use Path type instead of str

.github/poetry/cpu/poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 1.8.2 and should not be changed by hand.
+# This file is automatically @generated by Poetry 1.8.1 and should not be changed by hand.
 
 [[package]]
 name = "absl-py"
@@ -2903,7 +2903,7 @@ reference = "torch-cpu"
 
 [[package]]
 name = "torchrl"
-version = "0.4.0+c371266"
+version = "0.4.0+13bef42"
 description = ""
 optional = false
 python-versions = "*"
@@ -2918,13 +2918,13 @@ tensordict = ">=0.4.0"
 torch = ">=2.1.0"
 
 [package.extras]
-all = ["ale-py", "atari-py", "dm-control", "git", "gym", "gym[accept-rom-license]", "gymnasium", "h5py", "huggingface-hub", "hydra-core (>=1.1)", "hydra-submitit-launcher", "minari", "moviepy", "mujoco", "pandas", "pettingzoo (>=1.24.1)", "pillow", "pygame", "pytest", "pytest-instafail", "pyyaml", "requests", "scikit-learn", "scipy", "tensorboard", "torchsnapshot", "torchvision", "tqdm", "vmas (>=1.2.10)", "wandb"]
+all = ["ale-py", "atari-py", "dm_control", "git", "gym", "gym[accept-rom-license]", "gymnasium", "h5py", "huggingface_hub", "hydra-core (>=1.1)", "hydra-submitit-launcher", "minari", "moviepy", "mujoco", "pandas", "pettingzoo (>=1.24.1)", "pillow", "pygame", "pytest", "pytest-instafail", "pyyaml", "requests", "scikit-learn", "scipy", "tensorboard", "torchsnapshot", "torchvision", "tqdm", "vmas (>=1.2.10)", "wandb"]
 atari = ["ale-py", "atari-py", "gym", "gym[accept-rom-license]", "pygame"]
 checkpointing = ["torchsnapshot"]
-dm-control = ["dm-control"]
+dm-control = ["dm_control"]
 gym-continuous = ["gymnasium", "mujoco"]
 marl = ["pettingzoo (>=1.24.1)", "vmas (>=1.2.10)"]
-offline-data = ["h5py", "huggingface-hub", "minari", "pandas", "pillow", "requests", "scikit-learn", "torchvision", "tqdm"]
+offline-data = ["h5py", "huggingface_hub", "minari", "pandas", "pillow", "requests", "scikit-learn", "torchvision", "tqdm"]
 rendering = ["moviepy"]
 tests = ["pytest", "pytest-instafail", "pyyaml", "scipy"]
 utils = ["git", "hydra-core (>=1.1)", "hydra-submitit-launcher", "tensorboard", "tqdm", "wandb"]
@@ -2932,8 +2932,8 @@ utils = ["git", "hydra-core (>=1.1)", "hydra-submitit-launcher", "tensorboard",
 [package.source]
 type = "git"
 url = "https://github.com/pytorch/rl"
-reference = "HEAD"
-resolved_reference = "c371266ce5a71cb4b1a319cc56ad59d9b492cb9d"
+reference = "13bef426dcfa5887c6e5034a6e9697993fa92c37"
+resolved_reference = "13bef426dcfa5887c6e5034a6e9697993fa92c37"
 
 [[package]]
 name = "torchvision"
@@ -3123,4 +3123,4 @@ testing = ["big-O", "jaraco.functools", "jaraco.itertools", "more-itertools", "p
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.10"
-content-hash = "4dca0197360d62d56d870cd4791a9bb8bfa179645b892441a1296ac45285b19f"
+content-hash = "4aa6a1e3f29560dd4a1c24d493ee1154089da4aa8d2190ad1f786c125ab2b735"

.github/poetry/cpu/pyproject.toml

@@ -41,7 +41,7 @@ numba = "^0.59.0"
 mpmath = "^1.3.0"
 torch = {version = "^2.2.1", source = "torch-cpu"}
 tensordict = {git = "https://github.com/pytorch/tensordict"}
-torchrl = {git = "https://github.com/pytorch/rl"}
+torchrl = {git = "https://github.com/pytorch/rl", rev = "13bef426dcfa5887c6e5034a6e9697993fa92c37"}
 mujoco = "^3.1.2"
 mujoco-py = "^2.1.2.14"
 gym = "^0.26.2"

LICENSE

@@ -276,207 +276,3 @@ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 SOFTWARE.
-
-## Some of lerobot's code is derived from DETR, which is subject to the following copyright notice:
-
-                                 Apache License
-                           Version 2.0, January 2004
-                        http://www.apache.org/licenses/
-
-   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
-
-   1. Definitions.
-
-      "License" shall mean the terms and conditions for use, reproduction,
-      and distribution as defined by Sections 1 through 9 of this document.
-
-      "Licensor" shall mean the copyright owner or entity authorized by
-      the copyright owner that is granting the License.
-
-      "Legal Entity" shall mean the union of the acting entity and all
-      other entities that control, are controlled by, or are under common
-      control with that entity. For the purposes of this definition,
-      "control" means (i) the power, direct or indirect, to cause the
-      direction or management of such entity, whether by contract or
-      otherwise, or (ii) ownership of fifty percent (50%) or more of the
-      outstanding shares, or (iii) beneficial ownership of such entity.
-
-      "You" (or "Your") shall mean an individual or Legal Entity
-      exercising permissions granted by this License.
-
-      "Source" form shall mean the preferred form for making modifications,
-      including but not limited to software source code, documentation
-      source, and configuration files.
-
-      "Object" form shall mean any form resulting from mechanical
-      transformation or translation of a Source form, including but
-      not limited to compiled object code, generated documentation,
-      and conversions to other media types.
-
-      "Work" shall mean the work of authorship, whether in Source or
-      Object form, made available under the License, as indicated by a
-      copyright notice that is included in or attached to the work
-      (an example is provided in the Appendix below).
-
-      "Derivative Works" shall mean any work, whether in Source or Object
-      form, that is based on (or derived from) the Work and for which the
-      editorial revisions, annotations, elaborations, or other modifications
-      represent, as a whole, an original work of authorship. For the purposes
-      of this License, Derivative Works shall not include works that remain
-      separable from, or merely link (or bind by name) to the interfaces of,
-      the Work and Derivative Works thereof.
-
-      "Contribution" shall mean any work of authorship, including
-      the original version of the Work and any modifications or additions
-      to that Work or Derivative Works thereof, that is intentionally
-      submitted to Licensor for inclusion in the Work by the copyright owner
-      or by an individual or Legal Entity authorized to submit on behalf of
-      the copyright owner. For the purposes of this definition, "submitted"
-      means any form of electronic, verbal, or written communication sent
-      to the Licensor or its representatives, including but not limited to
-      communication on electronic mailing lists, source code control systems,
-      and issue tracking systems that are managed by, or on behalf of, the
-      Licensor for the purpose of discussing and improving the Work, but
-      excluding communication that is conspicuously marked or otherwise
-      designated in writing by the copyright owner as "Not a Contribution."
-
-      "Contributor" shall mean Licensor and any individual or Legal Entity
-      on behalf of whom a Contribution has been received by Licensor and
-      subsequently incorporated within the Work.
-
-   2. Grant of Copyright License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      copyright license to reproduce, prepare Derivative Works of,
-      publicly display, publicly perform, sublicense, and distribute the
-      Work and such Derivative Works in Source or Object form.
-
-   3. Grant of Patent License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      (except as stated in this section) patent license to make, have made,
-      use, offer to sell, sell, import, and otherwise transfer the Work,
-      where such license applies only to those patent claims licensable
-      by such Contributor that are necessarily infringed by their
-      Contribution(s) alone or by combination of their Contribution(s)
-      with the Work to which such Contribution(s) was submitted. If You
-      institute patent litigation against any entity (including a
-      cross-claim or counterclaim in a lawsuit) alleging that the Work
-      or a Contribution incorporated within the Work constitutes direct
-      or contributory patent infringement, then any patent licenses
-      granted to You under this License for that Work shall terminate
-      as of the date such litigation is filed.
-
-   4. Redistribution. You may reproduce and distribute copies of the
-      Work or Derivative Works thereof in any medium, with or without
-      modifications, and in Source or Object form, provided that You
-      meet the following conditions:
-
-      (a) You must give any other recipients of the Work or
-          Derivative Works a copy of this License; and
-
-      (b) You must cause any modified files to carry prominent notices
-          stating that You changed the files; and
-
-      (c) You must retain, in the Source form of any Derivative Works
-          that You distribute, all copyright, patent, trademark, and
-          attribution notices from the Source form of the Work,
-          excluding those notices that do not pertain to any part of
-          the Derivative Works; and
-
-      (d) If the Work includes a "NOTICE" text file as part of its
-          distribution, then any Derivative Works that You distribute must
-          include a readable copy of the attribution notices contained
-          within such NOTICE file, excluding those notices that do not
-          pertain to any part of the Derivative Works, in at least one
-          of the following places: within a NOTICE text file distributed
-          as part of the Derivative Works; within the Source form or
-          documentation, if provided along with the Derivative Works; or,
-          within a display generated by the Derivative Works, if and
-          wherever such third-party notices normally appear. The contents
-          of the NOTICE file are for informational purposes only and
-          do not modify the License. You may add Your own attribution
-          notices within Derivative Works that You distribute, alongside
-          or as an addendum to the NOTICE text from the Work, provided
-          that such additional attribution notices cannot be construed
-          as modifying the License.
-
-      You may add Your own copyright statement to Your modifications and
-      may provide additional or different license terms and conditions
-      for use, reproduction, or distribution of Your modifications, or
-      for any such Derivative Works as a whole, provided Your use,
-      reproduction, and distribution of the Work otherwise complies with
-      the conditions stated in this License.
-
-   5. Submission of Contributions. Unless You explicitly state otherwise,
-      any Contribution intentionally submitted for inclusion in the Work
-      by You to the Licensor shall be under the terms and conditions of
-      this License, without any additional terms or conditions.
-      Notwithstanding the above, nothing herein shall supersede or modify
-      the terms of any separate license agreement you may have executed
-      with Licensor regarding such Contributions.
-
-   6. Trademarks. This License does not grant permission to use the trade
-      names, trademarks, service marks, or product names of the Licensor,
-      except as required for reasonable and customary use in describing the
-      origin of the Work and reproducing the content of the NOTICE file.
-
-   7. Disclaimer of Warranty. Unless required by applicable law or
-      agreed to in writing, Licensor provides the Work (and each
-      Contributor provides its Contributions) on an "AS IS" BASIS,
-      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
-      implied, including, without limitation, any warranties or conditions
-      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
-      PARTICULAR PURPOSE. You are solely responsible for determining the
-      appropriateness of using or redistributing the Work and assume any
-      risks associated with Your exercise of permissions under this License.
-
-   8. Limitation of Liability. In no event and under no legal theory,
-      whether in tort (including negligence), contract, or otherwise,
-      unless required by applicable law (such as deliberate and grossly
-      negligent acts) or agreed to in writing, shall any Contributor be
-      liable to You for damages, including any direct, indirect, special,
-      incidental, or consequential damages of any character arising as a
-      result of this License or out of the use or inability to use the
-      Work (including but not limited to damages for loss of goodwill,
-      work stoppage, computer failure or malfunction, or any and all
-      other commercial damages or losses), even if such Contributor
-      has been advised of the possibility of such damages.
-
-   9. Accepting Warranty or Additional Liability. While redistributing
-      the Work or Derivative Works thereof, You may choose to offer,
-      and charge a fee for, acceptance of support, warranty, indemnity,
-      or other liability obligations and/or rights consistent with this
-      License. However, in accepting such obligations, You may act only
-      on Your own behalf and on Your sole responsibility, not on behalf
-      of any other Contributor, and only if You agree to indemnify,
-      defend, and hold each Contributor harmless for any liability
-      incurred by, or claims asserted against, such Contributor by reason
-      of your accepting any such warranty or additional liability.
-
-   END OF TERMS AND CONDITIONS
-
-   APPENDIX: How to apply the Apache License to your work.
-
-      To apply the Apache License to your work, attach the following
-      boilerplate notice, with the fields enclosed by brackets "[]"
-      replaced with your own identifying information. (Don't include
-      the brackets!)  The text should be enclosed in the appropriate
-      comment syntax for the file format. We also recommend that a
-      file or class name and description of purpose be included on the
-      same "printed page" as the copyright notice for easier
-      identification within third-party archives.
-
-   Copyright 2020 - present, Facebook, Inc
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.

