(add) EE-kinematics retargeting for cross-morphology policy rollout:

Add examples/openarm with a self-contained script that replays a recorded
bimanual-OpenArm episode into an mp4 by driving the official OpenArm MuJoCo
model (enactic/openarm_mujoco, v1) from a LeRobot dataset's observation.state,
plus a README covering model provenance, setup, and troubleshooting.

Co-authored-by: Cursor <cursoragent@cursor.com>

.github/workflows/documentation.yml

@@ -24,24 +24,19 @@ on:
         required: false
         type: string
 
-  # Triggers on pushes to main that touch the docs or the sources the API reference is generated from.
-  # `src/**` is included because the API reference is built from docstrings via `[[autodoc]]`: without it,
-  # published API pages would go stale as soon as a docstring changed.
+  # Triggers the workflow on push events to main for the docs folder
   push:
     branches:
       - main
     paths:
       - "docs/**"
-      - "src/**"
 
-  # Same for pull requests, so a docstring change gets a preview build and a broken `[[autodoc]]` path
-  # fails the PR rather than main.
+  # Triggers the workflow on pull request events targeting main for the docs folder
   pull_request:
     branches:
       - main
     paths:
       - "docs/**"
-      - "src/**"
 
   release:
     types: [published]
@@ -64,21 +59,12 @@ jobs:
     with:
       commit_sha: ${{ github.sha }}
       package: lerobot
-      # doc-builder ships a mock-deps registry entry for lerobot, so the reusable workflow takes its
-      # "light install" path: `pip install ./lerobot --no-deps` plus a handful of real dependencies.
-      # That is not enough to import lerobot — draccus runs `register_subclass` at import time and
-      # `processor/converters.py` calls `functools.singledispatch.register(torch.Tensor)`, neither of
-      # which works against a mock. Install the package for real before the build.
-      pre_command: uv pip install "./lerobot[dataset]"
-      # `--version main` is load-bearing: without `--not_python_module`, doc-builder falls back to
-      # `lerobot.__version__` and only maps that to the default branch when it contains "dev". Our main
-      # branch carries a release version (0.6.2), so omitting this would publish the main docs to
-      # /lerobot/v0.6.2/ instead of /lerobot/main/ and disable notebook building.
       additional_args: >-
+        --not_python_module
         ${{
           (github.event_name == 'release' && format('--version {0}', github.event.release.tag_name)) ||
           (inputs.version != '' && format('--version {0}', inputs.version)) ||
-          '--version main'
+          ''
         }}
     secrets:
       token: ${{ secrets.HUGGINGFACE_PUSH }}
@@ -97,6 +83,4 @@ jobs:
       commit_sha: ${{ github.event.pull_request.head.sha }}
       pr_number: ${{ github.event.number }}
       package: lerobot
-      # See the comment on build_main_docs. The PR workflow passes its own `--version pr_<n>`, so no
-      # additional_args are needed here.
-      pre_command: uv pip install "./lerobot[dataset]"
+      additional_args: --not_python_module

.github/workflows/quality.yml

@@ -56,41 +56,3 @@ jobs:
         uses: pre-commit/action@2c7b3805fd2a0fd8c1884dcaebf91fc102a13ecd  # v3.0.1
         with:
           extra_args: --all-files --show-diff-on-failure --color=always
-
-  # This job runs the examples in our docstrings and validates the doctest allowlist.
-  # See docs/source/writing_docstrings.mdx for the standard these enforce.
-  doc-checks:
-    name: Run Documentation Checks (Doctests)
-    runs-on: ubuntu-latest
-    env:
-      # Examples that need a physical robot, a serial port or a Hub download are skipped by content.
-      # Everything else has to actually run. See src/lerobot/utils/doctest_utils.py.
-      SKIP_HARDWARE_DOCTEST: "1"
-      SKIP_CUDA_DOCTEST: "1"
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1  # v7.0.1
-        with:
-          persist-credentials: false
-
-      - name: Setup uv and Python
-        uses: astral-sh/setup-uv@11f9893b081a58869d3b5fccaea48c9e9e46f990  # v8.3.2
-        with:
-          enable-cache: true
-          version: "0.11.30"
-          python-version: "3.12"
-
-      - name: Install dependencies
-        run: uv sync --locked --extra test --extra dataset
-
-      - name: Check the doctest list is sorted and its paths exist
-        run: make check-doctest-list
-
-      - name: Check documented arguments match their signatures
-        run: make check-docstrings
-
-      - name: Check docstring coverage has not regressed
-        run: uv run --with interrogate interrogate --config=pyproject.toml
-
-      - name: Run doctests
-        run: make doctest

.pre-commit-config.yaml

@@ -67,11 +67,7 @@ repos:
         args: [--prose-wrap=preserve]
         # Jinja2 model-card templates use a .md extension but contain {% ... %} /
         # {{ ... }} tags that prettier's Markdown formatter mangles (e.g. table loops).
-        #
-        # docs/source/api/ holds the generated API reference. Its `[[autodoc]]` blocks restrict output
-        # to an indented `- member` list, which prettier reads as a lazy paragraph continuation and
-        # joins onto one line — silently turning a member list into part of the directive.
-        exclude: ^(src/lerobot/templates/.*\.md|docs/source/api/.*\.mdx)$
+        exclude: ^src/lerobot/templates/.*\.md$
 
