fix(ci): create venv for release testing (#1652 )

fix(ci): remove uv run + bump minor (#1651 )
chore(ci): Bump to v0.3.0 (#1649 )
2026-05-15 08:39:49 +00:00 · 2025-08-01 21:04:47 +02:00 · 2025-08-01 20:52:10 +02:00 · 2025-08-01 18:30:33 +02:00 · 2025-08-01 18:21:37 +02:00 · 2025-08-01 18:11:08 +02:00
37 changed files with 1646 additions and 239 deletions
@@ -19,6 +19,11 @@ on:
    tags:
      - 'v*.*.*' # Trigger on tags like v0.1.0, v1.0.0

+# Sets up the environment variables
+env:
+  UV_VERSION: "0.8.0"
+  PYTHON_VERSION: "3.10"
+
 jobs:
  # This job builds the Python package and publishes it to PyPI
  build-and-publish:
@@ -50,6 +55,7 @@ jobs:
          VERSION_NUMBER=${VERSION#v}
          echo "tag_version=$VERSION_NUMBER" >> $GITHUB_OUTPUT
      - name: Check if version matches pyproject.toml
+        if: startsWith(github.ref, 'refs/tags/v') && !contains(github.ref, '-')
        # zizmor: ignore[template-injection]
        run: |
          TAG_VERSION=${{ steps.extract_info.outputs.tag_version }}
@@ -86,13 +92,29 @@ jobs:
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        # zizmor: ignore[template-injection]
-        run: gh release create ${{ github.ref_name }} --release-name "Release ${{ github.ref_name }}" --generate-notes ./dist/*
+        run: |
+          gh release create ${{ github.ref_name }} \
+            --title "Release ${{ github.ref_name }}" \
+            --generate-notes \
+            --draft=$([[ "${{ github.ref_name }}" == *-* ]] && echo true || echo false) \
+            --prerelease=$([[ "${{ github.ref_name }}" == *-* ]] && echo true || echo false) \
+            ./dist/*

-      - name: Publish to PyPI
-        if: startsWith(github.ref, 'refs/tags/v')
+      - name: Publish to TestPyPI for pre-releases
+        # True for tags like 'v0.2.0-rc1'
+        if: startsWith(github.ref, 'refs/tags/v') && contains(github.ref, '-')
        uses: pypa/gh-action-pypi-publish@v1.12.4 # zizmor: ignore[unpinned-uses, use-trusted-publishing]
        with:
-          password: ${{ secrets.PYPI_API_TOKEN }}
+          repository-url: https://test.pypi.org/legacy/
+          verbose: true
+          print-hash: true
+
+      - name: Publish to PyPI
+        if: startsWith(github.ref, 'refs/tags/v') && !contains(github.ref, '-')
+        uses: pypa/gh-action-pypi-publish@v1.12.4 # zizmor: ignore[unpinned-uses, use-trusted-publishing]
+        with:
+          verbose: true
+          print-hash: true

  # This job runs end-to-end tests on the release
  test-release:
@@ -119,15 +141,31 @@ jobs:
          enable-cache: true
          version: ${{ env.UV_VERSION }}
          python-version: ${{ env.PYTHON_VERSION }}
+      - name: Create uv virtual environment
+        run: uv venv
      - name: Install lerobot release
-        run: uv run pip install lerobot==${{ needs.build-and-publish.outputs.version }} # zizmor: ignore[template-injection]
-
+        # zizmor: ignore[template-injection]
+        run: |
+          VERSION="${{ needs.build-and-publish.outputs.version }}"
+          if [[ "$VERSION" == *-* ]]; then
+            BASE_VERSION="${VERSION%%-*}"
+            echo "Installing pre-release version $BASE_VERSION from TestPyPI..."
+            uv pip install \
+              --index-url https://test.pypi.org/simple/ \
+              --extra-index-url https://pypi.org/simple \
+              --index-strategy unsafe-best-match \
+               "lerobot[all]==$BASE_VERSION"
+          else
+            echo "Installing release version $VERSION from PyPI..."
+            uv pip install "lerobot[all]==$VERSION"
+          fi
      - name: Check lerobot version
-        run: uv run lerobot --version
+        run: uv run python -c "import lerobot; print(lerobot.__version__)"

      - name: Run end-to-end tests
        run: uv run make test-end-to-end


-# TODO(Steven): Publish draft/pre-release and to test pypi
+# TODO(Steven): Publish draft/pre-release and to test pypi weekly
+# TODO(Steven): Separate build and publish job
 # TODO(Steven): Tag documentation with the same version as the package
@@ -1,25 +1,21 @@
 <p align="center">
-  <picture>
-    <source media="(prefers-color-scheme: dark)" srcset="media/lerobot-logo-thumbnail.png">
-    <source media="(prefers-color-scheme: light)" srcset="media/lerobot-logo-thumbnail.png">
-    <img alt="LeRobot, Hugging Face Robotics Library" src="media/lerobot-logo-thumbnail.png" style="max-width: 100%;">
-  </picture>
+  <img alt="LeRobot, Hugging Face Robotics Library" src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/lerobot-logo-thumbnail.png" width="100%">
  <br/>
  <br/>
 </p>

 <div align="center">

-[![Tests](https://github.com/huggingface/lerobot/actions/workflows/nightly-tests.yml/badge.svg?branch=main)](https://github.com/huggingface/lerobot/actions/workflows/nightly-tests.yml?query=branch%3Amain)
-[![Coverage](https://codecov.io/gh/huggingface/lerobot/branch/main/graph/badge.svg?token=TODO)](https://codecov.io/gh/huggingface/lerobot)
+[![Tests](https://github.com/huggingface/lerobot/actions/workflows/nightly.yml/badge.svg?branch=main)](https://github.com/huggingface/lerobot/actions/workflows/nighty.yml?query=branch%3Amain)
 [![Python versions](https://img.shields.io/pypi/pyversions/lerobot)](https://www.python.org/downloads/)
 [![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://github.com/huggingface/lerobot/blob/main/LICENSE)
 [![Status](https://img.shields.io/pypi/status/lerobot)](https://pypi.org/project/lerobot/)
 [![Version](https://img.shields.io/pypi/v/lerobot)](https://pypi.org/project/lerobot/)
-[![Examples](https://img.shields.io/badge/Examples-green.svg)](https://github.com/huggingface/lerobot/tree/main/examples)
-[![Contributor Covenant](https://img.shields.io/badge/Contributor%20Covenant-v2.1%20adopted-ff69b4.svg)](https://github.com/huggingface/lerobot/blob/main/CODE_OF_CONDUCT.md)
+[![Contributor Covenant](https://img.shields.io/badge/Contributor%20Covenant-v2.1-ff69b4.svg)](https://github.com/huggingface/lerobot/blob/main/CODE_OF_CONDUCT.md)
 [![Discord](https://dcbadge.vercel.app/api/server/C5P34WJ68S?style=flat)](https://discord.gg/s3KuuzsPFb)

+<!-- [![Coverage](https://codecov.io/gh/huggingface/lerobot/branch/main/graph/badge.svg?token=TODO)](https://codecov.io/gh/huggingface/lerobot) -->
+
 </div>

 <h2 align="center">
@@ -29,10 +25,10 @@

 <div align="center">
  <img
-    src="media/hope_jr/hopejr.png?raw=true"
+    src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/hope_jr/hopejr.png"
    alt="HopeJR robot"
    title="HopeJR robot"
-    style="width: 60%;"
+    width="60%"
  />

  <p><strong>Meet HopeJR – A humanoid robot arm and hand for dexterous manipulation!</strong></p>
@@ -51,20 +47,12 @@
 </h2>

 <div align="center">
-  <div style="display: flex; gap: 1rem; justify-content: center; align-items: center;" >
-    <img
-      src="media/so101/so101.webp?raw=true"
-      alt="SO-101 follower arm"
-      title="SO-101 follower arm"
-      style="width: 40%;"
-    />
-    <img
-      src="media/so101/so101-leader.webp?raw=true"
-      alt="SO-101 leader arm"
-      title="SO-101 leader arm"
-      style="width: 40%;"
-    />
-  </div>
+  <table>
+    <tr>
+      <td align="center"><img src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/so101/so101.webp" alt="SO-101 follower arm" title="SO-101 follower arm" width="90%"/></td>
+      <td align="center"><img src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/so101/so101-leader.webp" alt="SO-101 leader arm" title="SO-101 leader arm" width="90%"/></td>
+    </tr>
+  </table>

  <p><strong>Meet the updated SO100, the SO-101 – Just €114 per arm!</strong></p>
  <p>Train it in minutes with a few simple moves on your laptop.</p>
@@ -76,7 +64,7 @@
  <p>Want to take it to the next level? Make your SO-101 mobile by building LeKiwi!</p>
  <p>Check out the <a href="https://huggingface.co/docs/lerobot/lekiwi">LeKiwi tutorial</a> and bring your robot to life on wheels.</p>

-  <img src="media/lekiwi/kiwi.webp?raw=true" alt="LeKiwi mobile robot" title="LeKiwi mobile robot" width="50%">
+  <img src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/lekiwi/kiwi.webp" alt="LeKiwi mobile robot" title="LeKiwi mobile robot" width="50%">
 </div>

 <br/>
@@ -99,9 +87,9 @@

 <table>
  <tr>
-    <td><img src="media/gym/aloha_act.gif" width="100%" alt="ACT policy on ALOHA env"/></td>
-    <td><img src="media/gym/simxarm_tdmpc.gif" width="100%" alt="TDMPC policy on SimXArm env"/></td>
-    <td><img src="media/gym/pusht_diffusion.gif" width="100%" alt="Diffusion policy on PushT env"/></td>
+    <td><img src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/gym/aloha_act.gif" width="100%" alt="ACT policy on ALOHA env"/></td>
+    <td><img src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/gym/simxarm_tdmpc.gif" width="100%" alt="TDMPC policy on SimXArm env"/></td>
+    <td><img src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/gym/pusht_diffusion.gif" width="100%" alt="Diffusion policy on PushT env"/></td>
  </tr>
  <tr>
    <td align="center">ACT policy on ALOHA env</td>
@@ -110,24 +98,9 @@
  </tr>
 </table>