README.md

@@ -1,21 +1,4 @@
-# Le Robot
-
-#### State-of-the-art machine learning for real-world robotics
-
-Le Robot aims to provide models, datasets, and tools for real-world robotics in PyTorch. The goal is to lower the barrier for entry to robotics so that everyone can contribute and benefit from sharing datasets and pretrained models.
-
-Le Robot contains state-of-the-art approaches that have been shown to transfer to the real-world with a focus on imitation learning and reinforcement learning.
-
-Le Robot already provides a set of pretrained models, datasets with human collected demonstrations, and simulated environments so that everyone can get started. In the coming weeks, the plan is to add more and more supports for real-world robotics on the most affordable and capable robots out there.
-
-Le Robot is built upon [TorchRL](https://github.com/pytorch/rl) which provides abstractions and utilities for Reinforcement Learning.
-
-## Acknowledgment
-
-- Our ACT policy and ALOHA environment are adapted from [ALOHA](https://tonyzhaozh.github.io/aloha/)
-- Our Diffusion policy and Pusht environment are adapted from [Diffusion Policy](https://diffusion-policy.cs.columbia.edu/)
-- Our TDMPC policy and Simxarm environment are adapted from [FOWM](https://www.yunhaifeng.com/FOWM/)
-
+# LeRobot
 
 ## Installation
 
@@ -155,7 +138,7 @@ git lfs pull
 
 When adding a new dataset, mock it with
 ```
-python tests/scripts/mock_dataset.py --in-data-dir data/$DATASET --out-data-dir tests/data/$DATASET
+python tests/scripts/mock_dataset.py --in-data-dir data/<dataset_id> --out-data-dir tests/data/<dataset_id>
 ```
 
 Run tests
@@ -165,61 +148,24 @@ DATA_DIR="tests/data" pytest -sx tests
 
 **Datasets**
 
-To add a dataset to the hub, first login and use a token generated from [huggingface settings](https://huggingface.co/settings/tokens) with write access:
+To add a pytorch rl dataset to the hub, first login and use a token generated from [huggingface settings](https://huggingface.co/settings/tokens) with write access:
 ```
-huggingface-cli login --token ${HUGGINGFACE_TOKEN} --add-to-git-credential
+huggingface-cli login --token $HUGGINGFACE_TOKEN --add-to-git-credential
 ```
 
 Then you can upload it to the hub with:
 ```
-HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli upload $HF_USER/$DATASET data/$DATASET \
---repo-type dataset  \
---revision v1.0
+HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli upload --repo-type dataset $HF_USER/$DATASET data/$DATASET
 ```
 
-You will need to set the corresponding version as a default argument in your dataset class:
-```python
-  version: str | None = "v1.0",
-```
-See: [`lerobot/common/datasets/pusht.py`](https://github.com/Cadene/lerobot/blob/main/lerobot/common/datasets/pusht.py)
-
 For instance, for [cadene/pusht](https://huggingface.co/datasets/cadene/pusht), we used:
 ```
 HF_USER=cadene
 DATASET=pusht
 ```
 
-If you want to improve an existing dataset, you can download it locally with:
-```
-mkdir -p data/$DATASET
-HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download ${HF_USER}/$DATASET \
---repo-type dataset \
---local-dir data/$DATASET \
---local-dir-use-symlinks=False \
---revision v1.0
-```
 
-Iterate on your code and dataset with:
-```
-DATA_DIR=data python train.py
-```
-
-Upload a new version (v2.0 or v1.1 if the changes are respectively more or less significant):
-```
-HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli upload $HF_USER/$DATASET data/$DATASET \
---repo-type dataset \
---revision v1.1 \
---delete "*"
-```
-
-Then you will need to set the corresponding version as a default argument in your dataset class:
-```python
-  version: str | None = "v1.1",
-```
-See: [`lerobot/common/datasets/pusht.py`](https://github.com/Cadene/lerobot/blob/main/lerobot/common/datasets/pusht.py)
-
-
-Finally, you might want to mock the dataset if you need to update the unit tests as well:
-```
-python tests/scripts/mock_dataset.py --in-data-dir data/$DATASET --out-data-dir tests/data/$DATASET
-```
+## Acknowledgment
+- Our Diffusion policy and Pusht environment are adapted from [Diffusion Policy](https://diffusion-policy.cs.columbia.edu/)
+- Our TDMPC policy and Simxarm environment are adapted from [FOWM](https://www.yunhaifeng.com/FOWM/)
+- Our ACT policy and ALOHA environment are adapted from [ALOHA](https://tonyzhaozh.github.io/aloha/)

lerobot/common/datasets/abstract.py

@@ -19,7 +19,6 @@ class AbstractExperienceReplay(TensorDictReplayBuffer):
     def __init__(
         self,
         dataset_id: str,
-        version: str | None = None,
         batch_size: int = None,
         *,
         shuffle: bool = True,
@@ -32,15 +31,8 @@ class AbstractExperienceReplay(TensorDictReplayBuffer):
         transform: "torchrl.envs.Transform" = None,
     ):
         self.dataset_id = dataset_id
-        self.version = version
         self.shuffle = shuffle
         self.root = root
-
-        if self.root is not None and self.version is not None:
-            logging.warning(
-                f"The version of the dataset ({self.version}) is not enforced when root is provided ({self.root})."
-            )
-
         storage = self._download_or_load_dataset()
 
         super().__init__(
@@ -57,9 +49,9 @@ class AbstractExperienceReplay(TensorDictReplayBuffer):
     @property
     def stats_patterns(self) -> dict:
         return {
-            ("observation", "state"): "b c -> c",
-            ("observation", "image"): "b c h w -> c 1 1",
-            ("action",): "b c -> c",
+            ("observation", "state"): "b c -> 1 c",
+            ("observation", "image"): "b c h w -> 1 c 1 1",
+            ("action",): "b c -> 1 c",
         }
 