   ##### Security #####
   - repo: https://github.com/gitleaks/gitleaks
@@ -108,13 +104,8 @@ repos:
   #       args: ["--docstring-style", "google", "-v", "2"]
   #       exclude: ^tests/.*$
 
-  # interrogate runs in CI (quality.yml, doc-checks job) rather than here. Its 1.7.0 release still imports
-  # the deprecated `py` package, which resolves against whatever `py` happens to be importable in
-  # pre-commit's isolated env — on a machine with miniconda on the path that is a stray `py.py` and the
-  # hook dies before it reads any config. The gate is the same either way; the CI step is just reliable.
   # - repo: https://github.com/econchick/interrogate
   #   rev: 1.7.0
   #   hooks:
   #     - id: interrogate
-  #       args: ["--config=pyproject.toml"]
-  #       pass_filenames: false
+  #       args: ["-vv", "--config=pyproject.toml"]

CONTRIBUTING.md

@@ -50,10 +50,6 @@ To run checks manually on all files:
 pre-commit run --all-files
 ```
 
-### Docstrings
-
-The API reference is generated from the docstrings in `src/lerobot/`. If you add or change anything public, follow the [docstring standard](https://huggingface.co/docs/lerobot/writing_docstrings) — the format is parsed by the renderer and checked in CI.
-
 ### Running Tests
 
 We use `pytest`. First, ensure you have test artifacts by installing **git-lfs**:

Makefile

@@ -184,29 +184,3 @@ test-smolvla-ete-eval:
 # backend, so it does not require a real model checkpoint or GPU.
 annotation-e2e:
 	uv run python -m tests.annotations.run_e2e_smoke
-
-# Docstring & doctest checks. See docs/source/writing_docstrings.mdx for the standard these enforce.
-
-# Run the examples in the docstrings listed in utils/documentation_tests.txt. Hardware and GPU examples are
-# skipped by content (see src/lerobot/utils/doctest_utils.py); CI sets both flags.
-doctest:
-	@files=$$(grep -v '^\s*#' utils/documentation_tests.txt | grep -v '^\s*$$'); \
-	if [ -z "$$files" ]; then \
-		echo "utils/documentation_tests.txt lists no files; nothing to run."; \
-	else \
-		SKIP_HARDWARE_DOCTEST=1 uv run pytest --doctest-modules --no-header -q $$files; \
-	fi
-
-check-doctest-list:
-	uv run python utils/check_doctest_list.py
-
-fix-doctest-list:
-	uv run python utils/check_doctest_list.py --fix_and_overwrite
-
-check-docstrings:
-	uv run python utils/check_docstrings.py
-	uv run python utils/check_config_docstrings.py
-
-fix-docstrings:
-	uv run python utils/check_docstrings.py --fix_and_overwrite
-	uv run python utils/check_doctest_list.py --fix_and_overwrite

README.md

@@ -128,23 +128,6 @@ lerobot-eval \
 
 Learn how to implement your own simulation environment or benchmark and distribute it from the HF Hub by following the [EnvHub Documentation](https://huggingface.co/docs/lerobot/envhub).
 
-### Third-Party Hardware
-
-Beyond the natively supported hardware, the community maintains a growing ecosystem of plugins for other robots, teleoperators, cameras, and sensors - UFACTORY xArm, Universal Robots UR5e, Franka, AgileX Piper, Trossen WidowX, ARX5, I2RT YAM, GELLO, SpaceMouse, Meta Quest, ROS 2 bridges, tactile and depth cameras, and more.
-
-Plugins are auto-discovered by package name: LeRobot imports any installed package prefixed with `lerobot_robot_`, `lerobot_teleoperator_`, or `lerobot_camera_`. Install one and use the `type` it registers straight from the CLI:
-
-```bash
-pip install lerobot_robot_<name> lerobot_teleoperator_<name>
-
-lerobot-record \
-  --robot.type=<robot_name> \
-  --teleop.type=<teleoperator_name> \
-  --dataset.repo_id=${HF_USER}/my-dataset
-```
-
-Browse the full list in the [Third-Party Robots & Teleoperators](https://huggingface.co/docs/lerobot/main/third_party_robots) and [Third-Party Cameras & Sensors](https://huggingface.co/docs/lerobot/main/third_party_sensors) documentation.
-
 ## Resources
 
 - **[Documentation](https://huggingface.co/docs/lerobot/index):** The complete guide to tutorials & API.

conftest.py

@@ -1,60 +0,0 @@
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-"""Root conftest: makes doctest collection use LeRobot's parser.
-
-This only affects `--doctest-modules` runs (see `make doctest`). The test suite itself is configured by
-`tests/conftest.py`.
-"""
-
-import doctest
-
-import _pytest.doctest
-
-from lerobot.utils.doctest_utils import LeRobotDoctestModule, LeRobotDocTestParser
-
-# Lets an example opt out of output comparison with `# doctest: +IGNORE_RESULT`, for calls whose output is
-# a progress bar or otherwise not reproducible.
-IGNORE_RESULT = doctest.register_optionflag("IGNORE_RESULT")
-
-OutputChecker = doctest.OutputChecker
-
-
-class CustomOutputChecker(OutputChecker):
-    """An output checker that honours the `IGNORE_RESULT` flag."""
-
-    def check_output(self, want, got, optionflags):
-        """Return `True` when `IGNORE_RESULT` is set, otherwise defer to stdlib.
-
-        Args:
-            want (`str`):
-                The expected output.
-            got (`str`):
-                The actual output.
-            optionflags (`int`):
-                Bitmask of active doctest option flags.
-
-        Returns:
-            `bool`: Whether the output is considered a match.
-        """
-        if IGNORE_RESULT & optionflags:
-            return True
-        return OutputChecker.check_output(self, want, got, optionflags)
-
-
-# Reassigning these module attributes is how doctest behaviour is customised; mypy sees it as assigning to
-# a type, which is exactly what is intended here.