-### Acknowledgment
-
- The LeRobot team 🤗 for building SmolVLA [Paper](https://arxiv.org/abs/2506.01844), [Blog](https://huggingface.co/blog/smolvla).
- Thanks to Tony Zhao, Zipeng Fu and colleagues for open sourcing ACT policy, ALOHA environments and datasets. Ours are adapted from [ALOHA](https://tonyzhaozh.github.io/aloha) and [Mobile ALOHA](https://mobile-aloha.github.io).
- Thanks to Cheng Chi, Zhenjia Xu and colleagues for open sourcing Diffusion policy, Pusht environment and datasets, as well as UMI datasets. Ours are adapted from [Diffusion Policy](https://diffusion-policy.cs.columbia.edu) and [UMI Gripper](https://umi-gripper.github.io).
- Thanks to Nicklas Hansen, Yunhai Feng and colleagues for open sourcing TDMPC policy, Simxarm environments and datasets. Ours are adapted from [TDMPC](https://github.com/nicklashansen/tdmpc) and [FOWM](https://www.yunhaifeng.com/FOWM).
- Thanks to Antonio Loquercio and Ashish Kumar for their early support.
- Thanks to [Seungjae (Jay) Lee](https://sjlee.cc/), [Mahi Shafiullah](https://mahis.life/) and colleagues for open sourcing [VQ-BeT](https://sjlee.cc/vq-bet/) policy and helping us adapt the codebase to our repository. The policy is adapted from [VQ-BeT repo](https://github.com/jayLEE0301/vq_bet_official).
-
 ## Installation

-Download our source code:
-
-```bash
-git clone https://github.com/huggingface/lerobot.git
-cd lerobot
-```
-
+LeRobot works with Python 3.10+ and PyTorch 2.2+.
 Create a virtual environment with Python 3.10 and activate it, e.g. with [`miniconda`](https://docs.anaconda.com/free/miniconda/index.html):

 ```bash
@@ -154,7 +127,7 @@ conda install ffmpeg -c conda-forge
 Install 🤗 LeRobot:

 ```bash
-pip install -e .
+pip install lerobot
 ```

 > **NOTE:** If you encounter build errors, you may need to install additional dependencies (`cmake`, `build-essential`, and `ffmpeg libs`). On Linux, run:
@@ -182,7 +155,7 @@ wandb login

 ### Visualize datasets

-Check out [example 1](./examples/1_load_lerobot_dataset.py) that illustrates how to use our dataset class which automatically downloads data from the Hugging Face hub.
+Check out [example 1](https://github.com/huggingface/lerobot/blob/main/examples/1_load_lerobot_dataset.py) that illustrates how to use our dataset class which automatically downloads data from the Hugging Face hub.

 You can also locally visualize episodes from a dataset on the hub by executing our script from the command line:

@@ -212,7 +185,7 @@ Our script can also visualize datasets stored on a distant server. See `python -

 A dataset in `LeRobotDataset` format is very simple to use. It can be loaded from a repository on the Hugging Face hub or a local folder simply with e.g. `dataset = LeRobotDataset("lerobot/aloha_static_coffee")` and can be indexed into like any Hugging Face and PyTorch dataset. For instance `dataset[0]` will retrieve a single temporal frame from the dataset containing observation(s) and an action as PyTorch tensors ready to be fed to a model.

-A specificity of `LeRobotDataset` is that, rather than retrieving a single frame by its index, we can retrieve several frames based on their temporal relationship with the indexed frame, by setting `delta_timestamps` to a list of relative times with respect to the indexed frame. For example, with `delta_timestamps = {"observation.image": [-1, -0.5, -0.2, 0]}` one can retrieve, for a given index, 4 frames: 3 "previous" frames 1 second, 0.5 seconds, and 0.2 seconds before the indexed frame, and the indexed frame itself (corresponding to the 0 entry). See example [1_load_lerobot_dataset.py](examples/1_load_lerobot_dataset.py) for more details on `delta_timestamps`.
+A specificity of `LeRobotDataset` is that, rather than retrieving a single frame by its index, we can retrieve several frames based on their temporal relationship with the indexed frame, by setting `delta_timestamps` to a list of relative times with respect to the indexed frame. For example, with `delta_timestamps = {"observation.image": [-1, -0.5, -0.2, 0]}` one can retrieve, for a given index, 4 frames: 3 "previous" frames 1 second, 0.5 seconds, and 0.2 seconds before the indexed frame, and the indexed frame itself (corresponding to the 0 entry). See example [1_load_lerobot_dataset.py](https://github.com/huggingface/lerobot/blob/main/examples/1_load_lerobot_dataset.py) for more details on `delta_timestamps`.

 Under the hood, the `LeRobotDataset` format makes use of several ways to serialize data which can be useful to understand if you plan to work more closely with this format. We tried to make a flexible yet simple dataset format that would cover most type of features and specificities present in reinforcement learning and robotics, in simulation and in real-world, with a focus on cameras and robot states but easily extended to other types of sensory inputs as long as they can be represented by a tensor.

@@ -256,7 +229,7 @@ Dataset can be uploaded/downloaded from the HuggingFace hub seamlessly. To work

 ### Evaluate a pretrained policy

-Check out [example 2](./examples/2_evaluate_pretrained_policy.py) that illustrates how to download a pretrained policy from Hugging Face hub, and run an evaluation on its corresponding environment.
+Check out [example 2](https://github.com/huggingface/lerobot/blob/main/examples/2_evaluate_pretrained_policy.py) that illustrates how to download a pretrained policy from Hugging Face hub, and run an evaluation on its corresponding environment.

 We also provide a more capable script to parallelize the evaluation over multiple environments during the same rollout. Here is an example with a pretrained model hosted on [lerobot/diffusion_pusht](https://huggingface.co/lerobot/diffusion_pusht):

@@ -280,13 +253,13 @@ See `python -m lerobot.scripts.eval --help` for more instructions.

 ### Train your own policy

-Check out [example 3](./examples/3_train_policy.py) that illustrates how to train a model using our core library in python, and [example 4](./examples/4_train_policy_with_script.md) that shows how to use our training script from command line.
+Check out [example 3](https://github.com/huggingface/lerobot/blob/main/examples/3_train_policy.py) that illustrates how to train a model using our core library in python, and [example 4](https://github.com/huggingface/lerobot/blob/main/examples/4_train_policy_with_script.md) that shows how to use our training script from command line.

 To use wandb for logging training and evaluation curves, make sure you've run `wandb login` as a one-time setup step. Then, when running the training command above, enable WandB in the configuration by adding `--wandb.enable=true`.

-A link to the wandb logs for the run will also show up in yellow in your terminal. Here is an example of what they look like in your browser. Please also check [here](./examples/4_train_policy_with_script.md#typical-logs-and-metrics) for the explanation of some commonly used metrics in logs.
+A link to the wandb logs for the run will also show up in yellow in your terminal. Here is an example of what they look like in your browser. Please also check [here](https://github.com/huggingface/lerobot/blob/main/examples/4_train_policy_with_script.md#typical-logs-and-metrics) for the explanation of some commonly used metrics in logs.

-![](media/wandb.png)
+\<img src="https://raw.githubusercontent.com/huggingface/lerobot/main/media/wandb.png" alt="WandB logs example"\>

 Note: For efficiency, during training every checkpoint is evaluated on a low number of episodes. You may use `--eval.n_episodes=500` to evaluate on more episodes than the default. Or, after training, you may want to re-evaluate your best checkpoints on more episodes or change the evaluation settings. See `python -m lerobot.scripts.eval --help` for more instructions.

@@ -305,26 +278,6 @@ reproduces SOTA results for Diffusion Policy on the PushT task.

 If you would like to contribute to 🤗 LeRobot, please check out our [contribution guide](https://github.com/huggingface/lerobot/blob/main/CONTRIBUTING.md).

-<!-- ### Add a new dataset
-
-To add a dataset to the hub, you need to login using a write-access token, which can be generated from the [Hugging Face settings](https://huggingface.co/settings/tokens):
-```bash
-huggingface-cli login --token ${HUGGINGFACE_TOKEN} --add-to-git-credential
-```
-
-Then point to your raw dataset folder (e.g. `data/aloha_static_pingpong_test_raw`), and push your dataset to the hub with:
-```bash
-python lerobot/scripts/push_dataset_to_hub.py \
--raw-dir data/aloha_static_pingpong_test_raw \
--out-dir data \
--repo-id lerobot/aloha_static_pingpong_test \
--raw-format aloha_hdf5
-```
-
-See `python lerobot/scripts/push_dataset_to_hub.py --help` for more instructions.
-
-If your dataset format is not supported, implement your own in `lerobot/datasets/push_dataset_to_hub/${raw_format}_format.py` by copying examples like [pusht_zarr](https://github.com/huggingface/lerobot/blob/main/lerobot/datasets/push_dataset_to_hub/pusht_zarr_format.py), [umi_zarr](https://github.com/huggingface/lerobot/blob/main/lerobot/datasets/push_dataset_to_hub/umi_zarr_format.py), [aloha_hdf5](https://github.com/huggingface/lerobot/blob/main/lerobot/datasets/push_dataset_to_hub/aloha_hdf5_format.py), or [xarm_pkl](https://github.com/huggingface/lerobot/blob/main/lerobot/datasets/push_dataset_to_hub/xarm_pkl_format.py). -->
-
 ### Add a pretrained policy

 Once you have trained a policy you may upload it to the Hugging Face hub using a hub id that looks like `${hf_user}/${repo_name}` (e.g. [lerobot/diffusion_pusht](https://huggingface.co/lerobot/diffusion_pusht)).
@@ -341,34 +294,16 @@ To upload these to the hub, run the following:
 huggingface-cli upload ${hf_user}/${repo_name} path/to/pretrained_model
 ```

-See [eval.py](https://github.com/huggingface/lerobot/blob/main/lerobot/scripts/eval.py) for an example of how other people may use your policy.
+See [eval.py](https://github.com/huggingface/lerobot/blob/main/src/lerobot/scripts/eval.py) for an example of how other people may use your policy.

-### Improve your code with profiling
+### Acknowledgment

-An example of a code snippet to profile the evaluation of a policy:
-
-<!-- prettier-ignore-start -->
-```python
-from torch.profiler import profile, record_function, ProfilerActivity
-
-def trace_handler(prof):
-    prof.export_chrome_trace(f"tmp/trace_schedule_{prof.step_num}.json")
-
-with profile(
-    activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
-    schedule=torch.profiler.schedule(
-        wait=2,
-        warmup=2,
-        active=3,
-    ),
-    on_trace_ready=trace_handler
-) as prof:
-    with record_function("eval_policy"):
-        for i in range(num_episodes):
-            prof.step()
-            # insert code to profile, potentially whole body of eval_policy function
-```
-<!-- prettier-ignore-end -->
+- The LeRobot team 🤗 for building SmolVLA [Paper](https://arxiv.org/abs/2506.01844), [Blog](https://huggingface.co/blog/smolvla).
+- Thanks to Tony Zhao, Zipeng Fu and colleagues for open sourcing ACT policy, ALOHA environments and datasets. Ours are adapted from [ALOHA](https://tonyzhaozh.github.io/aloha) and [Mobile ALOHA](https://mobile-aloha.github.io).
+- Thanks to Cheng Chi, Zhenjia Xu and colleagues for open sourcing Diffusion policy, Pusht environment and datasets, as well as UMI datasets. Ours are adapted from [Diffusion Policy](https://diffusion-policy.cs.columbia.edu) and [UMI Gripper](https://umi-gripper.github.io).
+- Thanks to Nicklas Hansen, Yunhai Feng and colleagues for open sourcing TDMPC policy, Simxarm environments and datasets. Ours are adapted from [TDMPC](https://github.com/nicklashansen/tdmpc) and [FOWM](https://www.yunhaifeng.com/FOWM).
+- Thanks to Antonio Loquercio and Ashish Kumar for their early support.
+- Thanks to [Seungjae (Jay) Lee](https://sjlee.cc/), [Mahi Shafiullah](https://mahis.life/) and colleagues for open sourcing [VQ-BeT](https://sjlee.cc/vq-bet/) policy and helping us adapt the codebase to our repository. The policy is adapted from [VQ-BeT repo](https://github.com/jayLEE0301/vq_bet_official).