     @property
@@ -93,7 +85,7 @@ class AbstractExperienceReplay(TensorDictReplayBuffer):
             self._transform = transform
 
     def compute_or_load_stats(self, num_batch=100, batch_size=32) -> TensorDict:
-        stats_path = self.data_dir / "stats.pth"
+        stats_path = Path(self.data_dir) / "stats.pth"
         if stats_path.exists():
             stats = torch.load(stats_path)
         else:
@@ -104,14 +96,10 @@ class AbstractExperienceReplay(TensorDictReplayBuffer):
 
     def _download_or_load_dataset(self) -> torch.StorageBase:
         if self.root is None:
-            self.data_dir = Path(
-                snapshot_download(
-                    repo_id=f"cadene/{self.dataset_id}", repo_type="dataset", revision=self.version
-                )
-            )
+            self.data_dir = snapshot_download(repo_id=f"cadene/{self.dataset_id}", repo_type="dataset")
         else:
             self.data_dir = self.root / self.dataset_id
-        return TensorStorage(TensorDict.load_memmap(self.data_dir / "replay_buffer"))
+        return TensorStorage(TensorDict.load_memmap(self.data_dir))
 
     def _compute_stats(self, num_batch=100, batch_size=32):
         rb = TensorDictReplayBuffer(

lerobot/common/datasets/aloha.py

@@ -84,7 +84,6 @@ class AlohaExperienceReplay(AbstractExperienceReplay):
     def __init__(
         self,
         dataset_id: str,
-        version: str | None = "v1.1",
         batch_size: int = None,
         *,
         shuffle: bool = True,
@@ -100,7 +99,6 @@ class AlohaExperienceReplay(AbstractExperienceReplay):
 
         super().__init__(
             dataset_id,
-            version,
             batch_size,
             shuffle=shuffle,
             root=root,
@@ -115,11 +113,11 @@ class AlohaExperienceReplay(AbstractExperienceReplay):
     @property
     def stats_patterns(self) -> dict:
         d = {
-            ("observation", "state"): "b c -> c",
-            ("action",): "b c -> c",
+            ("observation", "state"): "b c -> 1 c",
+            ("action",): "b c -> 1 c",
         }
         for cam in CAMERAS[self.dataset_id]:
-            d[("observation", "image", cam)] = "b c h w -> c 1 1"
+            d[("observation", "image", cam)] = "b c h w -> 1 c 1 1"
         return d
 
     @property

lerobot/common/datasets/pusht.py

@@ -87,7 +87,6 @@ class PushtExperienceReplay(AbstractExperienceReplay):
     def __init__(
         self,
         dataset_id: str,
-        version: str | None = "v1.1",
         batch_size: int = None,
         *,
         shuffle: bool = True,
@@ -101,7 +100,6 @@ class PushtExperienceReplay(AbstractExperienceReplay):
     ):
         super().__init__(
             dataset_id,
-            version,
             batch_size,
             shuffle=shuffle,
             root=root,

lerobot/common/datasets/simxarm.py

@@ -40,7 +40,6 @@ class SimxarmExperienceReplay(AbstractExperienceReplay):
     def __init__(
         self,
         dataset_id: str,
-        version: str | None = None,
         batch_size: int = None,
         *,
         shuffle: bool = True,
@@ -54,7 +53,6 @@ class SimxarmExperienceReplay(AbstractExperienceReplay):
     ):
         super().__init__(
             dataset_id,
-            version,
             batch_size,
             shuffle=shuffle,
             root=root,

lerobot/common/envs/abstract.py

@@ -1,3 +1,4 @@
+import abc
 from collections import deque
 from typing import Optional
 
@@ -26,6 +27,7 @@ class AbstractEnv(EnvBase):
         self.image_size = image_size
         self.num_prev_obs = num_prev_obs
         self.num_prev_action = num_prev_action
+        self._rendering_hooks = []
 
         if pixels_only:
             assert from_pixels
@@ -43,20 +45,36 @@ class AbstractEnv(EnvBase):
             raise NotImplementedError()
             # self._prev_action_queue = deque(maxlen=self.num_prev_action)
 
+    def register_rendering_hook(self, func):
+        self._rendering_hooks.append(func)
+
+    def call_rendering_hooks(self):
+        for func in self._rendering_hooks:
+            func(self)
+
+    def reset_rendering_hooks(self):
+        self._rendering_hooks = []
+
+    @abc.abstractmethod
     def render(self, mode="rgb_array", width=640, height=480):
-        raise NotImplementedError("Abstract method")
+        raise NotImplementedError()
 
+    @abc.abstractmethod
     def _reset(self, tensordict: Optional[TensorDict] = None):
-        raise NotImplementedError("Abstract method")
+        raise NotImplementedError()
 
+    @abc.abstractmethod
     def _step(self, tensordict: TensorDict):
-        raise NotImplementedError("Abstract method")
+        raise NotImplementedError()
 
+    @abc.abstractmethod
     def _make_env(self):
-        raise NotImplementedError("Abstract method")
+        raise NotImplementedError()
 
+    @abc.abstractmethod
     def _make_spec(self):
-        raise NotImplementedError("Abstract method")
+        raise NotImplementedError()
 
+    @abc.abstractmethod
     def _set_seed(self, seed: Optional[int]):
-        raise NotImplementedError("Abstract method")
+        raise NotImplementedError()

lerobot/common/envs/aloha/env.py

@@ -35,8 +35,6 @@ _has_gym = importlib.util.find_spec("gym") is not None
 
 
 class AlohaEnv(AbstractEnv):
-    _reset_warning_issued = False
-
     def __init__(
         self,
         task,
@@ -122,76 +120,90 @@ class AlohaEnv(AbstractEnv):
         return obs
 
     def _reset(self, tensordict: Optional[TensorDict] = None):
-        if tensordict is not None and not AlohaEnv._reset_warning_issued:
-            logging.warning(f"{self.__class__.__name__}._reset ignores the provided tensordict.")
-            AlohaEnv._reset_warning_issued = True
+        td = tensordict
+        if td is None or td.is_empty():
+            # we need to handle seed iteration, since self._env.reset() rely an internal _seed.
+            self._current_seed += 1
+            self.set_seed(self._current_seed)
 
-        # we need to handle seed iteration, since self._env.reset() rely an internal _seed.
-        self._current_seed += 1
-        self.set_seed(self._current_seed)
+            # TODO(rcadene): do not use global variable for this
+            if "sim_transfer_cube" in self.task:
+                BOX_POSE[0] = sample_box_pose()  # used in sim reset
+            elif "sim_insertion" in self.task:
+                BOX_POSE[0] = np.concatenate(sample_insertion_pose())  # used in sim reset
 
-        # TODO(rcadene): do not use global variable for this
-        if "sim_transfer_cube" in self.task:
-            BOX_POSE[0] = sample_box_pose()  # used in sim reset
-        elif "sim_insertion" in self.task:
-            BOX_POSE[0] = np.concatenate(sample_insertion_pose())  # used in sim reset
+            raw_obs = self._env.reset()
+            # TODO(rcadene): add assert
+            # assert self._current_seed == self._env._seed
 
-        raw_obs = self._env.reset()
-        # TODO(rcadene): add assert
-        # assert self._current_seed == self._env._seed
+            obs = self._format_raw_obs(raw_obs.observation)
 
-        obs = self._format_raw_obs(raw_obs.observation)
+            if self.num_prev_obs > 0:
+                stacked_obs = {}
+                if "image" in obs:
+                    self._prev_obs_image_queue = deque(
+                        [obs["image"]["top"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
+                    )
+                    stacked_obs["image"] = {"top": torch.stack(list(self._prev_obs_image_queue))}
+                if "state" in obs:
+                    self._prev_obs_state_queue = deque(
+                        [obs["state"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
+                    )
+                    stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
+                obs = stacked_obs
 
-        if self.num_prev_obs > 0:
-            stacked_obs = {}
-            if "image" in obs:
-                self._prev_obs_image_queue = deque(
-                    [obs["image"]["top"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
-                )
-                stacked_obs["image"] = {"top": torch.stack(list(self._prev_obs_image_queue))}
-            if "state" in obs:
-                self._prev_obs_state_queue = deque(
-                    [obs["state"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
-                )
-                stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
-            obs = stacked_obs
-
-        td = TensorDict(
-            {
-                "observation": TensorDict(obs, batch_size=[]),
-                "done": torch.tensor([False], dtype=torch.bool),
-            },
-            batch_size=[],
-        )
+            td = TensorDict(
+                {
+                    "observation": TensorDict(obs, batch_size=[]),
+                    "done": torch.tensor([False], dtype=torch.bool),
+                },
+                batch_size=[],
+            )
+        else:
+            raise NotImplementedError()
 
+        self.call_rendering_hooks()
         return td
 
     def _step(self, tensordict: TensorDict):
         td = tensordict
         action = td["action"].numpy()
-        assert action.ndim == 1
+        # step expects shape=(4,) so we pad if necessary
         # TODO(rcadene): add info["is_success"] and info["success"] ?
+        sum_reward = 0
 
-        _, reward, _, raw_obs = self._env.step(action)
+        if action.ndim == 1:
+            action = einops.repeat(action, "c -> t c", t=self.frame_skip)
+        else:
+            if self.frame_skip > 1:
+                raise NotImplementedError()
 
-        # TODO(rcadene): add an enum
-        success = done = reward == 4
-        obs = self._format_raw_obs(raw_obs)
+        num_action_steps = action.shape[0]
+        for i in range(num_action_steps):
+            _, reward, discount, raw_obs = self._env.step(action[i])
+            del discount  # not used
 
-        if self.num_prev_obs > 0:
-            stacked_obs = {}
-            if "image" in obs:
-                self._prev_obs_image_queue.append(obs["image"]["top"])
-                stacked_obs["image"] = {"top": torch.stack(list(self._prev_obs_image_queue))}
-            if "state" in obs:
-                self._prev_obs_state_queue.append(obs["state"])
-                stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
-            obs = stacked_obs
+            # TOOD(rcadene): add an enum
+            success = done = reward == 4
+            sum_reward += reward
+            obs = self._format_raw_obs(raw_obs)
+
+            if self.num_prev_obs > 0:
+                stacked_obs = {}
+                if "image" in obs:
+                    self._prev_obs_image_queue.append(obs["image"]["top"])
+                    stacked_obs["image"] = {"top": torch.stack(list(self._prev_obs_image_queue))}
+                if "state" in obs:
+                    self._prev_obs_state_queue.append(obs["state"])
+                    stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
+                obs = stacked_obs
+
+            self.call_rendering_hooks()
 
         td = TensorDict(
             {
                 "observation": TensorDict(obs, batch_size=[]),
-                "reward": torch.tensor([reward], dtype=torch.float32),
+                "reward": torch.tensor([sum_reward], dtype=torch.float32),
                 # succes and done are true when coverage > self.success_threshold in env
                 "done": torch.tensor([done], dtype=torch.bool),
                 "success": torch.tensor([success], dtype=torch.bool),

lerobot/common/envs/factory.py

@@ -1,18 +1,14 @@
-from torchrl.envs import SerialEnv
 from torchrl.envs.transforms import Compose, StepCounter, Transform, TransformedEnv
 
 
 def make_env(cfg, transform=None):
-    """
-    Note: The returned environment is wrapped in a torchrl.SerialEnv with cfg.rollout_batch_size underlying
-    environments. The env therefore returns batches.`
-    """
-
     kwargs = {
         "frame_skip": cfg.env.action_repeat,
         "from_pixels": cfg.env.from_pixels,
         "pixels_only": cfg.env.pixels_only,
         "image_size": cfg.env.image_size,
+        # TODO(rcadene): do we want a specific eval_env_seed?
+        "seed": cfg.seed,
         "num_prev_obs": cfg.n_obs_steps - 1,
     }
 
