.display_data)
 
     if cfg.push_to_hub:
         dataset.push_to_hub(tags=cfg.tags, private=cfg.private)
@@ -363,6 +368,40 @@ def replay(
         log_control_info(robot, dt_s, fps=cfg.fps)
 
 
+def _init_rerun(control_config: ControlConfig, session_name: str = "lerobot_control_loop") -> None:
+    """Initializes the Rerun SDK for visualizing the control loop.
+
+    Args:
+        control_config: Configuration determining data display and robot type.
+        session_name: Rerun session name. Defaults to "lerobot_control_loop".
+
+    Raises:
+        ValueError: If viewer IP is missing for non-remote configurations with display enabled.
+    """
+    if (control_config.display_data and not is_headless()) or (
+        control_config.display_data and isinstance(control_config, RemoteRobotConfig)
+    ):
+        # Configure Rerun flush batch size default to 8KB if not set
+        batch_size = os.getenv("RERUN_FLUSH_NUM_BYTES", "8000")
+        os.environ["RERUN_FLUSH_NUM_BYTES"] = batch_size
+
+        # Initialize Rerun based on configuration
+        rr.init(session_name)
+        if isinstance(control_config, RemoteRobotConfig):
+            viewer_ip = control_config.viewer_ip
+            viewer_port = control_config.viewer_port
+            if not viewer_ip or not viewer_port:
+                raise ValueError(
+                    "Viewer IP & Port are required for remote config. Set via config file/CLI or disable control_config.display_data."
+                )
+            logging.info(f"Connecting to viewer at {viewer_ip}:{viewer_port}")
+            rr.connect_tcp(f"{viewer_ip}:{viewer_port}")
+        else:
+            # Get memory limit for rerun viewer parameters
+            memory_limit = os.getenv("LEROBOT_RERUN_MEMORY_LIMIT", "10%")
+            rr.spawn(memory_limit=memory_limit)
+
+
 @parser.wrap()
 def control_robot(cfg: ControlPipelineConfig):
     init_logging()
@@ -370,17 +409,22 @@ def control_robot(cfg: ControlPipelineConfig):
 
     robot = make_robot_from_config(cfg.robot)
 
+    # TODO(Steven): Blueprint for fixed window size
+
     if isinstance(cfg.control, CalibrateControlConfig):
         calibrate(robot, cfg.control)
     elif isinstance(cfg.control, TeleoperateControlConfig):
+        _init_rerun(control_config=cfg.control, session_name="lerobot_control_loop_teleop")
         teleoperate(robot, cfg.control)
     elif isinstance(cfg.control, RecordControlConfig):
+        _init_rerun(control_config=cfg.control, session_name="lerobot_control_loop_record")
         record(robot, cfg.control)
     elif isinstance(cfg.control, ReplayControlConfig):
         replay(robot, cfg.control)
     elif isinstance(cfg.control, RemoteRobotConfig):
         from lerobot.common.robot_devices.robots.lekiwi_remote import run_lekiwi
 
+        _init_rerun(control_config=cfg.control, session_name="lerobot_control_loop_remote")
         run_lekiwi(cfg.robot)
 
     if robot.is_connected:

lerobot/scripts/visualize_dataset_html.py

@@ -174,7 +174,10 @@ def run_server(
                 dataset.meta.get_video_file_path(episode_id, key) for key in dataset.meta.video_keys
             ]
             videos_info = [
-                {"url": url_for("static", filename=video_path), "filename": video_path.parent.name}
+                {
+                    "url": url_for("static", filename=str(video_path).replace("\\", "/")),
+                    "filename": video_path.parent.name,
+                }
                 for video_path in video_paths
             ]
             tasks = dataset.meta.episodes[episode_id]["tasks"]
@@ -381,7 +384,7 @@ def visualize_dataset_html(
         if isinstance(dataset, LeRobotDataset):
             ln_videos_dir = static_dir / "videos"
             if not ln_videos_dir.exists():
-                ln_videos_dir.symlink_to((dataset.root / "videos").resolve())
+                ln_videos_dir.symlink_to((dataset.root / "videos").resolve().as_posix())
 
         if serve:
             run_server(dataset, episodes, host, port, static_dir, template_dir)

pyproject.toml

@@ -60,9 +60,9 @@ dependencies = [
     "jsonlines>=4.0.0",
     "numba>=0.59.0",
     "omegaconf>=2.3.0",
-    "opencv-python>=4.9.0",
+    "opencv-python-headless>=4.9.0",
     "packaging>=24.2",
-    "av>=12.0.5,<13.0.0",
+    "av>=12.0.5",
     "pymunk>=6.6.0",
     "pynput>=1.7.7",
     "pyzmq>=26.2.1",

tests/robots/test_control_robot.py

@@ -172,8 +172,7 @@ def test_record_and_replay_and_policy(tmp_path, request, robot_type, mock):
         push_to_hub=False,
         # TODO(rcadene, aliberts): test video=True
         video=False,
-        # TODO(rcadene): display cameras through cv2 sometimes crashes on mac
-        display_cameras=False,
+        display_data=False,
         play_sounds=False,
     )
     dataset = record(robot, rec_cfg)
@@ -226,7 +225,7 @@ def test_record_and_replay_and_policy(tmp_path, request, robot_type, mock):
         num_episodes=2,
         push_to_hub=False,
         video=False,
-        display_cameras=False,
+        display_data=False,
         play_sounds=False,
         num_image_writer_processes=num_image_writer_processes,
     )
@@ -273,7 +272,7 @@ def test_resume_record(tmp_path, request, robot_type, mock):
         episode_time_s=1,
         push_to_hub=False,
         video=False,
-        display_cameras=False,
+        display_data=False,
         play_sounds=False,
         num_episodes=1,
     )
@@ -330,7 +329,7 @@ def test_record_with_event_rerecord_episode(tmp_path, request, robot_type, mock)
             num_episodes=1,
             push_to_hub=False,
             video=False,
-            display_cameras=False,
+            display_data=False,
             play_sounds=False,
         )
         dataset = record(robot, rec_cfg)
@@ -380,7 +379,7 @@ def test_record_with_event_exit_early(tmp_path, request, robot_type, mock):
             num_episodes=1,
             push_to_hub=False,
             video=False,
-            display_cameras=False,
+            display_data=False,
             play_sounds=False,
         )
 
@@ -433,7 +432,7 @@ def test_record_with_event_stop_recording(tmp_path, request, robot_type, mock, n
             num_episodes=2,
             push_to_hub=False,
             video=False,
-            display_cameras=False,
+            display_data=False,
             play_sounds=False,
             num_image_writer_processes=num_image_writer_processes,
         )