 ## Citation

@@ -383,76 +318,6 @@ If you want, you can cite this work with:
 }
 ```

-Additionally, if you are using any of the particular policy architecture, pretrained models, or datasets, it is recommended to cite the original authors of the work as they appear below:
-
- [SmolVLA](https://arxiv.org/abs/2506.01844)
-
-```bibtex
-@article{shukor2025smolvla,
-  title={SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics},
-  author={Shukor, Mustafa and Aubakirova, Dana and Capuano, Francesco and Kooijmans, Pepijn and Palma, Steven and Zouitine, Adil and Aractingi, Michel and Pascal, Caroline and Russi, Martino and Marafioti, Andres and Alibert, Simon and Cord, Matthieu and Wolf, Thomas and Cadene, Remi},
-  journal={arXiv preprint arXiv:2506.01844},
-  year={2025}
-}
-```
-
- [Diffusion Policy](https://diffusion-policy.cs.columbia.edu)
-
-```bibtex
-@article{chi2024diffusionpolicy,
-	author = {Cheng Chi and Zhenjia Xu and Siyuan Feng and Eric Cousineau and Yilun Du and Benjamin Burchfiel and Russ Tedrake and Shuran Song},
-	title ={Diffusion Policy: Visuomotor Policy Learning via Action Diffusion},
-	journal = {The International Journal of Robotics Research},
-	year = {2024},
-}
-```
-
- [ACT or ALOHA](https://tonyzhaozh.github.io/aloha)
-
-```bibtex
-@article{zhao2023learning,
-  title={Learning fine-grained bimanual manipulation with low-cost hardware},
-  author={Zhao, Tony Z and Kumar, Vikash and Levine, Sergey and Finn, Chelsea},
-  journal={arXiv preprint arXiv:2304.13705},
-  year={2023}
-}
-```
-
- [TDMPC](https://www.nicklashansen.com/td-mpc/)
-
-```bibtex
-@inproceedings{Hansen2022tdmpc,
-	title={Temporal Difference Learning for Model Predictive Control},
-	author={Nicklas Hansen and Xiaolong Wang and Hao Su},
-	booktitle={ICML},
-	year={2022}
-}
-```
-
- [VQ-BeT](https://sjlee.cc/vq-bet/)
-
-```bibtex
-@article{lee2024behavior,
-  title={Behavior generation with latent actions},
-  author={Lee, Seungjae and Wang, Yibin and Etukuru, Haritheja and Kim, H Jin and Shafiullah, Nur Muhammad Mahi and Pinto, Lerrel},
-  journal={arXiv preprint arXiv:2403.03181},
-  year={2024}
-}
-```
-
- [HIL-SERL](https://hil-serl.github.io/)
-
-```bibtex
-@Article{luo2024hilserl,
-title={Precise and Dexterous Robotic Manipulation via Human-in-the-Loop Reinforcement Learning},
-author={Jianlan Luo and Charles Xu and Jeffrey Wu and Sergey Levine},
-year={2024},
-eprint={2410.21845},
-archivePrefix={arXiv},
-primaryClass={cs.RO}
-}
-```
-
 ## Star History

 [![Star History Chart](https://api.star-history.com/svg?repos=huggingface/lerobot&type=Timeline)](https://star-history.com/#huggingface/lerobot&Timeline)
@@ -0,0 +1,3 @@
+# docs-requirements.txt
+hf-doc-builder @ git+https://github.com/huggingface/doc-builder.git@main
+watchdog>=6.0.0
@@ -20,7 +20,7 @@ To generate the documentation, you first have to build it. Several packages are
 you can install them with the following command, at the root of the code repository:

 ```bash
-pip install -e ".[docs]"
+pip install -e . -r docs-requirements.txt
 ```

 You will also need `nodejs`. Please refer to their [installation page](https://nodejs.org/en/download)
@@ -462,9 +462,9 @@ If you do not want to push your model to the hub after training use `--policy.pu

 Additionally you can provide extra `tags` or specify a `license` for your model or make the model repo `private` by adding this: `--policy.private=true --policy.tags=\[ppo,rl\] --policy.license=mit`

-#### Train using Collab
+#### Train using Google Colab

-If your local computer doesn't have a powerful GPU you could utilize Google Collab to train your model by following the [ACT training notebook](./notebooks#training-act).
+If your local computer doesn't have a powerful GPU you could utilize Google Colab to train your model by following the [ACT training notebook](./notebooks#training-act).

 #### Upload policy checkpoints

@@ -0,0 +1,14 @@
+## Paper
+
+https://tonyzhaozh.github.io/aloha
+
+## Citation
+
+```bibtex
+@article{zhao2023learning,
+  title={Learning fine-grained bimanual manipulation with low-cost hardware},
+  author={Zhao, Tony Z and Kumar, Vikash and Levine, Sergey and Finn, Chelsea},
+  journal={arXiv preprint arXiv:2304.13705},
+  year={2023}
+}
+```
@@ -0,0 +1,14 @@
+## Paper
+
+https://diffusion-policy.cs.columbia.edu
+
+## Citation
+
+```bibtex
+@article{chi2024diffusionpolicy,
+	author = {Cheng Chi and Zhenjia Xu and Siyuan Feng and Eric Cousineau and Yilun Du and Benjamin Burchfiel and Russ Tedrake and Shuran Song},
+	title ={Diffusion Policy: Visuomotor Policy Learning via Action Diffusion},
+	journal = {The International Journal of Robotics Research},
+	year = {2024},
+}
+```
@@ -0,0 +1,14 @@
+## Paper
+
+https://arxiv.org/abs/2506.01844
+
+## Citation
+
+```bibtex
+@article{shukor2025smolvla,
+  title={SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics},
+  author={Shukor, Mustafa and Aubakirova, Dana and Capuano, Francesco and Kooijmans, Pepijn and Palma, Steven and Zouitine, Adil and Aractingi, Michel and Pascal, Caroline and Russi, Martino and Marafioti, Andres and Alibert, Simon and Cord, Matthieu and Wolf, Thomas and Cadene, Remi},
+  journal={arXiv preprint arXiv:2506.01844},
+  year={2025}
+}
+```
@@ -0,0 +1,14 @@
+## Paper
+
+https://www.nicklashansen.com/td-mpc/
+
+## Citation
+
+```bibtex
+@inproceedings{Hansen2022tdmpc,
+	title={Temporal Difference Learning for Model Predictive Control},
+	author={Nicklas Hansen and Xiaolong Wang and Hao Su},
+	booktitle={ICML},
+	year={2022}
+}
+```
@@ -0,0 +1,14 @@
+## Paper
+
+https://sjlee.cc/vq-bet/
+
+## Citation
+
+```bibtex
+@article{lee2024behavior,
+  title={Behavior generation with latent actions},
+  author={Lee, Seungjae and Wang, Yibin and Etukuru, Haritheja and Kim, H Jin and Shafiullah, Nur Muhammad Mahi and Pinto, Lerrel},
+  journal={arXiv preprint arXiv:2403.03181},
+  year={2024}
+}
+```
@@ -25,7 +25,7 @@ discord = "https://discord.gg/s3KuuzsPFb"

 [project]
 name = "lerobot"
-version = "0.2.0"
+version = "0.3.2"
 description = "🤗 LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch"
 readme = "README.md"
 license = { text = "Apache-2.0" }
@@ -61,7 +61,7 @@ dependencies = [
    # Hugging Face dependencies
    "datasets>=2.19.0,<=3.6.0", # TODO: Bumb dependency
    "diffusers>=0.27.2",
-    "huggingface-hub[hf-transfer,cli]>=0.27.1,<0.34.0",
+    "huggingface-hub[hf-transfer,cli]>=0.34.2",

    # Core dependencies
    "cmake>=3.29.0.1",
@@ -125,7 +125,6 @@ hilserl = ["lerobot[transformers-dep]", "gym-hil>=0.1.9", "lerobot[grpcio-dep]",
 async = ["lerobot[grpcio-dep]", "matplotlib>=3.10.3"]

 # Development
-docs = ["hf-doc-builder @ git+https://github.com/huggingface/doc-builder.git@main", "watchdog >= 6.0.0"]
 dev = ["pre-commit>=3.7.0", "debugpy>=1.8.1", "lerobot[grpcio-dep]", "grpcio-tools==1.73.1"]
 test = ["pytest>=8.1.0", "pytest-timeout>=2.4.0", "pytest-cov>=5.0.0", "mock-serial>=0.0.1 ; sys_platform != 'win32'"]
 video_benchmark = ["scikit-image>=0.23.2", "pandas>=2.2.2"]
@@ -147,7 +146,6 @@ all = [
    "lerobot[smolvla]",
    "lerobot[hilserl]",
    "lerobot[async]",
-    "lerobot[docs]",
    "lerobot[dev]",
    "lerobot[test]",
    "lerobot[video_benchmark]",
@@ -0,0 +1,625 @@
+# This file is autogenerated by pip-compile with Python 3.10
+# by the following command:
+#
+#    pip-compile --output-file=requirements-macos.txt requirements.in
+#
+-e .[all]
+    # via -[all]
+absl-py==2.3.1
+    # via
+    #   dm-control
+    #   dm-env
+    #   dm-tree
+    #   labmaze
+    #   mujoco
+accelerate==1.9.0
+    # via lerobot
+aiohappyeyeballs==2.6.1
+    # via aiohttp
+aiohttp==3.12.15
+    # via fsspec
+aiosignal==1.4.0
+    # via aiohttp
+annotated-types==0.7.0
+    # via pydantic
+asttokens==3.0.0
+    # via stack-data
+async-timeout==5.0.1
+    # via aiohttp
+attrs==25.3.0
+    # via
+    #   aiohttp
+    #   dm-tree
+    #   jsonlines
+    #   rerun-sdk
+av==15.0.0
+    # via lerobot
+blinker==1.9.0
+    # via flask
+certifi==2025.7.14
+    # via
+    #   requests
+    #   sentry-sdk
+cffi==1.17.1
+    # via pymunk
+cfgv==3.4.0
+    # via pre-commit
+charset-normalizer==3.4.2
+    # via requests
+click==8.2.1
+    # via
+    #   flask
+    #   wandb
+cloudpickle==3.1.1
+    # via gymnasium
+cmake==4.0.3
+    # via lerobot
+cmeel==0.57.3
+    # via
+    #   cmeel-assimp
+    #   cmeel-boost
+    #   cmeel-console-bridge
+    #   cmeel-octomap
+    #   cmeel-qhull
+    #   cmeel-tinyxml2
+    #   cmeel-urdfdom
+    #   cmeel-zlib
+    #   coal-library
+    #   eigenpy
+    #   eiquadprog
+    #   pin
+    #   placo
+    #   rhoban-cmeel-jsoncpp
+cmeel-assimp==5.4.3.1
+    # via coal-library
+cmeel-boost==1.87.0.1
+    # via
+    #   coal-library
+    #   eigenpy
+    #   eiquadprog
+    #   pin
+cmeel-console-bridge==1.0.2.3
+    # via cmeel-urdfdom
+cmeel-octomap==1.10.0
+    # via coal-library
+cmeel-qhull==8.0.2.1
+    # via coal-library
+cmeel-tinyxml2==10.0.0
+    # via cmeel-urdfdom
+cmeel-urdfdom==4.0.1
+    # via pin
+cmeel-zlib==1.3.1
+    # via cmeel-assimp
+coal-library==3.0.1
+    # via pin
+contourpy==1.3.2
+    # via matplotlib
+coverage[toml]==7.10.1
+    # via pytest-cov
+cycler==0.12.1
+    # via matplotlib
+datasets==3.6.0
+    # via lerobot
+debugpy==1.8.15
+    # via lerobot
+decorator==5.2.1
+    # via ipython
+deepdiff==8.5.0
+    # via lerobot
+diffusers==0.34.0
+    # via lerobot
+dill==0.3.8
+    # via
+    #   datasets
+    #   multiprocess
+distlib==0.4.0
+    # via virtualenv
+dm-control==1.0.14
+    # via gym-aloha
+dm-env==1.6
+    # via dm-control
+dm-tree==0.1.9
+    # via
+    #   dm-control
+    #   dm-env
+docopt==0.6.2
+    # via num2words
+draccus==0.10.0
+    # via lerobot
+dynamixel-sdk==3.7.31
+    # via lerobot
+eigenpy==3.10.3
+    # via coal-library
+einops==0.8.1
+    # via lerobot
+eiquadprog==1.2.9
+    # via placo
+exceptiongroup==1.3.0
+    # via
+    #   ipython
+    #   pytest
+executing==2.2.0
+    # via stack-data
+farama-notifications==0.0.4
+    # via gymnasium
+feetech-servo-sdk==1.0.0
+    # via lerobot
+filelock==3.18.0
+    # via
+    #   datasets
+    #   diffusers
+    #   huggingface-hub
+    #   torch
+    #   transformers
+    #   virtualenv
+flask==3.1.1
+    # via lerobot
+fonttools==4.59.0
+    # via matplotlib
+frozenlist==1.7.0
+    # via
+    #   aiohttp
+    #   aiosignal
+fsspec[http]==2025.3.0
+    # via
+    #   datasets
+    #   huggingface-hub
+    #   torch
+gitdb==4.0.12
+    # via gitpython
+gitpython==3.1.45
+    # via wandb
+glfw==2.9.0
+    # via
+    #   dm-control
+    #   mujoco
+grpcio==1.73.1
+    # via
+    #   grpcio-tools
+    #   lerobot
+grpcio-tools==1.73.1
+    # via lerobot
+gym-aloha==0.1.1
+    # via lerobot
+gym-hil==0.1.10
+    # via lerobot
+gym-pusht==0.1.5
+    # via lerobot
+gym-xarm==0.1.1
+    # via lerobot
+gymnasium==0.29.1
+    # via
+    #   gym-aloha
+    #   gym-hil
+    #   gym-pusht
+    #   gym-xarm
+    #   gymnasium-robotics
+    #   lerobot
+    #   pettingzoo
+gymnasium-robotics==1.2.4
+    # via gym-xarm
+hf-transfer==0.1.9
+    # via huggingface-hub
+hf-xet==1.1.5
+    # via huggingface-hub
+hidapi==0.14.0.post4
+    # via
+    #   gym-hil
+    #   lerobot
+huggingface-hub[cli,hf-transfer]==0.34.3
+    # via
+    #   accelerate
+    #   datasets
+    #   diffusers
+    #   lerobot
+    #   tokenizers
+    #   transformers
+identify==2.6.12
+    # via pre-commit
+idna==3.10
+    # via
+    #   requests
+    #   yarl
+imageio[ffmpeg]==2.37.0
+    # via
+    #   gym-aloha
+    #   gym-hil
+    #   gymnasium-robotics
+    #   lerobot
+    #   scikit-image
+imageio-ffmpeg==0.6.0
+    # via imageio
+importlib-metadata==8.7.0
+    # via diffusers
+iniconfig==2.1.0
+    # via pytest
+inquirerpy==0.3.4
+    # via huggingface-hub
+ipython==8.37.0
+    # via meshcat
+ischedule==1.2.7
+    # via placo
+itsdangerous==2.2.0
+    # via flask
+jedi==0.19.2
+    # via ipython
+jinja2==3.1.6
+    # via
+    #   flask
+    #   gymnasium-robotics
+    #   torch
+jsonlines==4.0.0
+    # via lerobot
+kiwisolver==1.4.8
+    # via matplotlib
+labmaze==1.0.6
+    # via dm-control
+lazy-loader==0.4
+    # via scikit-image
+lxml==6.0.0
+    # via dm-control
+markupsafe==3.0.2
+    # via
+    #   flask
+    #   jinja2
+    #   werkzeug
+matplotlib==3.10.5
+    # via lerobot
+matplotlib-inline==0.1.7
+    # via ipython
+mergedeep==1.3.4
+    # via draccus
+meshcat==0.3.2
+    # via placo
+mock-serial==0.0.1
+    # via lerobot
+mpmath==1.3.0
+    # via sympy
+mujoco==2.3.7
+    # via
+    #   dm-control
+    #   gym-aloha
+    #   gym-hil
+    #   gym-xarm
+    #   gymnasium-robotics
+multidict==6.6.3
+    # via
+    #   aiohttp
+    #   yarl
+multiprocess==0.70.16
+    # via datasets
+mypy-extensions==1.1.0
+    # via typing-inspect
+networkx==3.4.2
+    # via
+    #   scikit-image
+    #   torch
+nodeenv==1.9.1
+    # via pre-commit
+num2words==0.5.14
+    # via lerobot
+numpy==2.2.6
+    # via
+    #   accelerate
+    #   cmeel-boost