@@ -35,33 +31,22 @@ def make_env(cfg, transform=None):
     else:
         raise ValueError(cfg.env.name)
 
-    def _make_env(seed):
-        nonlocal kwargs
-        kwargs["seed"] = seed
-        env = clsfunc(**kwargs)
+    env = clsfunc(**kwargs)
 
-        # limit rollout to max_steps
-        env = TransformedEnv(env, StepCounter(max_steps=cfg.env.episode_length))
+    # limit rollout to max_steps
+    env = TransformedEnv(env, StepCounter(max_steps=cfg.env.episode_length))
 
-        if transform is not None:
-            # useful to add normalization
-            if isinstance(transform, Compose):
-                for tf in transform:
-                    env.append_transform(tf.clone())
-            elif isinstance(transform, Transform):
-                env.append_transform(transform.clone())
-            else:
-                raise NotImplementedError()
+    if transform is not None:
+        # useful to add normalization
+        if isinstance(transform, Compose):
+            for tf in transform:
+                env.append_transform(tf.clone())
+        elif isinstance(transform, Transform):
+            env.append_transform(transform.clone())
+        else:
+            raise NotImplementedError()
 
-        return env
-
-    return SerialEnv(
-        cfg.rollout_batch_size,
-        create_env_fn=_make_env,
-        create_env_kwargs=[
-            {"seed": env_seed} for env_seed in range(cfg.seed, cfg.seed + cfg.rollout_batch_size)
-        ],
-    )
+    return env
 
 
 # def make_env(env_name, frame_skip, device, is_test=False):

lerobot/common/envs/pusht/env.py

@@ -1,8 +1,8 @@
 import importlib
-import logging
 from collections import deque
 from typing import Optional
 
+import einops
 import torch
 from tensordict import TensorDict
 from torchrl.data.tensor_specs import (
@@ -20,8 +20,6 @@ _has_gym = importlib.util.find_spec("gym") is not None
 
 
 class PushtEnv(AbstractEnv):
-    _reset_warning_issued = False
-
     def __init__(
         self,
         task="pusht",
@@ -82,67 +80,80 @@ class PushtEnv(AbstractEnv):
         return obs
 
     def _reset(self, tensordict: Optional[TensorDict] = None):
-        if tensordict is not None and not PushtEnv._reset_warning_issued:
-            logging.warning(f"{self.__class__.__name__}._reset ignores the provided tensordict.")
-            PushtEnv._reset_warning_issued = True
+        td = tensordict
+        if td is None or td.is_empty():
+            # we need to handle seed iteration, since self._env.reset() rely an internal _seed.
+            self._current_seed += 1
+            self.set_seed(self._current_seed)
+            raw_obs = self._env.reset()
+            assert self._current_seed == self._env._seed
 
-        # we need to handle seed iteration, since self._env.reset() rely an internal _seed.
-        self._current_seed += 1
-        self.set_seed(self._current_seed)
-        raw_obs = self._env.reset()
-        assert self._current_seed == self._env._seed
+            obs = self._format_raw_obs(raw_obs)
 
-        obs = self._format_raw_obs(raw_obs)
+            if self.num_prev_obs > 0:
+                stacked_obs = {}
+                if "image" in obs:
+                    self._prev_obs_image_queue = deque(
+                        [obs["image"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
+                    )
+                    stacked_obs["image"] = torch.stack(list(self._prev_obs_image_queue))
+                if "state" in obs:
+                    self._prev_obs_state_queue = deque(
+                        [obs["state"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
+                    )
+                    stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
+                obs = stacked_obs
 
-        if self.num_prev_obs > 0:
-            stacked_obs = {}
-            if "image" in obs:
-                self._prev_obs_image_queue = deque(
-                    [obs["image"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
-                )
-                stacked_obs["image"] = torch.stack(list(self._prev_obs_image_queue))
-            if "state" in obs:
-                self._prev_obs_state_queue = deque(
-                    [obs["state"]] * (self.num_prev_obs + 1), maxlen=(self.num_prev_obs + 1)
-                )
-                stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
-            obs = stacked_obs
-
-        td = TensorDict(
-            {
-                "observation": TensorDict(obs, batch_size=[]),
-                "done": torch.tensor([False], dtype=torch.bool),
-            },
-            batch_size=[],
-        )
+            td = TensorDict(
+                {
+                    "observation": TensorDict(obs, batch_size=[]),
+                    "done": torch.tensor([False], dtype=torch.bool),
+                },
+                batch_size=[],
+            )
+        else:
+            raise NotImplementedError()
 
+        self.call_rendering_hooks()
         return td
 
     def _step(self, tensordict: TensorDict):
         td = tensordict
         action = td["action"].numpy()
-        assert action.ndim == 1
+        # step expects shape=(4,) so we pad if necessary
         # TODO(rcadene): add info["is_success"] and info["success"] ?
+        sum_reward = 0
 
-        raw_obs, reward, done, info = self._env.step(action)
+        if action.ndim == 1:
+            action = einops.repeat(action, "c -> t c", t=self.frame_skip)
+        else:
+            if self.frame_skip > 1:
+                raise NotImplementedError()
 
-        obs = self._format_raw_obs(raw_obs)
+        num_action_steps = action.shape[0]
+        for i in range(num_action_steps):
+            raw_obs, reward, done, info = self._env.step(action[i])
+            sum_reward += reward
 
-        if self.num_prev_obs > 0:
-            stacked_obs = {}
-            if "image" in obs:
-                self._prev_obs_image_queue.append(obs["image"])
-                stacked_obs["image"] = torch.stack(list(self._prev_obs_image_queue))
-            if "state" in obs:
-                self._prev_obs_state_queue.append(obs["state"])
-                stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
-            obs = stacked_obs
+            obs = self._format_raw_obs(raw_obs)
+
+            if self.num_prev_obs > 0:
+                stacked_obs = {}
+                if "image" in obs:
+                    self._prev_obs_image_queue.append(obs["image"])
+                    stacked_obs["image"] = torch.stack(list(self._prev_obs_image_queue))
+                if "state" in obs:
+                    self._prev_obs_state_queue.append(obs["state"])
+                    stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
+                obs = stacked_obs
+
+            self.call_rendering_hooks()
 
         td = TensorDict(
             {
                 "observation": TensorDict(obs, batch_size=[]),
-                "reward": torch.tensor([reward], dtype=torch.float32),
-                # success and done are true when coverage > self.success_threshold in env
+                "reward": torch.tensor([sum_reward], dtype=torch.float32),
+                # succes and done are true when coverage > self.success_threshold in env
                 "done": torch.tensor([done], dtype=torch.bool),
                 "success": torch.tensor([done], dtype=torch.bool),
             },

lerobot/common/envs/simxarm.py

@@ -118,6 +118,7 @@ class SimxarmEnv(AbstractEnv):
         else:
             raise NotImplementedError()
 
+        self.call_rendering_hooks()
         return td
 
     def _step(self, tensordict: TensorDict):
@@ -151,6 +152,8 @@ class SimxarmEnv(AbstractEnv):
                     stacked_obs["state"] = torch.stack(list(self._prev_obs_state_queue))
                 obs = stacked_obs
 
+            self.call_rendering_hooks()
+
         td = TensorDict(
             {
                 "observation": self._format_raw_obs(raw_obs),

lerobot/common/logger.py

@@ -30,7 +30,6 @@ class Logger:
         self._model_dir = self._log_dir / "models"
         self._buffer_dir = self._log_dir / "buffers"
         self._save_model = cfg.save_model
-        self._disable_wandb_artifact = cfg.wandb.disable_artifact
         self._save_buffer = cfg.save_buffer
         self._group = cfg_to_group(cfg)
         self._seed = cfg.seed
@@ -72,10 +71,9 @@ class Logger:
             self._model_dir.mkdir(parents=True, exist_ok=True)
             fp = self._model_dir / f"{str(identifier)}.pt"
             policy.save(fp)
-            if self._wandb and not self._disable_wandb_artifact:
-                # note wandb artifact does not accept ":" in its name
+            if self._wandb:
                 artifact = self._wandb.Artifact(
-                    self._group.replace(":", "_") + "-" + str(self._seed) + "-" + str(identifier),
+                    self._group + "-" + str(self._seed) + "-" + str(identifier),
                     type="model",
                 )
                 artifact.add_file(fp)