-doctest.OutputChecker = CustomOutputChecker  # type: ignore[misc]
-_pytest.doctest.DoctestModule = LeRobotDoctestModule
-doctest.DocTestParser = LeRobotDocTestParser  # type: ignore[misc]

docs/source/_toctree.yml

@@ -191,28 +191,6 @@
 - sections:
   - local: contributing
     title: Contribute to LeRobot
-  - local: writing_docstrings
-    title: Writing docstrings
   - local: backwardcomp
     title: Backward compatibility
   title: "About"
-- sections:
-  - local: api/robots
-    title: Robots
-  - local: api/teleoperators
-    title: Teleoperators
-  - local: api/cameras
-    title: Cameras
-  - local: api/motors
-    title: Motors
-  - local: api/datasets
-    title: Datasets
-  - local: api/policies
-    title: Policies
-  - local: api/processor
-    title: Processors
-  - local: api/envs
-    title: Environments
-  - local: api/configs
-    title: Configuration
-  title: "API Reference"

docs/source/api/cameras.mdx

@@ -1,24 +0,0 @@
-# Cameras
-
-Cameras supply the image observations a policy sees. Every backend — OpenCV, Intel RealSense, Reachy 2 —
-implements the [`Camera`] interface, so swapping hardware does not change the code that reads frames.
-
-See the [Cameras guide](../cameras) for choosing and configuring a camera, and
-[Third-Party Cameras & Sensors](../third_party_sensors) for devices outside the core set.
-
-## Camera
-
-[[autodoc]] lerobot.cameras.Camera
-    - connect
-    - disconnect
-    - read
-    - async_read
-    - find_cameras
-
-## CameraConfig
-
-[[autodoc]] lerobot.cameras.CameraConfig
-
-## make_cameras_from_configs
-
-[[autodoc]] lerobot.cameras.make_cameras_from_configs

docs/source/api/configs.mdx

@@ -1,27 +0,0 @@
-# Configuration
-
-LeRobot configuration is plain dataclasses parsed by [draccus](https://github.com/dlwh/draccus), so every
-field is settable from the CLI. [`TrainPipelineConfig`] is the top-level object for `lerobot-train`.
-
-Polymorphic configs (policies, robots, environments) use `draccus.ChoiceRegistry`: a subclass registers
-itself with `@register_subclass("name")` and is then selectable by that name on the command line.
-
-## TrainPipelineConfig
-
-[[autodoc]] lerobot.configs.train.TrainPipelineConfig
-
-## PreTrainedConfig
-
-[[autodoc]] lerobot.configs.PreTrainedConfig
-
-## DatasetConfig
-
-[[autodoc]] lerobot.configs.DatasetConfig
-
-## EvalConfig
-
-[[autodoc]] lerobot.configs.EvalConfig
-
-## WandBConfig
-
-[[autodoc]] lerobot.configs.WandBConfig

docs/source/api/datasets.mdx

@@ -1,57 +0,0 @@
-# Datasets
-
-[`LeRobotDataset`] is the format every LeRobot script reads and writes. It is episode-aware, decodes video
-observations on the fly, and round-trips to the Hugging Face Hub.
-
-See [Using LeRobotDataset](../lerobot-dataset-v3) for the format and the common operations,
-[Porting Large Datasets](../porting_datasets_v3) for migration, and [Tools](../tools) for the CLI.
-
-## LeRobotDataset
-
-[[autodoc]] lerobot.datasets.LeRobotDataset
-
-## LeRobotDatasetMetadata
-
-[[autodoc]] lerobot.datasets.LeRobotDatasetMetadata
-
-## MultiLeRobotDataset
-
-[[autodoc]] lerobot.datasets.MultiLeRobotDataset
-
-## StreamingLeRobotDataset
-
-[[autodoc]] lerobot.datasets.StreamingLeRobotDataset
-
-## EpisodeAwareSampler
-
-[[autodoc]] lerobot.datasets.sampler.EpisodeAwareSampler
-
-## Editing a dataset
-
-Functions in `lerobot.datasets.dataset_tools` for editing an existing `LeRobotDataset` on disk: adding,
-removing, or modifying features; splitting, merging, or deleting episodes; re-encoding video; and
-recomputing statistics. Each returns a new dataset rather than mutating the source in place.
-
-[[autodoc]] lerobot.datasets.dataset_tools.add_features
-
-[[autodoc]] lerobot.datasets.dataset_tools.remove_feature
-
-[[autodoc]] lerobot.datasets.dataset_tools.modify_features
-
-[[autodoc]] lerobot.datasets.dataset_tools.modify_tasks
-
-[[autodoc]] lerobot.datasets.dataset_tools.delete_episodes
-
-[[autodoc]] lerobot.datasets.dataset_tools.split_dataset
-
-[[autodoc]] lerobot.datasets.dataset_tools.merge_datasets
-
-[[autodoc]] lerobot.datasets.dataset_tools.recompute_stats
-
-[[autodoc]] lerobot.datasets.dataset_tools.reencode_dataset
-
-[[autodoc]] lerobot.datasets.dataset_tools.convert_image_to_video_dataset
-
-## Aggregating datasets
-
-[[autodoc]] lerobot.datasets.aggregate.aggregate_datasets

docs/source/api/envs.mdx

@@ -1,19 +0,0 @@
-# Environments
-
-Simulation environments are configured through [`EnvConfig`] and built by [`make_env`]. Each subclass
-declares its `gym_kwargs` and how to construct the vectorised environments.
-
-See [Environments from the Hub](../envhub) for using published environments and
-[Adding a New Benchmark](../adding_benchmarks) for contributing one.
-
-## EnvConfig
-
-[[autodoc]] lerobot.envs.EnvConfig
-
-## make_env
-
-[[autodoc]] lerobot.envs.make_env
-
-## make_env_config
-
-[[autodoc]] lerobot.envs.make_env_config

docs/source/api/motors.mdx

@@ -1,23 +0,0 @@
-# Motors
-
-`MotorsBus` is the low-level interface to a chain of servos on a serial bus. Robots use it to read positions
-and write goal positions; you rarely touch it directly unless you are adding hardware.
-
-See [Bring Your Own Hardware](../integrate_hardware) for adding a new bus, and
-[Updating Feetech Firmware](../feetech) and [Damiao Motors and CAN Bus](../damiao) for device-specific notes.
-
-## MotorsBus
-
-[[autodoc]] lerobot.motors.motors_bus.MotorsBus
-
-## Motor
-
-[[autodoc]] lerobot.motors.Motor
-
-## MotorCalibration
-
-[[autodoc]] lerobot.motors.MotorCalibration
-
-## MotorNormMode
-