+    #   contourpy
+    #   datasets
+    #   diffusers
+    #   dm-control
+    #   dm-env
+    #   dm-tree
+    #   gymnasium
+    #   gymnasium-robotics
+    #   imageio
+    #   labmaze
+    #   matplotlib
+    #   meshcat
+    #   mujoco
+    #   opencv-python
+    #   opencv-python-headless
+    #   pandas
+    #   pettingzoo
+    #   rerun-sdk
+    #   scikit-image
+    #   scipy
+    #   shapely
+    #   tifffile
+    #   torchvision
+    #   transformers
+opencv-python==4.12.0.88
+    # via gym-pusht
+opencv-python-headless==4.12.0.88
+    # via lerobot
+orderly-set==5.5.0
+    # via deepdiff
+packaging==25.0
+    # via
+    #   accelerate
+    #   datasets
+    #   huggingface-hub
+    #   lazy-loader
+    #   lerobot
+    #   matplotlib
+    #   pytest
+    #   scikit-image
+    #   transformers
+    #   wandb
+pandas==2.3.1
+    # via
+    #   datasets
+    #   lerobot
+parso==0.8.4
+    # via jedi
+pettingzoo==1.24.3
+    # via gymnasium-robotics
+pexpect==4.9.0
+    # via ipython
+pfzy==0.3.4
+    # via inquirerpy
+pillow==11.3.0
+    # via
+    #   diffusers
+    #   imageio
+    #   matplotlib
+    #   meshcat
+    #   rerun-sdk
+    #   scikit-image
+    #   torchvision
+pin==3.4.0
+    # via placo
+placo==0.9.14
+    # via lerobot
+platformdirs==4.3.8
+    # via
+    #   virtualenv
+    #   wandb
+pluggy==1.6.0
+    # via
+    #   pytest
+    #   pytest-cov
+pre-commit==4.2.0
+    # via lerobot
+prompt-toolkit==3.0.51
+    # via
+    #   inquirerpy
+    #   ipython
+propcache==0.3.2
+    # via
+    #   aiohttp
+    #   yarl
+protobuf==6.31.0
+    # via
+    #   dm-control
+    #   grpcio-tools
+    #   lerobot
+    #   wandb
+psutil==7.0.0
+    # via
+    #   accelerate
+    #   imageio
+ptyprocess==0.7.0
+    # via pexpect
+pure-eval==0.2.3
+    # via stack-data
+pyarrow==21.0.0
+    # via
+    #   datasets
+    #   rerun-sdk
+pycparser==2.22
+    # via cffi
+pydantic==2.11.7
+    # via wandb
+pydantic-core==2.33.2
+    # via pydantic
+pygame==2.6.1
+    # via
+    #   gym-hil
+    #   gym-pusht
+    #   lerobot
+pygments==2.19.2
+    # via
+    #   ipython
+    #   pytest
+pymunk==6.11.1
+    # via
+    #   gym-pusht
+    #   lerobot
+pyngrok==7.2.12
+    # via meshcat
+pynput==1.8.1
+    # via
+    #   gym-hil
+    #   lerobot
+pyobjc-core==11.1
+    # via
+    #   pyobjc-framework-applicationservices
+    #   pyobjc-framework-cocoa
+    #   pyobjc-framework-coretext
+    #   pyobjc-framework-quartz
+pyobjc-framework-applicationservices==11.1
+    # via pynput
+pyobjc-framework-cocoa==11.1
+    # via
+    #   pyobjc-framework-applicationservices
+    #   pyobjc-framework-coretext
+    #   pyobjc-framework-quartz
+pyobjc-framework-coretext==11.1
+    # via pyobjc-framework-applicationservices
+pyobjc-framework-quartz==11.1
+    # via
+    #   pynput
+    #   pyobjc-framework-applicationservices
+    #   pyobjc-framework-coretext
+pyopengl==3.1.9
+    # via
+    #   dm-control
+    #   mujoco
+pyparsing==3.2.3
+    # via
+    #   dm-control
+    #   matplotlib
+pyrealsense2-macosx==2.54.2
+    # via lerobot
+pyserial==3.5
+    # via
+    #   dynamixel-sdk
+    #   feetech-servo-sdk
+    #   lerobot
+pytest==8.4.1
+    # via
+    #   lerobot
+    #   pytest-cov
+    #   pytest-timeout
+pytest-cov==6.2.1
+    # via lerobot
+pytest-timeout==2.4.0
+    # via lerobot
+python-dateutil==2.9.0.post0
+    # via
+    #   matplotlib
+    #   pandas
+pytz==2025.2
+    # via pandas
+pyyaml==6.0.2
+    # via
+    #   accelerate
+    #   datasets
+    #   draccus
+    #   huggingface-hub
+    #   pre-commit
+    #   pyngrok
+    #   pyyaml-include
+    #   transformers
+    #   wandb
+pyyaml-include==1.4.1
+    # via draccus
+pyzmq==27.0.0
+    # via
+    #   lerobot
+    #   meshcat
+regex==2025.7.34
+    # via
+    #   diffusers
+    #   transformers
+requests==2.32.4
+    # via
+    #   datasets
+    #   diffusers
+    #   dm-control
+    #   huggingface-hub
+    #   transformers
+    #   wandb
+rerun-sdk==0.22.1
+    # via lerobot
+rhoban-cmeel-jsoncpp==1.9.4.9
+    # via placo
+safetensors==0.5.3
+    # via
+    #   accelerate
+    #   diffusers
+    #   lerobot
+    #   transformers
+scikit-image==0.25.2
+    # via
+    #   gym-pusht
+    #   lerobot
+scipy==1.15.3
+    # via
+    #   dm-control
+    #   scikit-image
+sentry-sdk==2.34.1
+    # via wandb
+shapely==2.1.1
+    # via gym-pusht
+six==1.17.0
+    # via
+    #   pynput
+    #   python-dateutil
+smmap==5.0.2
+    # via gitdb
+stack-data==0.6.3
+    # via ipython
+sympy==1.14.0
+    # via torch
+termcolor==3.1.0
+    # via lerobot
+tifffile==2025.5.10
+    # via scikit-image
+tokenizers==0.21.4
+    # via transformers
+toml==0.10.2
+    # via draccus
+tomli==2.2.1
+    # via
+    #   cmeel
+    #   coverage
+    #   pytest
+torch==2.7.1
+    # via
+    #   accelerate
+    #   lerobot
+    #   torchvision
+torchcodec==0.5
+    # via lerobot
+torchvision==0.22.1
+    # via lerobot
+tornado==6.5.1
+    # via meshcat
+tqdm==4.67.1
+    # via
+    #   datasets
+    #   dm-control
+    #   huggingface-hub
+    #   transformers
+traitlets==5.14.3
+    # via
+    #   ipython
+    #   matplotlib-inline
+transformers==4.51.3
+    # via lerobot
+typing-extensions==4.14.1
+    # via
+    #   aiosignal
+    #   exceptiongroup
+    #   gymnasium
+    #   huggingface-hub
+    #   ipython
+    #   multidict
+    #   pydantic
+    #   pydantic-core
+    #   rerun-sdk
+    #   torch
+    #   typing-inspect
+    #   typing-inspection
+    #   wandb
+typing-inspect==0.9.0
+    # via draccus
+typing-inspection==0.4.1
+    # via pydantic
+tzdata==2025.2
+    # via pandas
+u-msgpack-python==2.8.0
+    # via meshcat
+urllib3==2.5.0
+    # via
+    #   requests
+    #   sentry-sdk
+virtualenv==20.32.0
+    # via pre-commit
+wandb==0.21.0
+    # via lerobot
+wcwidth==0.2.13
+    # via prompt-toolkit
+werkzeug==3.1.3
+    # via flask
+wrapt==1.17.2
+    # via dm-tree
+xxhash==3.5.0
+    # via datasets
+yarl==1.20.1
+    # via aiohttp
+zipp==3.23.0
+    # via importlib-metadata
+
+# The following packages are considered to be unsafe in a requirements file:
+# setuptools
@@ -0,0 +1,650 @@
+#
+# This file is autogenerated by pip-compile with Python 3.10
+# by the following command:
+#
+#    pip-compile --output-file=requirements-ubuntu.txt requirements.in
+#
+-e .[all]
+    # via -[all]
+absl-py==2.3.1
+    # via
+    #   dm-control
+    #   dm-env
+    #   dm-tree
+    #   labmaze
+    #   mujoco
+accelerate==1.9.0
+    # via lerobot
+aiohappyeyeballs==2.6.1
+    # via aiohttp
+aiohttp==3.12.15
+    # via fsspec
+aiosignal==1.4.0
+    # via aiohttp
+annotated-types==0.7.0
+    # via pydantic
+asttokens==3.0.0
+    # via stack-data
+async-timeout==5.0.1
+    # via aiohttp
+attrs==25.3.0
+    # via
+    #   aiohttp
+    #   dm-tree
+    #   jsonlines
+    #   rerun-sdk
+av==15.0.0
+    # via lerobot
+blinker==1.9.0
+    # via flask
+certifi==2025.7.14
+    # via
+    #   requests
+    #   sentry-sdk
+cffi==1.17.1
+    # via pymunk
+cfgv==3.4.0
+    # via pre-commit
+charset-normalizer==3.4.2
+    # via requests
+click==8.2.1
+    # via
+    #   flask
+    #   wandb
+cloudpickle==3.1.1
+    # via gymnasium
+cmake==4.0.3
+    # via lerobot
+cmeel==0.57.3
+    # via
+    #   cmeel-assimp
+    #   cmeel-boost
+    #   cmeel-console-bridge
+    #   cmeel-octomap
+    #   cmeel-qhull
+    #   cmeel-tinyxml2
+    #   cmeel-urdfdom
+    #   cmeel-zlib
+    #   coal-library
+    #   eigenpy
+    #   eiquadprog
+    #   pin
+    #   placo
+    #   rhoban-cmeel-jsoncpp
+cmeel-assimp==5.4.3.1
+    # via coal-library
+cmeel-boost==1.87.0.1
+    # via
+    #   coal-library
+    #   eigenpy
+    #   eiquadprog
+    #   pin
+cmeel-console-bridge==1.0.2.3
+    # via cmeel-urdfdom
+cmeel-octomap==1.10.0
+    # via coal-library
+cmeel-qhull==8.0.2.1
+    # via coal-library
+cmeel-tinyxml2==10.0.0
+    # via cmeel-urdfdom
+cmeel-urdfdom==4.0.1
+    # via pin
+cmeel-zlib==1.3.1
+    # via cmeel-assimp
+coal-library==3.0.1
+    # via pin
+contourpy==1.3.2
+    # via matplotlib
+coverage[toml]==7.10.1
+    # via pytest-cov
+cycler==0.12.1
+    # via matplotlib
+datasets==3.6.0
+    # via lerobot
+debugpy==1.8.15
+    # via lerobot
+decorator==5.2.1
+    # via ipython
+deepdiff==8.5.0
+    # via lerobot
+diffusers==0.34.0
+    # via lerobot
+dill==0.3.8
+    # via
+    #   datasets
+    #   multiprocess
+distlib==0.4.0
+    # via virtualenv
+dm-control==1.0.14
+    # via gym-aloha
+dm-env==1.6
+    # via dm-control
+dm-tree==0.1.9
+    # via
+    #   dm-control
+    #   dm-env
+docopt==0.6.2
+    # via num2words
+draccus==0.10.0
+    # via lerobot
+dynamixel-sdk==3.7.31
+    # via lerobot
+eigenpy==3.10.3
+    # via coal-library
+einops==0.8.1
+    # via lerobot
+eiquadprog==1.2.9
+    # via placo
+evdev==1.9.2
+    # via pynput
+exceptiongroup==1.3.0
+    # via
+    #   ipython
+    #   pytest
+executing==2.2.0
+    # via stack-data
+farama-notifications==0.0.4
+    # via gymnasium
+feetech-servo-sdk==1.0.0
+    # via lerobot
+filelock==3.18.0
+    # via
+    #   datasets
+    #   diffusers
+    #   huggingface-hub
+    #   torch
+    #   transformers
+    #   virtualenv
+flask==3.1.1
+    # via lerobot
+fonttools==4.59.0
+    # via matplotlib
+frozenlist==1.7.0
+    # via
+    #   aiohttp
+    #   aiosignal
+fsspec[http]==2025.3.0
+    # via
+    #   datasets
+    #   huggingface-hub
+    #   torch
+gitdb==4.0.12
+    # via gitpython
+gitpython==3.1.45
+    # via wandb
+glfw==2.9.0
+    # via
+    #   dm-control
+    #   mujoco
+grpcio==1.73.1
+    # via
+    #   grpcio-tools
+    #   lerobot
+grpcio-tools==1.73.1
+    # via lerobot
+gym-aloha==0.1.1
+    # via lerobot
+gym-hil==0.1.10
+    # via lerobot
+gym-pusht==0.1.5
+    # via lerobot
+gym-xarm==0.1.1
+    # via lerobot
+gymnasium==0.29.1
+    # via
+    #   gym-aloha
+    #   gym-hil
+    #   gym-pusht
+    #   gym-xarm
+    #   gymnasium-robotics
+    #   lerobot
+    #   pettingzoo
+gymnasium-robotics==1.2.4
+    # via gym-xarm
+hf-transfer==0.1.9