lerobot/common/policies/abstract.py

@@ -1,70 +0,0 @@
-from collections import deque
-
-import torch
-from torch import Tensor, nn
-
-
-class AbstractPolicy(nn.Module):
-    """Base policy which all policies should be derived from.
-
-    The forward method should generally not be overriden as it plays the role of handling multi-step policies. See its
-    documentation for more information.
-    """
-
-    def __init__(self, n_action_steps: int | None):
-        """
-        n_action_steps: Sets the cache size for storing action trajectories. If None, it is assumed that a single
-            action is returned by `select_actions` and that doesn't have a horizon dimension. The `forward` method then
-            adds that dimension.
-        """
-        super().__init__()
-        self.n_action_steps = n_action_steps
-        self.clear_action_queue()
-
-    def update(self, replay_buffer, step):
-        """One step of the policy's learning algorithm."""
-        raise NotImplementedError("Abstract method")
-
-    def save(self, fp):
-        torch.save(self.state_dict(), fp)
-
-    def load(self, fp):
-        d = torch.load(fp)
-        self.load_state_dict(d)
-
-    def select_actions(self, observation) -> Tensor:
-        """Select an action (or trajectory of actions) based on an observation during rollout.
-
-        If n_action_steps was provided at initialization, this should return a (batch_size, n_action_steps, *) tensor of
-        actions. Otherwise if n_actions_steps is None, this should return a (batch_size, *) tensor of actions.
-        """
-        raise NotImplementedError("Abstract method")
-
-    def clear_action_queue(self):
-        """This should be called whenever the environment is reset."""
-        if self.n_action_steps is not None:
-            self._action_queue = deque([], maxlen=self.n_action_steps)
-
-    def forward(self, *args, **kwargs) -> Tensor:
-        """Inference step that makes multi-step policies compatible with their single-step environments.
-
-        WARNING: In general, this should not be overriden.
-
-        Consider a "policy" that observes the environment then charts a course of N actions to take. To make this fit
-        into the formalism of a TorchRL environment, we view it as being effectively a policy that (1) makes an
-        observation and prepares a queue of actions, (2) consumes that queue when queried, regardless of the environment
-        observation, (3) repopulates the action queue when empty. This method handles the aforementioned logic so that
-        the subclass doesn't have to.
-
-        This method effectively wraps the `select_actions` method of the subclass. The following assumptions are made:
-        1. The `select_actions` method returns a Tensor of actions with shape (B, H, *) where B is the batch size, H is
-           the action trajectory horizon and * is the action dimensions.
-        2. Prior to the `select_actions` method being called, theres is an `n_action_steps` instance attribute defined.
-        """
-        if self.n_action_steps is None:
-            return self.select_actions(*args, **kwargs)
-        if len(self._action_queue) == 0:
-            # `select_actions` returns a (batch_size, n_action_steps, *) tensor, but the queue effectively has shape
-            # (n_action_steps, batch_size, *), hence the transpose.
-            self._action_queue.extend(self.select_actions(*args, **kwargs).transpose(0, 1))
-        return self._action_queue.popleft()

lerobot/common/policies/act/policy.py

@@ -2,10 +2,10 @@ import logging
 import time
 
 import torch
+import torch.nn as nn
 import torch.nn.functional as F  # noqa: N812
 import torchvision.transforms as transforms
 
-from lerobot.common.policies.abstract import AbstractPolicy
 from lerobot.common.policies.act.detr_vae import build
 
 
@@ -40,9 +40,9 @@ def kl_divergence(mu, logvar):
     return total_kld, dimension_wise_kld, mean_kld
 
 
-class ActionChunkingTransformerPolicy(AbstractPolicy):
+class ActionChunkingTransformerPolicy(nn.Module):
     def __init__(self, cfg, device, n_action_steps=1):
-        super().__init__(n_action_steps)
+        super().__init__()
         self.cfg = cfg
         self.n_action_steps = n_action_steps
         self.device = device
@@ -147,15 +147,16 @@ class ActionChunkingTransformerPolicy(AbstractPolicy):
         return loss
 
     @torch.no_grad()
-    def select_actions(self, observation, step_count):
-        if observation["image"].shape[0] != 1:
-            raise NotImplementedError("Batch size > 1 not handled")
-
+    def forward(self, observation, step_count):
         # TODO(rcadene): remove unused step_count
         del step_count
 
         self.eval()
 
+        # TODO(rcadene): remove unsqueeze hack to add bsize=1
+        observation["image", "top"] = observation["image", "top"].unsqueeze(0)
+        # observation["state"] = observation["state"].unsqueeze(0)
+
         # TODO(rcadene): remove hack
         # add 1 camera dimension
         observation["image", "top"] = observation["image", "top"].unsqueeze(1)
@@ -179,8 +180,11 @@ class ActionChunkingTransformerPolicy(AbstractPolicy):
             # exp_weights = torch.from_numpy(exp_weights).cuda().unsqueeze(dim=1)
             # raw_action = (actions_for_curr_step * exp_weights).sum(dim=0, keepdim=True)
 
+        # remove bsize=1
+        action = action.squeeze(0)
+
         # take first predicted action or n first actions
-        action = action[: self.n_action_steps]
+        action = action[0] if self.n_action_steps == 1 else action[: self.n_action_steps]
         return action
 
     def _forward(self, qpos, image, actions=None, is_pad=None):

lerobot/common/policies/diffusion/policy.py

@@ -3,14 +3,14 @@ import time
 
 import hydra
 import torch
+import torch.nn as nn
 
-from lerobot.common.policies.abstract import AbstractPolicy
 from lerobot.common.policies.diffusion.diffusion_unet_image_policy import DiffusionUnetImagePolicy
 from lerobot.common.policies.diffusion.model.lr_scheduler import get_scheduler
 from lerobot.common.policies.diffusion.model.multi_image_obs_encoder import MultiImageObsEncoder
 
 
-class DiffusionPolicy(AbstractPolicy):
+class DiffusionPolicy(nn.Module):
     def __init__(
         self,
         cfg,
@@ -34,7 +34,7 @@ class DiffusionPolicy(AbstractPolicy):
         # parameters passed to step
         **kwargs,
     ):
-        super().__init__(n_action_steps)
+        super().__init__()
         self.cfg = cfg
 
         noise_scheduler = hydra.utils.instantiate(cfg_noise_scheduler)
@@ -93,16 +93,21 @@ class DiffusionPolicy(AbstractPolicy):
         )
 
     @torch.no_grad()
-    def select_actions(self, observation, step_count):
+    def forward(self, observation, step_count):
         # TODO(rcadene): remove unused step_count
         del step_count
 
+        # TODO(rcadene): remove unsqueeze hack to add bsize=1
+        observation["image"] = observation["image"].unsqueeze(0)
+        observation["state"] = observation["state"].unsqueeze(0)
+
         obs_dict = {
             "image": observation["image"],
             "agent_pos": observation["state"],
         }
         out = self.diffusion.predict_action(obs_dict)
-        action = out["action"]
+
+        action = out["action"].squeeze(0)
         return action
 
     def update(self, replay_buffer, step):

lerobot/common/policies/factory.py

@@ -1,7 +1,4 @@
 def make_policy(cfg):
-    if cfg.policy.name != "diffusion" and cfg.rollout_batch_size > 1:
-        raise NotImplementedError("Only diffusion policy supports rollout_batch_size > 1 for the time being.")
-
     if cfg.policy.name == "tdmpc":
         from lerobot.common.policies.tdmpc.policy import TDMPC
 

lerobot/common/policies/tdmpc/policy.py

@@ -9,7 +9,6 @@ import torch
 import torch.nn as nn
 
 import lerobot.common.policies.tdmpc.helper as h
-from lerobot.common.policies.abstract import AbstractPolicy
 
 FIRST_FRAME = 0
 
@@ -86,11 +85,11 @@ class TOLD(nn.Module):
         return torch.min(Q1, Q2) if return_type == "min" else (Q1 + Q2) / 2
 
 
-class TDMPC(AbstractPolicy):
+class TDMPC(nn.Module):
     """Implementation of TD-MPC learning + inference."""
 
     def __init__(self, cfg, device):
-        super().__init__(None)
+        super().__init__()
         self.action_dim = cfg.action_dim
 
         self.cfg = cfg
@@ -125,19 +124,20 @@ class TDMPC(AbstractPolicy):
         self.model_target.load_state_dict(d["model_target"])
 
     @torch.no_grad()
-    def select_actions(self, observation, step_count):
-        if observation["image"].shape[0] != 1:
-            raise NotImplementedError("Batch size > 1 not handled")
-
+    def forward(self, observation, step_count):
         t0 = step_count.item() == 0
 
+        # TODO(rcadene): remove unsqueeze hack...
+        if observation["image"].ndim == 3:
+            observation["image"] = observation["image"].unsqueeze(0)
+            observation["state"] = observation["state"].unsqueeze(0)
+
         obs = {
             # TODO(rcadene): remove contiguous hack...
             "rgb": observation["image"].contiguous(),
             "state": observation["state"].contiguous(),
         }
-        # Note: unsqueeze needed because `act` still uses non-batch logic.
-        action = self.act(obs, t0=t0, step=self.step.item()).unsqueeze(0)
+        action = self.act(obs, t0=t0, step=self.step.item())
         return action
 
     @torch.no_grad()

lerobot/configs/default.yaml

@@ -10,9 +10,6 @@ hydra:
     name: default
 
 seed: 1337
-# batch size for TorchRL SerialEnv. Each underlying env will get the seed = seed + env_index
-# NOTE: only diffusion policy supports rollout_batch_size > 1
-rollout_batch_size: 1
 device: cuda  # cpu
 prefetch: 4
 eval_freq: ???
@@ -33,7 +30,5 @@ policy: ???
 
 wandb:
   enable: true
-  # Set to true to disable saving an artifact despite save_model == True
-  disable_artifact: false
   project: lerobot
   notes: ""

lerobot/scripts/eval.py

@@ -3,7 +3,6 @@ import threading
 import time
 from pathlib import Path
 
-import einops
 import hydra
 import imageio
 import numpy as np
@@ -11,12 +10,10 @@ import torch
 import tqdm
 from tensordict.nn import TensorDictModule
 from torchrl.envs import EnvBase
-from torchrl.envs.batched_envs import BatchedEnvBase
 
 from lerobot.common.datasets.factory import make_offline_buffer
 from lerobot.common.envs.factory import make_env
 from lerobot.common.logger import log_output_dir
-from lerobot.common.policies.abstract import AbstractPolicy
 from lerobot.common.policies.factory import make_policy
 from lerobot.common.utils import init_logging, set_seed
 