-[[autodoc]] lerobot.motors.MotorNormMode

docs/source/api/policies.mdx

@@ -1,20 +0,0 @@
-# Policies
-
-Every policy inherits [`PreTrainedPolicy`], which combines a `torch.nn.Module` with the Hub mixin, so any
-policy can be pushed to and loaded from the Hugging Face Hub with the same two calls.
-
-Each policy has its own guide with training recipes and results — [ACT](../act), [SmolVLA](../smolvla),
-[π₀](../pi0), [π₀.₅](../pi05) and the rest are listed under Policies. To add one, see
-[Adding a Policy](../bring_your_own_policies).
-
-## PreTrainedPolicy
-
-[[autodoc]] lerobot.policies.pretrained.PreTrainedPolicy
-
-## PreTrainedConfig
-
-[[autodoc]] lerobot.configs.PreTrainedConfig
-
-## make_policy
-
-[[autodoc]] lerobot.policies.factory.make_policy

docs/source/api/processor.mdx

@@ -1,20 +0,0 @@
-# Processors
-
-Processors are the data transformation layer between a robot, a dataset and a policy. A pipeline is a chain
-of [`ProcessorStep`]s; each step declares how it transforms both the data and the feature contract.
-
-See [Introduction to Robot Processors](../introduction_processors) for the concepts,
-[Implement your own processor](../implement_your_own_processor) to write a step, and
-[Debug your processor pipeline](../debug_processor_pipeline) when a pipeline misbehaves.
-
-## ProcessorStep
-
-[[autodoc]] lerobot.processor.pipeline.ProcessorStep
-
-## DataProcessorPipeline
-
-[[autodoc]] lerobot.processor.pipeline.DataProcessorPipeline
-
-## PolicyProcessorPipeline
-
-[[autodoc]] lerobot.processor.pipeline.PolicyProcessorPipeline

docs/source/api/robots.mdx

@@ -1,147 +0,0 @@
-# Robots
-
-Every robot in LeRobot implements the [`Robot`] interface: connect, read an observation, send an action,
-disconnect. Writing a policy or a recording script against that interface means it works with any supported
-arm without change.
-
-This page is the generated reference. For wiring, calibration and first-run instructions, start with the
-hardware guides — [SO-101](../so101), [LeKiwi](../lekiwi), [Hope Jr](../hope_jr), [Reachy 2](../reachy2),
-[OpenArm](../openarm) — or [Imitation Learning for Robots](../il_robots) for the end-to-end workflow. To add
-a robot of your own, see [Bring Your Own Hardware](../integrate_hardware).
-
-## Robot
-
-The abstract base class. Subclasses implement every method below; the contract described here is what a
-policy or recording loop can rely on.
-
-[[autodoc]] lerobot.robots.Robot
-    - connect
-    - disconnect
-    - configure
-    - calibrate
-    - get_observation
-    - send_action
-    - observation_features
-    - action_features
-    - is_connected
-    - is_calibrated
-
-## RobotConfig
-
-[[autodoc]] lerobot.robots.RobotConfig
-
-## make_robot_from_config
-
-[[autodoc]] lerobot.robots.make_robot_from_config
-
-## SO-100 and SO-101 followers
-
-`SO100Follower` and `SO101Follower` are aliases of the same `SOFollower` class; the two arms differ in their
-configuration, not their control code. `SO100FollowerConfig` and `SO101FollowerConfig` are likewise aliases
-of `SOFollowerRobotConfig`.
-
-[[autodoc]] lerobot.robots.so_follower.SOFollower
-    - all
-
-[[autodoc]] lerobot.robots.so_follower.SOFollowerRobotConfig
-
-## BiSOFollower
-
-Two SO followers driven as one bimanual robot.
-
-[[autodoc]] lerobot.robots.bi_so_follower.BiSOFollower
-    - all
-
-[[autodoc]] lerobot.robots.bi_so_follower.BiSOFollowerConfig
-
-## KochFollower
-
-[[autodoc]] lerobot.robots.koch_follower.KochFollower
-    - all
-
-[[autodoc]] lerobot.robots.koch_follower.KochFollowerConfig
-
-## LeKiwi
-
-`LeKiwi` runs on the robot itself. `LeKiwiClient` is the host-side proxy that talks to it over the network
-and presents the same [`Robot`] interface.
-
-[[autodoc]] lerobot.robots.lekiwi.LeKiwi
-    - all
-
-[[autodoc]] lerobot.robots.lekiwi.LeKiwiConfig
-
-[[autodoc]] lerobot.robots.lekiwi.LeKiwiClient
-    - all
-
-[[autodoc]] lerobot.robots.lekiwi.LeKiwiClientConfig
-
-## OpenArmFollower
-
-[[autodoc]] lerobot.robots.openarm_follower.OpenArmFollower
-    - all
-
-[[autodoc]] lerobot.robots.openarm_follower.OpenArmFollowerConfig
-
-## BiOpenArmFollower
-
-[[autodoc]] lerobot.robots.bi_openarm_follower.BiOpenArmFollower
-    - all
-
-[[autodoc]] lerobot.robots.bi_openarm_follower.BiOpenArmFollowerConfig
-
-## OmxFollower
-
-[[autodoc]] lerobot.robots.omx_follower.OmxFollower
-    - all
-
-[[autodoc]] lerobot.robots.omx_follower.OmxFollowerConfig
-
-## Reachy2Robot
-
-[[autodoc]] lerobot.robots.reachy2.Reachy2Robot
-    - all
-
-[[autodoc]] lerobot.robots.reachy2.Reachy2RobotConfig
-
-## UnitreeG1
-
-[[autodoc]] lerobot.robots.unitree_g1.UnitreeG1
-    - all
-
-[[autodoc]] lerobot.robots.unitree_g1.UnitreeG1Config
-
-## Hope Jr
-
-The Hope Jr humanoid is exposed as two independent robots, an arm and a hand.
-
-[[autodoc]] lerobot.robots.hope_jr.HopeJrArm
-    - all
-
-[[autodoc]] lerobot.robots.hope_jr.HopeJrArmConfig
-
-[[autodoc]] lerobot.robots.hope_jr.HopeJrHand
-    - all
-
-[[autodoc]] lerobot.robots.hope_jr.HopeJrHandConfig
-
-## RebotB601Follower
-
-[[autodoc]] lerobot.robots.rebot_b601_follower.RebotB601Follower
-    - all
-
-[[autodoc]] lerobot.robots.rebot_b601_follower.RebotB601FollowerRobotConfig
-
-## BiRebotB601Follower
-
-[[autodoc]] lerobot.robots.bi_rebot_b601_follower.BiRebotB601Follower
-    - all
-
-[[autodoc]] lerobot.robots.bi_rebot_b601_follower.BiRebotB601FollowerConfig
-
-## EarthRoverMiniPlus
-
-[[autodoc]] lerobot.robots.earthrover_mini_plus.EarthRoverMiniPlus
-    - all
-