+    # via huggingface-hub
+hf-xet==1.1.5
+    # via huggingface-hub
+hidapi==0.14.0.post4
+    # via
+    #   gym-hil
+    #   lerobot
+huggingface-hub[cli,hf-transfer]==0.34.3
+    # via
+    #   accelerate
+    #   datasets
+    #   diffusers
+    #   lerobot
+    #   tokenizers
+    #   transformers
+identify==2.6.12
+    # via pre-commit
+idna==3.10
+    # via
+    #   requests
+    #   yarl
+imageio[ffmpeg]==2.37.0
+    # via
+    #   gym-aloha
+    #   gym-hil
+    #   gymnasium-robotics
+    #   lerobot
+    #   scikit-image
+imageio-ffmpeg==0.6.0
+    # via imageio
+importlib-metadata==8.7.0
+    # via diffusers
+iniconfig==2.1.0
+    # via pytest
+inquirerpy==0.3.4
+    # via huggingface-hub
+ipython==8.37.0
+    # via meshcat
+ischedule==1.2.7
+    # via placo
+itsdangerous==2.2.0
+    # via flask
+jedi==0.19.2
+    # via ipython
+jinja2==3.1.6
+    # via
+    #   flask
+    #   gymnasium-robotics
+    #   torch
+jsonlines==4.0.0
+    # via lerobot
+kiwisolver==1.4.8
+    # via matplotlib
+labmaze==1.0.6
+    # via dm-control
+lazy-loader==0.4
+    # via scikit-image
+lxml==6.0.0
+    # via dm-control
+markupsafe==3.0.2
+    # via
+    #   flask
+    #   jinja2
+    #   werkzeug
+matplotlib==3.10.5
+    # via lerobot
+matplotlib-inline==0.1.7
+    # via ipython
+mergedeep==1.3.4
+    # via draccus
+meshcat==0.3.2
+    # via placo
+mock-serial==0.0.1
+    # via lerobot
+mpmath==1.3.0
+    # via sympy
+mujoco==2.3.7
+    # via
+    #   dm-control
+    #   gym-aloha
+    #   gym-hil
+    #   gym-xarm
+    #   gymnasium-robotics
+multidict==6.6.3
+    # via
+    #   aiohttp
+    #   yarl
+multiprocess==0.70.16
+    # via datasets
+mypy-extensions==1.1.0
+    # via typing-inspect
+networkx==3.4.2
+    # via
+    #   scikit-image
+    #   torch
+nodeenv==1.9.1
+    # via pre-commit
+num2words==0.5.14
+    # via lerobot
+numpy==2.2.6
+    # via
+    #   accelerate
+    #   cmeel-boost
+    #   contourpy
+    #   datasets
+    #   diffusers
+    #   dm-control
+    #   dm-env
+    #   dm-tree
+    #   gymnasium
+    #   gymnasium-robotics
+    #   imageio
+    #   labmaze
+    #   matplotlib
+    #   meshcat
+    #   mujoco
+    #   opencv-python
+    #   opencv-python-headless
+    #   pandas
+    #   pettingzoo
+    #   rerun-sdk
+    #   scikit-image
+    #   scipy
+    #   shapely
+    #   tifffile
+    #   torchvision
+    #   transformers
+nvidia-cublas-cu12==12.6.4.1
+    # via
+    #   nvidia-cudnn-cu12
+    #   nvidia-cusolver-cu12
+    #   torch
+nvidia-cuda-cupti-cu12==12.6.80
+    # via torch
+nvidia-cuda-nvrtc-cu12==12.6.77
+    # via torch
+nvidia-cuda-runtime-cu12==12.6.77
+    # via torch
+nvidia-cudnn-cu12==9.5.1.17
+    # via torch
+nvidia-cufft-cu12==11.3.0.4
+    # via torch
+nvidia-cufile-cu12==1.11.1.6
+    # via torch
+nvidia-curand-cu12==10.3.7.77
+    # via torch
+nvidia-cusolver-cu12==11.7.1.2
+    # via torch
+nvidia-cusparse-cu12==12.5.4.2
+    # via
+    #   nvidia-cusolver-cu12
+    #   torch
+nvidia-cusparselt-cu12==0.6.3
+    # via torch
+nvidia-nccl-cu12==2.26.2
+    # via torch
+nvidia-nvjitlink-cu12==12.6.85
+    # via
+    #   nvidia-cufft-cu12
+    #   nvidia-cusolver-cu12
+    #   nvidia-cusparse-cu12
+    #   torch
+nvidia-nvtx-cu12==12.6.77
+    # via torch
+opencv-python==4.12.0.88
+    # via gym-pusht
+opencv-python-headless==4.12.0.88
+    # via lerobot
+orderly-set==5.5.0
+    # via deepdiff
+packaging==25.0
+    # via
+    #   accelerate
+    #   datasets
+    #   huggingface-hub
+    #   lazy-loader
+    #   lerobot
+    #   matplotlib
+    #   pytest
+    #   scikit-image
+    #   transformers
+    #   wandb
+pandas==2.3.1
+    # via
+    #   datasets
+    #   lerobot
+parso==0.8.4
+    # via jedi
+pettingzoo==1.24.3
+    # via gymnasium-robotics
+pexpect==4.9.0
+    # via ipython
+pfzy==0.3.4
+    # via inquirerpy
+pillow==11.3.0
+    # via
+    #   diffusers
+    #   imageio
+    #   matplotlib
+    #   meshcat
+    #   rerun-sdk
+    #   scikit-image
+    #   torchvision
+pin==3.4.0
+    # via placo
+placo==0.9.14
+    # via lerobot
+platformdirs==4.3.8
+    # via
+    #   virtualenv
+    #   wandb
+pluggy==1.6.0
+    # via
+    #   pytest
+    #   pytest-cov
+pre-commit==4.2.0
+    # via lerobot
+prompt-toolkit==3.0.51
+    # via
+    #   inquirerpy
+    #   ipython
+propcache==0.3.2
+    # via
+    #   aiohttp
+    #   yarl
+protobuf==6.31.0
+    # via
+    #   dm-control
+    #   grpcio-tools
+    #   lerobot
+    #   wandb
+psutil==7.0.0
+    # via
+    #   accelerate
+    #   imageio
+ptyprocess==0.7.0
+    # via pexpect
+pure-eval==0.2.3
+    # via stack-data
+pyarrow==21.0.0
+    # via
+    #   datasets
+    #   rerun-sdk
+pycparser==2.22
+    # via cffi
+pydantic==2.11.7
+    # via wandb
+pydantic-core==2.33.2
+    # via pydantic
+pygame==2.6.1
+    # via
+    #   gym-hil
+    #   gym-pusht
+    #   lerobot
+pygments==2.19.2
+    # via
+    #   ipython
+    #   pytest
+pymunk==6.11.1
+    # via
+    #   gym-pusht
+    #   lerobot
+pyngrok==7.2.12
+    # via meshcat
+pynput==1.8.1
+    # via
+    #   gym-hil
+    #   lerobot
+pyopengl==3.1.9
+    # via
+    #   dm-control
+    #   mujoco
+pyparsing==3.2.3
+    # via
+    #   dm-control
+    #   matplotlib
+pyrealsense2==2.56.5.9235
+    # via lerobot
+pyserial==3.5
+    # via
+    #   dynamixel-sdk
+    #   feetech-servo-sdk
+    #   lerobot
+pytest==8.4.1
+    # via
+    #   lerobot
+    #   pytest-cov
+    #   pytest-timeout
+pytest-cov==6.2.1
+    # via lerobot
+pytest-timeout==2.4.0
+    # via lerobot
+python-dateutil==2.9.0.post0
+    # via
+    #   matplotlib
+    #   pandas
+python-xlib==0.33
+    # via pynput
+pytz==2025.2
+    # via pandas
+pyyaml==6.0.2
+    # via
+    #   accelerate
+    #   datasets
+    #   draccus
+    #   huggingface-hub
+    #   pre-commit
+    #   pyngrok
+    #   pyyaml-include
+    #   transformers
+    #   wandb
+pyyaml-include==1.4.1
+    # via draccus
+pyzmq==27.0.0
+    # via
+    #   lerobot
+    #   meshcat
+regex==2025.7.34
+    # via
+    #   diffusers
+    #   transformers
+requests==2.32.4
+    # via
+    #   datasets
+    #   diffusers
+    #   dm-control
+    #   huggingface-hub
+    #   transformers
+    #   wandb
+rerun-sdk==0.22.1
+    # via lerobot
+rhoban-cmeel-jsoncpp==1.9.4.9
+    # via placo
+safetensors==0.5.3
+    # via
+    #   accelerate
+    #   diffusers
+    #   lerobot
+    #   transformers
+scikit-image==0.25.2
+    # via
+    #   gym-pusht
+    #   lerobot
+scipy==1.15.3
+    # via
+    #   dm-control
+    #   scikit-image
+sentry-sdk==2.34.1
+    # via wandb
+shapely==2.1.1
+    # via gym-pusht
+six==1.17.0
+    # via
+    #   pynput
+    #   python-dateutil
+    #   python-xlib
+smmap==5.0.2
+    # via gitdb
+stack-data==0.6.3
+    # via ipython
+sympy==1.14.0
+    # via torch
+termcolor==3.1.0
+    # via lerobot
+tifffile==2025.5.10
+    # via scikit-image
+tokenizers==0.21.4
+    # via transformers
+toml==0.10.2
+    # via draccus
+tomli==2.2.1
+    # via
+    #   cmeel
+    #   coverage
+    #   pytest
+torch==2.7.1
+    # via
+    #   accelerate
+    #   lerobot
+    #   torchvision
+torchcodec==0.5
+    # via lerobot
+torchvision==0.22.1
+    # via lerobot
+tornado==6.5.1
+    # via meshcat
+tqdm==4.67.1
+    # via
+    #   datasets
+    #   dm-control
+    #   huggingface-hub
+    #   transformers
+traitlets==5.14.3
+    # via
+    #   ipython
+    #   matplotlib-inline
+transformers==4.51.3
+    # via lerobot
+triton==3.3.1
+    # via torch
+typing-extensions==4.14.1
+    # via
+    #   aiosignal
+    #   exceptiongroup
+    #   gymnasium
+    #   huggingface-hub
+    #   ipython
+    #   multidict
+    #   pydantic
+    #   pydantic-core
+    #   rerun-sdk
+    #   torch
+    #   typing-inspect
+    #   typing-inspection
+    #   wandb
+typing-inspect==0.9.0
+    # via draccus
+typing-inspection==0.4.1
+    # via pydantic
+tzdata==2025.2
+    # via pandas
+u-msgpack-python==2.8.0
+    # via meshcat
+urllib3==2.5.0
+    # via
+    #   requests
+    #   sentry-sdk
+virtualenv==20.32.0
+    # via pre-commit
+wandb==0.21.0
+    # via lerobot
+wcwidth==0.2.13
+    # via prompt-toolkit
+werkzeug==3.1.3
+    # via flask
+wrapt==1.17.2
+    # via dm-tree
+xxhash==3.5.0
+    # via datasets
+yarl==1.20.1
+    # via aiohttp
+zipp==3.23.0
+    # via importlib-metadata
+
+# The following packages are considered to be unsafe in a requirements file:
+# setuptools
@@ -0,0 +1,9 @@
+# requirements.in
+
+# requirements-macos.txt was generated on macOS and is platform-specific (macOS 15.5 24F74 arm64).
+# Darwin MacBook-Pro.local 24.5.0 Darwin Kernel Version 24.5.0: Tue Apr 22 19:54:43 PDT 2025; root:xnu-11417.121.6~2/RELEASE_ARM64_T8132 arm64
+
+# requirements-ubuntu.txt was generated on Linux and is platform-specific (Ubuntu 24.04.2 LTS x86_64).
+# Linux mlerobot-linux 6.14.0-27-generic #27~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Tue Jul 22 17:38:49 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
+
+-e .[all]
@@ -82,5 +82,9 @@ def calibrate(cfg: CalibrateConfig):
    device.disconnect()


-if __name__ == "__main__":
+def main():
    calibrate()
+
+
+if __name__ == "__main__":
+    main()
@@ -368,7 +368,7 @@ class OpenCVCamera(Camera):
        if requested_color_mode == ColorMode.RGB:
            processed_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

-        if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+        if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE, cv2.ROTATE_180]:
            processed_image = cv2.rotate(processed_image, self.rotation)

        return processed_image
@@ -434,7 +434,7 @@ class RealSenseCamera(Camera):
        if self.color_mode == ColorMode.BGR:
            processed_image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)

-        if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE]:
+        if self.rotation in [cv2.ROTATE_90_CLOCKWISE, cv2.ROTATE_90_COUNTERCLOCKWISE, cv2.ROTATE_180]:
            processed_image = cv2.rotate(processed_image, self.rotation)