@@ -26,8 +23,8 @@ def write_video(video_path, stacked_frames, fps):
 
 
 def eval_policy(
-    env: BatchedEnvBase,
-    policy: AbstractPolicy,
+    env: EnvBase,
+    policy: TensorDictModule = None,
     num_episodes: int = 10,
     max_steps: int = 30,
     save_video: bool = False,
@@ -35,80 +32,59 @@ def eval_policy(
     fps: int = 15,
     return_first_video: bool = False,
 ):
-    policy.eval()
     start = time.time()
     sum_rewards = []
     max_rewards = []
     successes = []
-    threads = []  # for video saving threads
-    episode_counter = 0  # for saving the correct number of videos
-
-    # TODO(alexander-soare): if num_episodes is not evenly divisible by the batch size, this will do more work than
-    # needed as I'm currently taking a ceil.
-    for i in tqdm.tqdm(range(-(-num_episodes // env.batch_size[0]))):
+    threads = []
+    for i in tqdm.tqdm(range(num_episodes)):
         ep_frames = []
+        if save_video or (return_first_video and i == 0):
 
-        def maybe_render_frame(env: EnvBase, _):
-            if save_video or (return_first_video and i == 0):  # noqa: B023
+            def render_frame(env):
                 ep_frames.append(env.render())  # noqa: B023
 
+            env.register_rendering_hook(render_frame)
+
         with torch.inference_mode():
-            # TODO(alexander-soare): When `break_when_any_done == False` this rolls out for max_steps even when all
-            # envs are done the first time. But we only use the first rollout. This is a waste of compute.
-            policy.clear_action_queue()
             rollout = env.rollout(
                 max_steps=max_steps,
                 policy=policy,
                 auto_cast_to_device=True,
-                callback=maybe_render_frame,
-                break_when_any_done=env.batch_size[0] == 1,
             )
-        # Figure out where in each rollout sequence the first done condition was encountered (results after this won't
-        # be included).
-        # Note: this assumes that the shape of the done key is (batch_size, max_steps, 1).
-        # Note: this relies on a property of argmax: that it returns the first occurrence as a tiebreaker.
-        rollout_steps = rollout["next", "done"].shape[1]
-        done_indices = torch.argmax(rollout["next", "done"].to(int), axis=1)  # (batch_size, rollout_steps)
-        mask = (torch.arange(rollout_steps) <= done_indices).unsqueeze(-1)  # (batch_size, rollout_steps, 1)
-        batch_sum_reward = einops.reduce((rollout["next", "reward"] * mask), "b n 1 -> b", "sum")
-        batch_max_reward = einops.reduce((rollout["next", "reward"] * mask), "b n 1 -> b", "max")
-        batch_success = einops.reduce((rollout["next", "success"] * mask), "b n 1 -> b", "any")
-        sum_rewards.extend(batch_sum_reward.tolist())
-        max_rewards.extend(batch_max_reward.tolist())
-        successes.extend(batch_success.tolist())
+        # print(", ".join([f"{x:.3f}" for x in rollout["next", "reward"][:,0].tolist()]))
+        ep_sum_reward = rollout["next", "reward"].sum()
+        ep_max_reward = rollout["next", "reward"].max()
+        ep_success = rollout["next", "success"].any()
+        sum_rewards.append(ep_sum_reward.item())
+        max_rewards.append(ep_max_reward.item())
+        successes.append(ep_success.item())
 
         if save_video or (return_first_video and i == 0):
-            batch_stacked_frames = np.stack(ep_frames)  # (t, b, *)
-            batch_stacked_frames = batch_stacked_frames.transpose(
-                1, 0, *range(2, batch_stacked_frames.ndim)
-            )  # (b, t, *)
+            stacked_frames = np.stack(ep_frames)
 
             if save_video:
-                for stacked_frames, done_index in zip(
-                    batch_stacked_frames, done_indices.flatten().tolist(), strict=False
-                ):
-                    if episode_counter >= num_episodes:
-                        continue
-                    video_dir.mkdir(parents=True, exist_ok=True)
-                    video_path = video_dir / f"eval_episode_{episode_counter}.mp4"
-                    thread = threading.Thread(
-                        target=write_video,
-                        args=(str(video_path), stacked_frames[:done_index], fps),
-                    )
-                    thread.start()
-                    threads.append(thread)
-                    episode_counter += 1
+                video_dir.mkdir(parents=True, exist_ok=True)
+                video_path = video_dir / f"eval_episode_{i}.mp4"
+                thread = threading.Thread(
+                    target=write_video,
+                    args=(str(video_path), stacked_frames, fps),
+                )
+                thread.start()
+                threads.append(thread)
 
             if return_first_video and i == 0:
-                first_video = batch_stacked_frames[0].transpose(0, 3, 1, 2)
+                first_video = stacked_frames.transpose(0, 3, 1, 2)
+
+    env.reset_rendering_hooks()
 
     for thread in threads:
         thread.join()
 
     info = {
-        "avg_sum_reward": np.nanmean(sum_rewards[:num_episodes]),
-        "avg_max_reward": np.nanmean(max_rewards[:num_episodes]),
-        "pc_success": np.nanmean(successes[:num_episodes]) * 100,
+        "avg_sum_reward": np.nanmean(sum_rewards),
+        "avg_max_reward": np.nanmean(max_rewards),
+        "pc_success": np.nanmean(successes) * 100,
         "eval_s": time.time() - start,
         "eval_ep_s": (time.time() - start) / num_episodes,
     }
@@ -162,7 +138,7 @@ def eval(cfg: dict, out_dir=None):
         save_video=True,
         video_dir=Path(out_dir) / "eval",
         fps=cfg.env.fps,
-        max_steps=cfg.env.episode_length,
+        max_steps=cfg.env.episode_length // cfg.n_action_steps,
         num_episodes=cfg.eval_episodes,
     )
     print(metrics)

lerobot/scripts/train.py

@@ -112,8 +112,6 @@ def train(cfg: dict, out_dir=None, job_name=None):
         raise NotImplementedError()
     if job_name is None:
         raise NotImplementedError()
-    if cfg.online_steps > 0:
-        assert cfg.rollout_batch_size == 1, "rollout_batch_size > 1 not supported for online training steps"
 
     init_logging()
 
@@ -183,7 +181,6 @@ def train(cfg: dict, out_dir=None, job_name=None):
         if offline_step == 0:
             logging.info("Start offline training on a fixed dataset")
         # TODO(rcadene): is it ok if step_t=0 = 0 and not 1 as previously done?
-        policy.train()
         train_info = policy.update(offline_buffer, step)
         if step % cfg.log_freq == 0:
             log_train_info(logger, train_info, step, cfg, offline_buffer, is_offline)
@@ -194,7 +191,7 @@ def train(cfg: dict, out_dir=None, job_name=None):
                 env,
                 td_policy,
                 num_episodes=cfg.eval_episodes,
-                max_steps=cfg.env.episode_length,
+                max_steps=cfg.env.episode_length // cfg.n_action_steps,
                 return_first_video=True,
                 video_dir=Path(out_dir) / "eval",
                 save_video=True,
@@ -220,11 +217,11 @@ def train(cfg: dict, out_dir=None, job_name=None):
         # TODO: add configurable number of rollout? (default=1)
         with torch.no_grad():
             rollout = env.rollout(
-                max_steps=cfg.env.episode_length,
+                max_steps=cfg.env.episode_length // cfg.n_action_steps,
                 policy=td_policy,
                 auto_cast_to_device=True,
             )
-        assert len(rollout) <= cfg.env.episode_length
+        assert len(rollout) <= cfg.env.episode_length // cfg.n_action_steps
         # set same episode index for all time steps contained in this rollout
         rollout["episode"] = torch.tensor([env_step] * len(rollout), dtype=torch.int)
         online_buffer.extend(rollout)

poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 1.8.2 and should not be changed by hand.
+# This file is automatically @generated by Poetry 1.8.1 and should not be changed by hand.
 