-[[autodoc]] lerobot.robots.earthrover_mini_plus.EarthRoverMiniPlusConfig

docs/source/api/teleoperators.mdx

@@ -1,30 +0,0 @@
-# Teleoperators
-
-A teleoperator produces actions for a robot to follow — a leader arm, a gamepad, a keyboard, a phone. All of
-them implement the [`Teleoperator`] interface, so a recording script written against it works with any input
-device.
-
-See [Phone teleoperation](../phone_teleop) and [Isaac Teleop](../isaac_teleop) for setup guides, and
-[Imitation Learning for Robots](../il_robots) for the recording workflow.
-
-## Teleoperator
-
-[[autodoc]] lerobot.teleoperators.Teleoperator
-    - connect
-    - disconnect
-    - configure
-    - calibrate
-    - get_action
-    - send_feedback
-    - action_features
-    - feedback_features
-    - is_connected
-    - is_calibrated
-
-## TeleoperatorConfig
-
-[[autodoc]] lerobot.teleoperators.TeleoperatorConfig
-
-## make_teleoperator_from_config
-
-[[autodoc]] lerobot.teleoperators.make_teleoperator_from_config

docs/source/bring_your_own_policies.mdx

@@ -161,16 +161,6 @@ The methods called by the train/eval loops:
 
 Batches are flat dictionaries keyed by the constants in [`lerobot.utils.constants`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/utils/constants.py): `OBS_STATE` (`observation.state.<motor>`), `OBS_IMAGES` (`observation.images.<camera>`), `OBS_LANGUAGE`, `ACTION`, etc. Reuse the constants — don't invent new prefixes.
 
-If your model is large enough to warrant [sharded multi-GPU training](./multi_gpu_training#sharded-training-fsdp), also declare its FSDP wrap units — the repeated block classes sharding operates on:
-
-```python
-class MyPolicy(PreTrainedPolicy):
-    ...
-    _fsdp_wrap_modules = ["MyTransformerBlock"]
-```
-
-With this one declaration, `--parallelism.dp_shard=N` works out of the box for your policy (users can still override it with `--accelerator.fsdp.wrap_modules`). Without any wrap source, sharded runs fail at startup by design.
-
 ### Processor functions
 
 LeRobot uses `PolicyProcessorPipeline`s to normalize inputs and de-normalize outputs around your policy. For a concrete reference, see [`processor_act.py`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/act/processor_act.py) or [`processor_diffusion.py`](https://github.com/huggingface/lerobot/blob/main/src/lerobot/policies/diffusion/processor_diffusion.py).
@@ -310,7 +300,7 @@ The file names are load-bearing: the factory does lazy imports by name, and the
 Two places need to know about your policy. All by name.
 
 1. **`policies/__init__.py`** — re-export `MyPolicyConfig` and add it to `__all__`. This import is what registers your policy: `@PreTrainedConfig.register_subclass("my_policy")` runs, and from then on the factory resolves everything by convention. **Don't** re-export the modeling class; it loads lazily through the factory (so `import lerobot` stays fast).
-2. **`templates/lerobot_modelcard_template.md` and the root `README.md`** — the template is what the end-of-training publisher renders into the model card of every checkpoint trained with your policy: add a one-line description of your policy in the `model_name` branches, map it in `policy_docs` so cards link to your MDX guide, and optionally add an architecture image to `diagrams`. Then add your policy to the models table in the root `README.md`, under the right category, linking to your doc page.
+2. **`templates/lerobot_modelcard_template.md` and the root `README.md`** — the template is what `push_model_to_hub` renders into the model card of every checkpoint trained with your policy: add a one-line description of your policy in the `model_name` branches, map it in `policy_docs` so cards link to your MDX guide, and optionally add an architecture image to `diagrams`. Then add your policy to the models table in the root `README.md`, under the right category, linking to your doc page.
 
 Mirror an existing policy that's structurally similar to yours; the diff is small.
 
@@ -354,7 +344,7 @@ A new policy is much easier to review — and far more useful — when it ships
 
 **Pick at least one in-tree benchmark.** LeRobot ships sim benchmarks with per-benchmark Docker images (LIBERO, LIBERO-plus, Meta-World, RoboTwin 2.0, RoboCasa365, RoboCerebra, RoboMME, VLABench and more). Pick the one that matches your policy's modality — VLAs usually go to LIBERO or VLABench; image-only BC to LIBERO or Meta-World. The full list lives under [Benchmarks](./libero) in the docs sidebar.
 
-**Push the checkpoint & processors** to the Hub under `lerobot/<policy>_<benchmark>` (or your namespace if you don't have write access; a maintainer can mirror it). The easiest way is training with `--policy.repo_id=<namespace>/<repo>` and `--policy.push_to_hub=true`: `lerobot-train` publishes the model, both processors, and a model card at the end of the run. To publish an existing checkpoint after the fact, upload its `pretrained_model/` directory (e.g. `huggingface-cli upload`), or use `lerobot-convert-dcp --push_to_hub=...` for sharded-format checkpoints.
+**Push the checkpoint & processors** to the Hub under `lerobot/<policy>_<benchmark>` (or your namespace if you don't have write access; a maintainer can mirror it). Use `PreTrainedPolicy.push_model_to_hub` so the repo gets `config.json`, `model.safetensors`, and a model card.
 
 **Report results in your policy's MDX**, with the exact `lerobot-eval` command and hardware so anyone can re-run:
 

docs/source/language_and_recipes.mdx

@@ -108,7 +108,6 @@ own binding plus a matching image block, e.g.
 