        return processed_image
@@ -44,7 +44,7 @@ class EnvConfig(draccus.ChoiceRegistry, abc.ABC):
@EnvConfig.register_subclass("aloha")
@dataclass
 class AlohaEnv(EnvConfig):
-    task: str = "AlohaInsertion-v0"
+    task: str | None = "AlohaInsertion-v0"
    fps: int = 50
    episode_length: int = 400
    obs_type: str = "pixels_agent_pos"
@@ -82,7 +82,7 @@ class AlohaEnv(EnvConfig):
@EnvConfig.register_subclass("pusht")
@dataclass
 class PushtEnv(EnvConfig):
-    task: str = "PushT-v0"
+    task: str | None = "PushT-v0"
    fps: int = 10
    episode_length: int = 300
    obs_type: str = "pixels_agent_pos"
@@ -124,7 +124,7 @@ class PushtEnv(EnvConfig):
@EnvConfig.register_subclass("xarm")
@dataclass
 class XarmEnv(EnvConfig):
-    task: str = "XarmLift-v0"
+    task: str | None = "XarmLift-v0"
    fps: int = 15
    episode_length: int = 200
    obs_type: str = "pixels_agent_pos"
@@ -200,10 +200,10 @@ class HILSerlRobotEnvConfig(EnvConfig):
    wrapper: EnvTransformConfig | None = None
    fps: int = 10
    name: str = "real_robot"
-    mode: str = None  # Either "record", "replay", None
+    mode: str | None = None  # Either "record", "replay", None
    repo_id: str | None = None
    dataset_root: str | None = None
-    task: str = ""
+    task: str | None = ""
    num_episodes: int = 10  # only for record mode
    episode: int = 0
    device: str = "cuda"
@@ -213,6 +213,7 @@ class HILSerlRobotEnvConfig(EnvConfig):
    # For the reward classifier, to record more positive examples after a success
    number_of_steps_after_success: int = 0

+    @property
    def gym_kwargs(self) -> dict:
        return {}

@@ -222,9 +223,8 @@ class HILSerlRobotEnvConfig(EnvConfig):
 class HILEnvConfig(EnvConfig):
    """Configuration for the HIL environment."""

-    type: str = "hil"
    name: str = "PandaPickCube"
-    task: str = "PandaPickCubeKeyboard-v0"
+    task: str | None = "PandaPickCubeKeyboard-v0"
    use_viewer: bool = True
    gripper_penalty: float = 0.0
    use_gamepad: bool = True
@@ -252,7 +252,7 @@ class HILEnvConfig(EnvConfig):
    robot_config: RobotConfig | None = None
    teleop_config: TeleoperatorConfig | None = None
    wrapper: EnvTransformConfig | None = None
-    mode: str = None  # Either "record", "replay", None
+    mode: str | None = None  # Either "record", "replay", None
    repo_id: str | None = None
    dataset_root: str | None = None
    num_episodes: int = 10  # only for record mode
@@ -286,7 +286,7 @@ def save_images_from_all_cameras(
            print(f"Image capture finished. Images saved to {output_dir}")


-if __name__ == "__main__":
+def main():
    parser = argparse.ArgumentParser(
        description="Unified camera utility script for listing cameras and capturing images."
    )
@@ -313,3 +313,7 @@ if __name__ == "__main__":
    )
    args = parser.parse_args()
    save_images_from_all_cameras(**vars(args))
+
+
+if __name__ == "__main__":
+    main()
@@ -61,5 +61,9 @@ def find_port():
        raise OSError(f"Could not detect the port. More than one port was found ({ports_diff}).")


-if __name__ == "__main__":
+def main():
    find_port()
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1 @@
+../../../../docs/source/policy_act_README.md
@@ -0,0 +1 @@
+../../../../docs/source/policy_diffusion_README.md
@@ -217,12 +217,13 @@ class DiffusionConfig(PreTrainedConfig):
                    )

        # Check that all input images have the same shape.
-        first_image_key, first_image_ft = next(iter(self.image_features.items()))
-        for key, image_ft in self.image_features.items():
-            if image_ft.shape != first_image_ft.shape:
-                raise ValueError(
-                    f"`{key}` does not match `{first_image_key}`, but we expect all image shapes to match."
-                )
+        if len(self.image_features) > 0:
+            first_image_key, first_image_ft = next(iter(self.image_features.items()))
+            for key, image_ft in self.image_features.items():
+                if image_ft.shape != first_image_ft.shape:
+                    raise ValueError(
+                        f"`{key}` does not match `{first_image_key}`, but we expect all image shapes to match."
+                    )

    @property
    def observation_delta_indices(self) -> list:
@@ -288,7 +288,7 @@ class DiffusionModel(nn.Module):

            "observation.images": (B, n_obs_steps, num_cameras, C, H, W)
                AND/OR
-            "observation.environment_state": (B, environment_dim)
+            "observation.environment_state": (B, n_obs_steps, environment_dim)
        }
        """
        batch_size, n_obs_steps = batch["observation.state"].shape[:2]
@@ -315,7 +315,7 @@ class DiffusionModel(nn.Module):

            "observation.images": (B, n_obs_steps, num_cameras, C, H, W)
                AND/OR
-            "observation.environment_state": (B, environment_dim)
+            "observation.environment_state": (B, n_obs_steps, environment_dim)

            "action": (B, horizon, action_dim)
            "action_is_pad": (B, horizon)
@@ -66,7 +66,8 @@ from lerobot.policies.pi0.paligemma_with_expert import (
    PaliGemmaWithExpertModel,
 )
 from lerobot.policies.pretrained import PreTrainedPolicy
-from lerobot.utils.utils import get_safe_dtype
+from lerobot.policies.utils import log_model_loading_keys
+from lerobot.utils.utils import get_safe_dtype, init_logging


 def create_sinusoidal_pos_embedding(
@@ -90,12 +91,6 @@ def create_sinusoidal_pos_embedding(
    return pos_emb


-def sample_beta(alpha, beta, bsize, device):
-    gamma1 = torch.empty((bsize,), device=device).uniform_(0, 1).pow(1 / alpha)
-    gamma2 = torch.empty((bsize,), device=device).uniform_(0, 1).pow(1 / beta)
-    return gamma1 / (gamma1 + gamma2)
-
-
 def make_att_2d_masks(pad_masks, att_masks):
    """Copied from big_vision.

@@ -258,6 +253,99 @@ class PI0Policy(PreTrainedPolicy):
        """This should be called whenever the environment is reset."""
        self._action_queue = deque([], maxlen=self.config.n_action_steps)

+    @classmethod
+    def _transform_state_dict_keys(cls, state_dict: dict) -> dict:
+        """
+        Transform state dict keys to match expected model structure.
+
+        Transformations:
+        - model.paligemma_with_expert.paligemma.language_model.lm_head ->
+          model.paligemma_with_expert.paligemma.lm_head
+        - model.paligemma_with_expert.paligemma.language_model.model ->
+          model.paligemma_with_expert.paligemma.model.language_model
+        - model.paligemma_with_expert.paligemma.vision_tower ->
+          model.paligemma_with_expert.paligemma.model.vision_tower
+        - model.paligemma_with_expert.paligemma.multi_modal_projector ->
+          model.paligemma_with_expert.paligemma.model.multi_modal_projector
+
+        Also handles tied weights between lm_head.weight and
+        embed_tokens.weight.
+        """
+        import re
+
+        transformed_dict = {}
+
+        transformations = [
+            (
+                re.compile(r"\.paligemma_with_expert\.paligemma\.language_model\.lm_head"),
+                ".paligemma_with_expert.paligemma.lm_head",
+            ),
+            (
+                re.compile(r"\.paligemma_with_expert\.paligemma\.language_model\.model"),
+                ".paligemma_with_expert.paligemma.model.language_model",
+            ),
+            (
+                re.compile(r"\.paligemma_with_expert\.paligemma\.vision_tower"),
+                ".paligemma_with_expert.paligemma.model.vision_tower",
+            ),
+            (
+                re.compile(r"\.paligemma_with_expert\.paligemma\.multi_modal_projector"),
+                ".paligemma_with_expert.paligemma.model.multi_modal_projector",
+            ),
+        ]
+
+        for key, value in state_dict.items():
+            new_key = key
+            for pattern, replacement in transformations:
+                new_key = pattern.sub(replacement, new_key)
+            transformed_dict[new_key] = value
+
+        # Handle tied weights: lm_head.weight and embed_tokens.weight share memory
+        lm_head_key = None
+        embed_tokens_key = None
+
+        for key in transformed_dict:
+            if key.endswith(".paligemma_with_expert.paligemma.lm_head.weight"):
+                lm_head_key = key
+            elif key.endswith(".paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight"):
+                embed_tokens_key = key
+            if lm_head_key and embed_tokens_key:
+                break
+
+        if lm_head_key and not embed_tokens_key:
+            embed_tokens_key = lm_head_key.replace(
+                ".lm_head.weight", ".model.language_model.embed_tokens.weight"
+            )
+            transformed_dict[embed_tokens_key] = transformed_dict[lm_head_key]
+        elif embed_tokens_key and not lm_head_key:
+            lm_head_key = embed_tokens_key.replace(
+                ".model.language_model.embed_tokens.weight", ".lm_head.weight"
+            )
+            transformed_dict[lm_head_key] = transformed_dict[embed_tokens_key]
+
+        return transformed_dict
+
+    @classmethod
+    def _load_as_safetensor(
+        cls, model: "PI0Policy", model_file: str, map_location: str, strict: bool
+    ) -> "PI0Policy":
+        """Override to apply key transformations before loading."""
+        from safetensors.torch import load_file
+
+        init_logging()
+        # Load the state dict from file safely
+        state_dict = load_file(model_file, device=map_location)
+
+        # Apply key transformations
+        transformed_state_dict = cls._transform_state_dict_keys(state_dict)
+
+        # Load the transformed state dict
+        msg = model.load_state_dict(transformed_state_dict, strict=strict)
+
+        # Log message
+        log_model_loading_keys(msg.missing_keys, msg.unexpected_keys)
+        return model
+
    def get_optim_params(self) -> dict:
        return self.parameters()

@@ -515,9 +603,10 @@ class PI0FlowMatching(nn.Module):