 [[package]]
 name = "absl-py"
@@ -658,13 +658,13 @@ typing = ["typing-extensions (>=4.8)"]
 
 [[package]]
 name = "fsspec"
-version = "2024.3.1"
+version = "2024.2.0"
 description = "File-system specification"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "fsspec-2024.3.1-py3-none-any.whl", hash = "sha256:918d18d41bf73f0e2b261824baeb1b124bcf771767e3a26425cd7dec3332f512"},
-    {file = "fsspec-2024.3.1.tar.gz", hash = "sha256:f39780e282d7d117ffb42bb96992f8a90795e4d0fb0f661a70ca39fe9c43ded9"},
+    {file = "fsspec-2024.2.0-py3-none-any.whl", hash = "sha256:817f969556fa5916bc682e02ca2045f96ff7f586d45110fcb76022063ad2c7d8"},
+    {file = "fsspec-2024.2.0.tar.gz", hash = "sha256:b6ad1a679f760dda52b1168c859d01b7b80648ea6f7f7c7f5a8a91dc3f3ecb84"},
 ]
 
 [package.extras]
@@ -1468,32 +1468,32 @@ setuptools = "*"
 
 [[package]]
 name = "numba"
-version = "0.59.1"
+version = "0.59.0"
 description = "compiling Python code using LLVM"
 optional = false
 python-versions = ">=3.9"
 files = [
-    {file = "numba-0.59.1-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:97385a7f12212c4f4bc28f648720a92514bee79d7063e40ef66c2d30600fd18e"},
-    {file = "numba-0.59.1-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:0b77aecf52040de2a1eb1d7e314497b9e56fba17466c80b457b971a25bb1576d"},
-    {file = "numba-0.59.1-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:3476a4f641bfd58f35ead42f4dcaf5f132569c4647c6f1360ccf18ee4cda3990"},
-    {file = "numba-0.59.1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:525ef3f820931bdae95ee5379c670d5c97289c6520726bc6937a4a7d4230ba24"},
-    {file = "numba-0.59.1-cp310-cp310-win_amd64.whl", hash = "sha256:990e395e44d192a12105eca3083b61307db7da10e093972ca285c85bef0963d6"},
-    {file = "numba-0.59.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:43727e7ad20b3ec23ee4fc642f5b61845c71f75dd2825b3c234390c6d8d64051"},
-    {file = "numba-0.59.1-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:411df625372c77959570050e861981e9d196cc1da9aa62c3d6a836b5cc338966"},
-    {file = "numba-0.59.1-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:2801003caa263d1e8497fb84829a7ecfb61738a95f62bc05693fcf1733e978e4"},
-    {file = "numba-0.59.1-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:dd2842fac03be4e5324ebbbd4d2d0c8c0fc6e0df75c09477dd45b288a0777389"},
-    {file = "numba-0.59.1-cp311-cp311-win_amd64.whl", hash = "sha256:0594b3dfb369fada1f8bb2e3045cd6c61a564c62e50cf1f86b4666bc721b3450"},
-    {file = "numba-0.59.1-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:1cce206a3b92836cdf26ef39d3a3242fec25e07f020cc4feec4c4a865e340569"},
-    {file = "numba-0.59.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:8c8b4477763cb1fbd86a3be7050500229417bf60867c93e131fd2626edb02238"},
-    {file = "numba-0.59.1-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:7d80bce4ef7e65bf895c29e3889ca75a29ee01da80266a01d34815918e365835"},
-    {file = "numba-0.59.1-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:f7ad1d217773e89a9845886401eaaab0a156a90aa2f179fdc125261fd1105096"},
-    {file = "numba-0.59.1-cp312-cp312-win_amd64.whl", hash = "sha256:5bf68f4d69dd3a9f26a9b23548fa23e3bcb9042e2935257b471d2a8d3c424b7f"},
-    {file = "numba-0.59.1-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:4e0318ae729de6e5dbe64c75ead1a95eb01fabfe0e2ebed81ebf0344d32db0ae"},
-    {file = "numba-0.59.1-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:0f68589740a8c38bb7dc1b938b55d1145244c8353078eea23895d4f82c8b9ec1"},
-    {file = "numba-0.59.1-cp39-cp39-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:649913a3758891c77c32e2d2a3bcbedf4a69f5fea276d11f9119677c45a422e8"},
-    {file = "numba-0.59.1-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:9712808e4545270291d76b9a264839ac878c5eb7d8b6e02c970dc0ac29bc8187"},
-    {file = "numba-0.59.1-cp39-cp39-win_amd64.whl", hash = "sha256:8d51ccd7008a83105ad6a0082b6a2b70f1142dc7cfd76deb8c5a862367eb8c86"},
-    {file = "numba-0.59.1.tar.gz", hash = "sha256:76f69132b96028d2774ed20415e8c528a34e3299a40581bae178f0994a2f370b"},
+    {file = "numba-0.59.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:8d061d800473fb8fef76a455221f4ad649a53f5e0f96e3f6c8b8553ee6fa98fa"},
+    {file = "numba-0.59.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:c086a434e7d3891ce5dfd3d1e7ee8102ac1e733962098578b507864120559ceb"},
+    {file = "numba-0.59.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:9e20736bf62e61f8353fb71b0d3a1efba636c7a303d511600fc57648b55823ed"},
+    {file = "numba-0.59.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:e86e6786aec31d2002122199486e10bbc0dc40f78d76364cded375912b13614c"},
+    {file = "numba-0.59.0-cp310-cp310-win_amd64.whl", hash = "sha256:0307ee91b24500bb7e64d8a109848baf3a3905df48ce142b8ac60aaa406a0400"},
+    {file = "numba-0.59.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:d540f69a8245fb714419c2209e9af6104e568eb97623adc8943642e61f5d6d8e"},
+    {file = "numba-0.59.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:1192d6b2906bf3ff72b1d97458724d98860ab86a91abdd4cfd9328432b661e31"},
+    {file = "numba-0.59.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:90efb436d3413809fcd15298c6d395cb7d98184350472588356ccf19db9e37c8"},
+    {file = "numba-0.59.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:cd3dac45e25d927dcb65d44fb3a973994f5add2b15add13337844afe669dd1ba"},
+    {file = "numba-0.59.0-cp311-cp311-win_amd64.whl", hash = "sha256:753dc601a159861808cc3207bad5c17724d3b69552fd22768fddbf302a817a4c"},
+    {file = "numba-0.59.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:ce62bc0e6dd5264e7ff7f34f41786889fa81a6b860662f824aa7532537a7bee0"},
+    {file = "numba-0.59.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:8cbef55b73741b5eea2dbaf1b0590b14977ca95a13a07d200b794f8f6833a01c"},
+    {file = "numba-0.59.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:70d26ba589f764be45ea8c272caa467dbe882b9676f6749fe6f42678091f5f21"},
+    {file = "numba-0.59.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:e125f7d69968118c28ec0eed9fbedd75440e64214b8d2eac033c22c04db48492"},
+    {file = "numba-0.59.0-cp312-cp312-win_amd64.whl", hash = "sha256:4981659220b61a03c1e557654027d271f56f3087448967a55c79a0e5f926de62"},
+    {file = "numba-0.59.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:fe4d7562d1eed754a7511ed7ba962067f198f86909741c5c6e18c4f1819b1f47"},
+    {file = "numba-0.59.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:6feb1504bb432280f900deaf4b1dadcee68812209500ed3f81c375cbceab24dc"},
+    {file = "numba-0.59.0-cp39-cp39-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:944faad25ee23ea9dda582bfb0189fb9f4fc232359a80ab2a028b94c14ce2b1d"},
+    {file = "numba-0.59.0-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:5516a469514bfae52a9d7989db4940653a5cbfac106f44cb9c50133b7ad6224b"},
+    {file = "numba-0.59.0-cp39-cp39-win_amd64.whl", hash = "sha256:32bd0a41525ec0b1b853da244808f4e5333867df3c43c30c33f89cf20b9c2b63"},
+    {file = "numba-0.59.0.tar.gz", hash = "sha256:12b9b064a3e4ad00e2371fc5212ef0396c80f41caec9b5ec391c8b04b6eaf2a8"},
 ]
 
 [package.dependencies]
@@ -2684,13 +2684,13 @@ test = ["asv", "gmpy2", "hypothesis", "mpmath", "pooch", "pytest", "pytest-cov",
 
 [[package]]
 name = "sentry-sdk"
-version = "1.42.0"
+version = "1.41.0"
 description = "Python client for Sentry (https://sentry.io)"
 optional = false
 python-versions = "*"
 files = [
-    {file = "sentry-sdk-1.42.0.tar.gz", hash = "sha256:4a8364b8f7edbf47f95f7163e48334c96100d9c098f0ae6606e2e18183c223e6"},
-    {file = "sentry_sdk-1.42.0-py2.py3-none-any.whl", hash = "sha256:a654ee7e497a3f5f6368b36d4f04baeab1fe92b3105f7f6965d6ef0de35a9ba4"},
+    {file = "sentry-sdk-1.41.0.tar.gz", hash = "sha256:4f2d6c43c07925d8cd10dfbd0970ea7cb784f70e79523cca9dbcd72df38e5a46"},
+    {file = "sentry_sdk-1.41.0-py2.py3-none-any.whl", hash = "sha256:be4f8f4b29a80b6a3b71f0f31487beb9e296391da20af8504498a328befed53f"},
 ]
 
 [package.dependencies]
@@ -2714,7 +2714,6 @@ grpcio = ["grpcio (>=1.21.1)"]
 httpx = ["httpx (>=0.16.0)"]
 huey = ["huey (>=2)"]
 loguru = ["loguru (>=0.5)"]
-openai = ["openai (>=1.0.0)", "tiktoken (>=0.3.0)"]
 opentelemetry = ["opentelemetry-distro (>=0.35b0)"]
 opentelemetry-experimental = ["opentelemetry-distro (>=0.40b0,<1.0)", "opentelemetry-instrumentation-aiohttp-client (>=0.40b0,<1.0)", "opentelemetry-instrumentation-django (>=0.40b0,<1.0)", "opentelemetry-instrumentation-fastapi (>=0.40b0,<1.0)", "opentelemetry-instrumentation-flask (>=0.40b0,<1.0)", "opentelemetry-instrumentation-requests (>=0.40b0,<1.0)", "opentelemetry-instrumentation-sqlite3 (>=0.40b0,<1.0)", "opentelemetry-instrumentation-urllib (>=0.40b0,<1.0)"]
 pure-eval = ["asttokens", "executing", "pure-eval"]
@@ -2830,18 +2829,18 @@ test = ["pytest"]
 
 [[package]]
 name = "setuptools"
-version = "69.2.0"
+version = "69.1.1"
 description = "Easily download, build, install, upgrade, and uninstall Python packages"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "setuptools-69.2.0-py3-none-any.whl", hash = "sha256:c21c49fb1042386df081cb5d86759792ab89efca84cf114889191cd09aacc80c"},
-    {file = "setuptools-69.2.0.tar.gz", hash = "sha256:0ff4183f8f42cd8fa3acea16c45205521a4ef28f73c6391d8a25e92893134f2e"},
+    {file = "setuptools-69.1.1-py3-none-any.whl", hash = "sha256:02fa291a0471b3a18b2b2481ed902af520c69e8ae0919c13da936542754b4c56"},
+    {file = "setuptools-69.1.1.tar.gz", hash = "sha256:5c0806c7d9af348e6dd3777b4f4dbb42c7ad85b190104837488eab9a7c945cf8"},
 ]
 