 ```yaml
 ask_vqa_top:
-  route: vqa
   bindings:
     vqa_query: "emitted_at(t, style=vqa, role=user, camera=observation.images.top)"
     vqa: "emitted_at(t, style=vqa, role=assistant, camera=observation.images.top)"
@@ -128,9 +127,7 @@ ask_vqa_top:
       }
 ```
 
-Add one such sub-recipe per camera the dataset records. The explicit
-`route: vqa` marker makes a matching sparse VQA annotation take precedence
-over normal weighted blend selection; component names are purely descriptive.
+Add one such sub-recipe per camera the dataset records.
 
 ## Layer 3 — training format
 
@@ -144,20 +141,7 @@ sample["target_message_indices"]
 
 The renderer does not apply a tokenizer chat template. Policy processors decide how to serialize the messages for their backbone, which keeps the same dataset usable across SmolVLA, Pi0.5, and any future VLM that expects OpenAI-style chat messages.
 
-## Blends
-
-Blend recipes select one weighted sub-recipe deterministically from the sample index.
-`recipes/subtask_mem.yaml` trains the compact core blend — high-level subtask prediction, low-level execution, and memory. `recipes/subtask_mem_vqa_speech.yaml` is the fuller variant that also adds VQA and spoken interjection responses.
-
-`recipes/subtask_joint.yaml` demonstrates joint sequence training rather than a
-weighted blend. For the same sample, its assistant subtask is supervised with
-text cross-entropy on the `low_level` stream while action prediction remains
-active, matching the joint setup from the π0.5 paper. Enable
-`--policy.joint_subtask_conditioning=true` to use that subtask conditioning at inference.
-
 ## Graceful absence
 
-If both language columns are missing, `None`, or empty, `RenderMessagesStep` uses
-the task string as low-level supervision when available and otherwise leaves the
-sample unchanged. For an annotated sample, if no recipe branch applies and no
-task fallback exists, rendering returns `None`, allowing a loader to retry another sample.
+If both language columns are missing, `None`, or empty, `RenderMessagesStep` is a no-op.
+If an event-scoped branch is selected on a frame without the required event row, rendering returns `None`, allowing a loader to retry another sample.

docs/source/lerobot-dataset-v3.mdx

@@ -142,22 +142,6 @@ repo_id = "yaak-ai/L2D-v3"
 dataset = StreamingLeRobotDataset(repo_id)  # streams directly from the Hub
 ```
 