        return noise

    def sample_time(self, bsize, device):
-        time_beta = sample_beta(1.5, 1.0, bsize, device)
+        beta_dist = torch.distributions.Beta(concentration1=1.5, concentration0=1.0)
+        time_beta = beta_dist.sample((bsize,)).to(device=device, dtype=torch.float32)
        time = time_beta * 0.999 + 0.001
-        return time.to(dtype=torch.float32, device=device)
+        return time

    def embed_prefix(
        self, images, img_masks, lang_tokens, lang_masks
@@ -30,6 +30,7 @@ from torch import Tensor, nn

 from lerobot.configs.policies import PreTrainedConfig
 from lerobot.configs.train import TrainPipelineConfig
+from lerobot.policies.utils import log_model_loading_keys
 from lerobot.utils.hub import HubMixin

 T = TypeVar("T", bound="PreTrainedPolicy")
@@ -128,18 +129,26 @@ class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):

    @classmethod
    def _load_as_safetensor(cls, model: T, model_file: str, map_location: str, strict: bool) -> T:
-        if packaging.version.parse(safetensors.__version__) < packaging.version.parse("0.4.3"):
-            load_model_as_safetensor(model, model_file, strict=strict)
-            if map_location != "cpu":
-                logging.warning(
-                    "Loading model weights on other devices than 'cpu' is not supported natively in your version of safetensors."
-                    " This means that the model is loaded on 'cpu' first and then copied to the device."
-                    " This leads to a slower loading time."
-                    " Please update safetensors to version 0.4.3 or above for improved performance."
-                )
-                model.to(map_location)
-        else:
-            safetensors.torch.load_model(model, model_file, strict=strict, device=map_location)
+        # Create base kwargs
+        kwargs = {"strict": strict}
+
+        # Add device parameter for newer versions that support it
+        if packaging.version.parse(safetensors.__version__) >= packaging.version.parse("0.4.3"):
+            kwargs["device"] = map_location
+
+        # Load the model with appropriate kwargs
+        missing_keys, unexpected_keys = load_model_as_safetensor(model, model_file, **kwargs)
+        log_model_loading_keys(missing_keys, unexpected_keys)
+
+        # For older versions, manually move to device if needed
+        if "device" not in kwargs and map_location != "cpu":
+            logging.warning(
+                "Loading model weights on other devices than 'cpu' is not supported natively in your version of safetensors."
+                " This means that the model is loaded on 'cpu' first and then copied to the device."
+                " This leads to a slower loading time."
+                " Please update safetensors to version 0.4.3 or above for improved performance."
+            )
+            model.to(map_location)
        return model

    @abc.abstractmethod
@@ -0,0 +1 @@
+../../../../docs/source/policy_smolvla_README.md
@@ -194,12 +194,6 @@ def create_sinusoidal_pos_embedding(
    return pos_emb


-def sample_beta(alpha, beta, bsize, device):
-    gamma1 = torch.empty((bsize,), device=device).uniform_(0, 1).pow(1 / alpha)
-    gamma2 = torch.empty((bsize,), device=device).uniform_(0, 1).pow(1 / beta)
-    return gamma1 / (gamma1 + gamma2)
-
-
 def make_att_2d_masks(pad_masks, att_masks):
    """Copied from big_vision.

@@ -690,9 +684,10 @@ class VLAFlowMatching(nn.Module):
        return noise

    def sample_time(self, bsize, device):
-        time_beta = sample_beta(1.5, 1.0, bsize, device)
+        beta_dist = torch.distributions.Beta(concentration1=1.5, concentration0=1.0)
+        time_beta = beta_dist.sample((bsize,)).to(device=device, dtype=torch.float32)
        time = time_beta * 0.999 + 0.001
-        return time.to(dtype=torch.float32, device=device)
+        return time

    def embed_prefix(
        self, images, img_masks, lang_tokens, lang_masks, state: torch.Tensor = None
@@ -0,0 +1 @@
+../../../../docs/source/policy_tdmpc_README.md
@@ -14,6 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.

+import logging
 from collections import deque

 import torch
@@ -71,3 +72,16 @@ def get_output_shape(module: nn.Module, input_shape: tuple) -> tuple:
    with torch.inference_mode():
        output = module(dummy_input)
    return tuple(output.shape)
+
+
+def log_model_loading_keys(missing_keys: list[str], unexpected_keys: list[str]) -> None:
+    """Log missing and unexpected keys when loading a model.
+
+    Args:
+        missing_keys (list[str]): Keys that were expected but not found.
+        unexpected_keys (list[str]): Keys that were found but not expected.
+    """
+    if missing_keys:
+        logging.warning(f"Missing key(s) when loading model: {missing_keys}")
+    if unexpected_keys:
+        logging.warning(f"Unexpected key(s) when loading model: {unexpected_keys}")
@@ -0,0 +1 @@
+../../../../docs/source/policy_vqbet_README.md
@@ -393,5 +393,9 @@ def record(cfg: RecordConfig) -> LeRobotDataset:
    return dataset


-if __name__ == "__main__":
+def main():
    record()
+
+
+if __name__ == "__main__":
+    main()
@@ -112,5 +112,9 @@ def replay(cfg: ReplayConfig):
    robot.disconnect()


-if __name__ == "__main__":
+def main():
    replay()
+
+
+if __name__ == "__main__":
+    main()
@@ -501,6 +501,10 @@ def eval_main(cfg: EvalPipelineConfig):
    logging.info("End of eval")


-if __name__ == "__main__":
+def main():
    init_logging()
    eval_main()
+
+
+if __name__ == "__main__":
+    main()
@@ -286,6 +286,10 @@ def train(cfg: TrainPipelineConfig):
        policy.push_model_to_hub(cfg)


-if __name__ == "__main__":
+def main():
    init_logging()
    train()
+
+
+if __name__ == "__main__":
+    main()
@@ -80,5 +80,9 @@ def setup_motors(cfg: SetupConfig):
    device.setup_motors()


-if __name__ == "__main__":
+def main():
    setup_motors()
+
+
+if __name__ == "__main__":
+    main()
@@ -153,5 +153,9 @@ def teleoperate(cfg: TeleoperateConfig):
        robot.disconnect()


-if __name__ == "__main__":
+def main():
    teleoperate()
+
+
+if __name__ == "__main__":
+    main()
Author	SHA1	Message	Date
Steven Palma	f771e3eaf1	fix(ci): create venv for release testing (#1652 )	2025-08-01 21:04:47 +02:00
Steven Palma	240a3892ae	fix(ci): remove uv run + bump minor (#1651 )	2025-08-01 20:52:10 +02:00
Steven Palma	3e24ecaf54	chore(ci): Bump to v0.3.0 (#1649 )	2025-08-01 18:30:33 +02:00
Steven Palma	60dc8e3a5d	fix(ci): use base tag for testpy to mimic the pyproject.toml version (#1648 )	2025-08-01 18:21:37 +02:00
Steven Palma	dcb305ffb2	fix(ci): change release-name to title (#1647 )	2025-08-01 18:11:08 +02:00
Steven Palma	11525cedeb	fix(ci): change steps based on wheter it is a -rc tag (#1646 )	2025-08-01 18:05:20 +02:00
Simon Alibert	2f8d98b05e	Update readme (#1570 ) * Cleanup badges * Remove comment * Remove profiling section * Move acknowledgment * Move citations * Fix badge display * Move build your robot section * Fix nightly badge * Revert `be13b3f` * Update README.md Co-authored-by: HUANG TZU-CHUN <tzu.chun.huang.tw@gmail.com> Signed-off-by: Simon Alibert <75076266+aliberts@users.noreply.github.com> * chore(docs): optimize readme for PyPI rendering * chore(docs): move policy readme to docs folder + symlink in policy dirs * fix(docs): max width og lerobot logo + url in citation block --------- Signed-off-by: Simon Alibert <75076266+aliberts@users.noreply.github.com> Co-authored-by: HUANG TZU-CHUN <tzu.chun.huang.tw@gmail.com> Co-authored-by: Steven Palma <steven.palma@huggingface.co>	2025-08-01 17:39:39 +02:00
Steven Palma	1baaa77a86	feat(ci): release workflow publish to pypi test + lock files (#1643 ) * chore(ci): add some release stuff * chore(ci): add requirements-macos * chore(ci): added lockfiles for future reference * feat(ci): add draft & prerelease option to release workflow tag	2025-08-01 17:14:15 +02:00
Steven Palma	91ed6097bc	fix(ci): declare entrypoints + fix testing release (#1642 )	2025-08-01 12:04:34 +02:00
Francesco Capuano	945e1ff266	fix colab typo (#1629 ) Signed-off-by: Francesco Capuano <74058581+fracapuano@users.noreply.github.com>	2025-07-31 11:08:12 +02:00
Yushun Xiang	71eff183ff	Fix pi0 checkpoint state map (#1415 ) Co-authored-by: Michel Aractingi <michel.aractingi@huggingface.co>	2025-07-30 17:38:32 +02:00
Rayen Ghali	67196c9d53	fix(180-degree rotation): Add `cv2.ROTATE_180` to rotation checks in both OpenCV and RealSense camera implementations	2025-07-29 13:54:43 +02:00
Abhay Deshpande	5695432142	fix(DiffusionPolicy): Fix bug where training without image features would crash with exception, fix environment state docs (#1617 ) * Fix bug in diffusion config validation when not using image features * Fix DiffusionPolicy docstring about shape of env state	2025-07-29 13:40:16 +02:00
Caroline Pascal	c14ab9e97b	fix(dependencies): removing versions ceilings on tokenizers and huggingface_hub dependencies (#1618 )	2025-07-29 10:59:23 +02:00
Michel Aractingi	c7c3b477d6	Fix sample beta for smolvla as done for pi0, remove sample_beta func (#1611 )	2025-07-28 17:28:55 +02:00
Caroline Pascal	b267cd40f7	fix(tokenizers dependency): adding ceiling version on tokenizers (#1612 )	2025-07-28 17:05:44 +02:00
Lumen Yang	7fe6adaf61	fix(config): typing correction on config.py (#1320 ) Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Michel Aractingi <michel.aractingi@huggingface.co>	2025-07-28 15:22:37 +02:00
Kleist Bond	4b88842d20	fix bug about sampling time from beta distribution (#1605 ) * fix bug about sampling t from beta distribution * fix: address review comments ---------	2025-07-28 15:17:30 +02:00
				`@@ -0,0 +1 @@`
				`../../../../docs/source/policy_act_README.md`
				`@@ -0,0 +1 @@`
				`../../../../docs/source/policy_diffusion_README.md`
				`@@ -0,0 +1 @@`
				`../../../../docs/source/policy_smolvla_README.md`
				`@@ -0,0 +1 @@`
				`../../../../docs/source/policy_tdmpc_README.md`
				`@@ -0,0 +1 @@`
				`../../../../docs/source/policy_vqbet_README.md`