 [package.extras]
 docs = ["furo", "jaraco.packaging (>=9.3)", "jaraco.tidelift (>=1.4)", "pygments-github-lexers (==0.0.5)", "rst.linker (>=1.9)", "sphinx (<7.2.5)", "sphinx (>=3.5)", "sphinx-favicon", "sphinx-inline-tabs", "sphinx-lint", "sphinx-notfound-page (>=1,<2)", "sphinx-reredirects", "sphinxcontrib-towncrier"]
-testing = ["build[virtualenv]", "filelock (>=3.4.0)", "importlib-metadata", "ini2toml[lite] (>=0.9)", "jaraco.develop (>=7.21)", "jaraco.envs (>=2.2)", "jaraco.path (>=3.2.0)", "mypy (==1.9)", "packaging (>=23.2)", "pip (>=19.1)", "pytest (>=6)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-home (>=0.5)", "pytest-mypy (>=0.9.1)", "pytest-perf", "pytest-ruff (>=0.2.1)", "pytest-timeout", "pytest-xdist (>=3)", "tomli", "tomli-w (>=1.0.0)", "virtualenv (>=13.0.0)", "wheel"]
+testing = ["build[virtualenv]", "filelock (>=3.4.0)", "flake8-2020", "ini2toml[lite] (>=0.9)", "jaraco.develop (>=7.21)", "jaraco.envs (>=2.2)", "jaraco.path (>=3.2.0)", "packaging (>=23.2)", "pip (>=19.1)", "pytest (>=6)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-home (>=0.5)", "pytest-mypy (>=0.9.1)", "pytest-perf", "pytest-ruff (>=0.2.1)", "pytest-timeout", "pytest-xdist", "tomli-w (>=1.0.0)", "virtualenv (>=13.0.0)", "wheel"]
 testing-integration = ["build[virtualenv] (>=1.0.3)", "filelock (>=3.4.0)", "jaraco.envs (>=2.2)", "jaraco.path (>=3.2.0)", "packaging (>=23.2)", "pytest", "pytest-enabler", "pytest-xdist", "tomli", "virtualenv (>=13.0.0)", "wheel"]
 
 [[package]]
@@ -2950,7 +2949,7 @@ mpmath = ">=0.19"
 
 [[package]]
 name = "tensordict"
-version = "0.4.0+f1c833e"
+version = "0.4.0+551331d"
 description = ""
 optional = false
 python-versions = "*"
@@ -2971,7 +2970,7 @@ tests = ["pytest", "pytest-benchmark", "pytest-instafail", "pytest-rerunfailures
 type = "git"
 url = "https://github.com/pytorch/tensordict"
 reference = "HEAD"
-resolved_reference = "f1c833ecf495aa61f3f76bf09f94dd708db496ec"
+resolved_reference = "ed22554d6860731610df784b2f5d09f31d3dbc7a"
 
 [[package]]
 name = "termcolor"
@@ -3104,8 +3103,8 @@ utils = ["git", "hydra-core (>=1.1)", "hydra-submitit-launcher", "tensorboard",
 [package.source]
 type = "git"
 url = "https://github.com/pytorch/rl"
-reference = "HEAD"
-resolved_reference = "c371266ce5a71cb4b1a319cc56ad59d9b492cb9d"
+reference = "13bef426dcfa5887c6e5034a6e9697993fa92c37"
+resolved_reference = "13bef426dcfa5887c6e5034a6e9697993fa92c37"
 
 [[package]]
 name = "torchvision"
@@ -3312,20 +3311,20 @@ jupyter = ["ipytree (>=0.2.2)", "ipywidgets (>=8.0.0)", "notebook"]
 
 [[package]]
 name = "zipp"
-version = "3.18.1"
+version = "3.17.0"
 description = "Backport of pathlib-compatible object wrapper for zip files"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "zipp-3.18.1-py3-none-any.whl", hash = "sha256:206f5a15f2af3dbaee80769fb7dc6f249695e940acca08dfb2a4769fe61e538b"},
-    {file = "zipp-3.18.1.tar.gz", hash = "sha256:2884ed22e7d8961de1c9a05142eb69a247f120291bc0206a00a7642f09b5b715"},
+    {file = "zipp-3.17.0-py3-none-any.whl", hash = "sha256:0e923e726174922dce09c53c59ad483ff7bbb8e572e00c7f7c46b88556409f31"},
+    {file = "zipp-3.17.0.tar.gz", hash = "sha256:84e64a1c28cf7e91ed2078bb8cc8c259cb19b76942096c8d7b84947690cabaf0"},
 ]
 
 [package.extras]
-docs = ["furo", "jaraco.packaging (>=9.3)", "jaraco.tidelift (>=1.4)", "rst.linker (>=1.9)", "sphinx (>=3.5)", "sphinx-lint"]
-testing = ["big-O", "jaraco.functools", "jaraco.itertools", "more-itertools", "pytest (>=6)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-ignore-flaky", "pytest-mypy", "pytest-ruff (>=0.2.1)"]
+docs = ["furo", "jaraco.packaging (>=9.3)", "jaraco.tidelift (>=1.4)", "rst.linker (>=1.9)", "sphinx (<7.2.5)", "sphinx (>=3.5)", "sphinx-lint"]
+testing = ["big-O", "jaraco.functools", "jaraco.itertools", "more-itertools", "pytest (>=6)", "pytest-black (>=0.3.7)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-ignore-flaky", "pytest-mypy (>=0.9.1)", "pytest-ruff"]
 
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.10"
-content-hash = "e9f0d66c6b070050c1cc29b3bfbb0a1383bd356536091dbebf664c9a0db1f57c"
+content-hash = "ee86b84a795e6a3e9c2d79f244a87b55589adbe46d549ac38adf48be27c04cf9"

pyproject.toml

@@ -41,7 +41,7 @@ numba = "^0.59.0"
 mpmath = "^1.3.0"
 torch = "^2.2.1"
 tensordict = {git = "https://github.com/pytorch/tensordict"}
-torchrl = {git = "https://github.com/pytorch/rl"}
+torchrl = {git = "https://github.com/pytorch/rl", rev = "13bef426dcfa5887c6e5034a6e9697993fa92c37"}
 mujoco = "2.3.7"
 mujoco-py = "^2.1.2.14"
 gym = "^0.26.2"

tests/scripts/mock_dataset.py

@@ -1,18 +1,5 @@
 """
-This script is designed to facilitate the creation of a subset of an existing dataset by selecting a specific number of frames from the original dataset.
-This subset can then be used for running quick unit tests.
-The script takes an input directory containing the original dataset and an output directory where the subset of the dataset will be saved.
-Additionally, the number of frames to include in the subset can be specified.
-The script ensures that the subset is a representative sample of the original dataset by copying the specified number of frames and retaining the structure and format of the data.
-
-Usage:
-    Run the script with the following command, specifying the path to the input data directory,
-    the path to the output data directory, and optionally the number of frames to include in the subset dataset:
-
-    `python tests/scripts/mock_dataset.py --in-data-dir path/to/input_data --out-data-dir path/to/output_data`
-
-Example:
-    `python tests/scripts/mock_dataset.py --in-data-dir data/pusht --out-data-dir tests/data/pusht`
+    usage: `python tests/scripts/mock_dataset.py --in-data-dir data/pusht --out-data-dir tests/data/pusht`
 """
 
 import argparse
@@ -22,16 +9,13 @@ from tensordict import TensorDict
 from pathlib import Path
 
 
-def mock_dataset(in_data_dir, out_data_dir, num_frames):
-    in_data_dir = Path(in_data_dir)
-    out_data_dir = Path(out_data_dir)
-
+def mock_dataset(in_data_dir, out_data_dir, num_frames=50):
     # load full dataset as a tensor dict
-    in_td_data = TensorDict.load_memmap(in_data_dir / "replay_buffer")
+    in_td_data = TensorDict.load_memmap(in_data_dir)
 
     # use 1 frame to know the specification of the dataset
     # and copy it over `n` frames in the test artifact directory
-    out_td_data = in_td_data[0].expand(num_frames).memmap_like(out_data_dir / "replay_buffer")
+    out_td_data = in_td_data[0].expand(num_frames).memmap_like(out_data_dir)
 
     # copy the first `n` frames so that we have real data
     out_td_data[:num_frames] = in_td_data[:num_frames].clone()
@@ -40,19 +24,18 @@ def mock_dataset(in_data_dir, out_data_dir, num_frames):
     out_td_data.lock_()
 
     # copy the full statistics of dataset since it's pretty small
-    in_stats_path = in_data_dir / "stats.pth"
-    out_stats_path = out_data_dir / "stats.pth"
+    in_stats_path = Path(in_data_dir) / "stats.pth"
+    out_stats_path = Path(out_data_dir) / "stats.pth"
     shutil.copy(in_stats_path, out_stats_path)
 
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser(description="Create a dataset with a subset of frames for quick testing.")
+    parser = argparse.ArgumentParser(description="Create dataset")
 
     parser.add_argument("--in-data-dir", type=str, help="Path to input data")
     parser.add_argument("--out-data-dir", type=str, help="Path to save the output data")
-    parser.add_argument("--num-frames", type=int, default=50, help="Number of frames to copy over")
 
     args = parser.parse_args()
 
-    mock_dataset(args.in_data_dir, args.out_data_dir, args.num_frames)
+    mock_dataset(args.in_data_dir, args.out_data_dir)