-Datasets stored in an [HF Storage Bucket](https://huggingface.co/docs/hub/storage-buckets) (`hf://buckets/`) can be streamed the same way by passing `repo_type="bucket"`:
-
-```python
-dataset = StreamingLeRobotDataset("my-org/my-bucket", repo_type="bucket")
-```
-
-Both options are available in `lerobot-train` through `--dataset.streaming=true`, and `--dataset.repo_type=bucket` to stream from a bucket instead of a Hub dataset repo:
-
-```bash
-lerobot-train \
-  --dataset.repo_id=my-org/my-bucket \
-  --dataset.repo_type=bucket \
-  --dataset.streaming=true \
-  ...
-```
-
 <div style="display:flex; justify-content:center; gap:12px; flex-wrap:wrap;">
   <figure style="margin:0; text-align:center;">
     <img

docs/source/multi_gpu_training.mdx

@@ -1,29 +1,28 @@
 # Multi-GPU Training
 
-LeRobot trains on multiple GPUs through [Hugging Face Accelerate](https://huggingface.co/docs/accelerate). Three data-parallel layouts are supported:
-
-| Layout   | What it does                                                  | Config                                                  |
-| -------- | ------------------------------------------------------------- | ------------------------------------------------------- |
-| **DDP**  | Replicates the full model on every GPU                        | default on any multi-GPU launch                         |
-| **FSDP** | Shards parameters, gradients, and optimizer state across GPUs | `--parallelism.dp_shard=N`                              |
-| **HSDP** | Shards within groups of GPUs, replicates across groups        | `--parallelism.dp_replicate=R --parallelism.dp_shard=S` |
+This guide shows you how to train policies on multiple GPUs using [Hugging Face Accelerate](https://huggingface.co/docs/accelerate).
 
 ## Installation
 
-`accelerate` is included in the `training` extra:
+`accelerate` is included in the `training` extra. Install it with:
 
 ```bash
 pip install 'lerobot[training]'
 ```
 
-## Launching
+## Training with Multiple GPUs
 
-Distributed training can be launched through both `torchrun` and `accelerate launch`. Accelerate is used as a plain launcher: it does not manage the training configuration, and every distributed training setting lives in LeRobot's own config system.
+You can launch training in two ways:
 
-With `torchrun`:
+### Option 1: Without config (specify parameters directly)
+
+You can specify all parameters directly in the command without running `accelerate config`:
 
 ```bash
-torchrun --nproc-per-node=2 $(which lerobot-train) \
+accelerate launch \
+  --multi_gpu \
+  --num_processes=2 \
+  $(which lerobot-train) \
   --dataset.repo_id=${HF_USER}/my_dataset \
   --policy.type=act \
   --policy.repo_id=${HF_USER}/my_trained_policy \
@@ -32,145 +31,150 @@ torchrun --nproc-per-node=2 $(which lerobot-train) \
   --wandb.enable=true
 ```
 
-With `accelerate launch` (as a plain launcher):
+**Key accelerate parameters:**
+
+- `--multi_gpu`: Enable multi-GPU training
+- `--num_processes=2`: Number of GPUs to use
+- `--mixed_precision=fp16`: Use fp16 mixed precision (or `bf16` if supported)
+
+### Option 2: Using accelerate config
+
+If you prefer to save your configuration, you can optionally configure accelerate for your hardware setup by running:
 
 ```bash
+accelerate config
+```
+
+This interactive setup will ask you questions about your training environment (number of GPUs, mixed precision settings, etc.) and saves the configuration for future use. For a simple multi-GPU setup on a single machine, you can use these recommended settings:
+
+- Compute environment: This machine
+- Number of machines: 1
+- Number of processes: (number of GPUs you want to use)
+- GPU ids to use: (leave empty to use all)
+- Mixed precision: fp16 or bf16 (recommended for faster training)
+
+Then launch training with:
+
+```bash
+accelerate launch $(which lerobot-train) \
+  --dataset.repo_id=${HF_USER}/my_dataset \
+  --policy.type=act \
+  --policy.repo_id=${HF_USER}/my_trained_policy \
+  --output_dir=outputs/train/act_multi_gpu \
+  --job_name=act_multi_gpu \
+  --wandb.enable=true
+```
+
+## How It Works
+
+When you launch training with accelerate:
+
+1. **Automatic detection**: LeRobot automatically detects if it's running under accelerate
+2. **Data distribution**: Your batch is automatically split across GPUs
+3. **Gradient synchronization**: Gradients are synchronized across GPUs during backpropagation
+4. **Single process logging**: Only the main process logs to wandb and saves checkpoints
+
+## Learning Rate and Training Steps Scaling
+
+**Important:** LeRobot does **NOT** automatically scale learning rates or training steps based on the number of GPUs. This gives you full control over your training hyperparameters.
+
+### Why No Automatic Scaling?
+
+Many distributed training frameworks automatically scale the learning rate by the number of GPUs (e.g., `lr = base_lr × num_gpus`).
+However, LeRobot keeps the learning rate exactly as you specify it.
+
+### When and How to Scale
+
+If you want to scale your hyperparameters when using multiple GPUs, you should do it manually:
+
+**Learning Rate Scaling:**
+
+```bash
+# Example: 2 GPUs with linear LR scaling
+# Base LR: 1e-4, with 2 GPUs -> 2e-4
 accelerate launch --num_processes=2 $(which lerobot-train) \
-  --dataset.repo_id=${HF_USER}/my_dataset \
-  --policy.type=act \
-  --policy.repo_id=${HF_USER}/my_trained_policy \
-  --output_dir=outputs/train/act_multi_gpu \
-  --job_name=act_multi_gpu \
-  --wandb.enable=true
+  --optimizer.lr=2e-4 \
+  --dataset.repo_id=lerobot/pusht \
+  --policy.type=act
 ```
 
-With no `--parallelism.*` flags, a multi-process launch runs plain DDP. Multi-node runs use the standard `torchrun --nnodes/--node-rank/--rdzv-endpoint` flags (or `accelerate launch --num_machines/--machine_rank/--main_process_ip`).
+**Training Steps Scaling:**
 
-> [!WARNING]
-> Accelerate's YAML config files (`accelerate launch --config_file some.yaml`, `accelerate config`) are not supported. They configure the engine through environment variables, bypassing LeRobot's configuration system, so `train_config.json` would no longer describe the settings a run actually used. `lerobot-train` therefore refuses to start when [accelerate environment variables](https://huggingface.co/docs/accelerate/usage_guides/fsdp) are set. Put the settings in `--parallelism.*` / `--accelerator.*` flags instead, or set `LEROBOT_ALLOW_ACCELERATE_ENV=1` to acknowledge the override and proceed anyway.
-
-## Batch semantics, learning rate, and steps
-
-Each of the `dp_replicate × dp_shard` data-parallel workers loads its own `--batch_size` micro-batch every step, so one training step consumes `batch_size × dp_world_size` samples, and `× gradient_accumulation_steps` of those go into each optimizer update:
-
-```
-effective_batch_size = batch_size × dp_world_size × gradient_accumulation_steps
-```
-
-The training banner prints this factorization at startup. `--steps` counts loop steps (micro-batches per worker), not optimizer updates.
-
-Gradient accumulation is a first-class flag:
+Since the effective batch size `bs` increases with multiple GPUs (batch_size × num_gpus), you may want to reduce the number of training steps proportionally:
 
 ```bash
-torchrun --nproc-per-node=2 $(which lerobot-train) \
-  --batch_size=8 --accelerator.gradient_accumulation.steps=4 ...
+# Example: 2 GPUs with effective batch size 2x larger
+# Original: batch_size=8, steps=100000
+# With 2 GPUs: batch_size=8 (16 in total), steps=50000
+accelerate launch --num_processes=2 $(which lerobot-train) \
+  --batch_size=8 \
+  --steps=50000 \
+  --dataset.repo_id=lerobot/pusht \
+  --policy.type=act
 ```
 
-**LeRobot does not auto-scale the learning rate or the number of steps** when the effective batch size grows. If you scale out and want equivalent training, please adjust manually, e.g. with 2 GPUs: double `--optimizer.lr` (linear scaling), or halve `--steps`.
+## Training Large Models with FSDP
 
-## Sharded training (FSDP)
+DDP replicates the full model on every GPU, so a model that doesn't fit on one GPU won't fit under
+DDP either. For large models, use **FSDP** (Fully Sharded Data Parallel), which shards parameters,
+gradients, and optimizer state across GPUs. See the [accelerate FSDP guide](https://huggingface.co/docs/accelerate/usage_guides/fsdp) for background.
 
-If a model is too large to train with DDP, shard it with FSDP2:
+An example on how to launch LeRobot training with FSDP across 4 GPUs (1 machine):
 
 ```bash
-torchrun --nproc-per-node=4 $(which lerobot-train) \
+accelerate launch --config_file fsdp.yaml --num_processes=4 $(which lerobot-train) \
   --dataset.repo_id=${HF_USER}/my_dataset \
   --policy.type=<your_policy> \
-  --parallelism.dp_shard=4 \
-  --accelerator.mixed_precision=bf16 \
   --output_dir=outputs/train/my_policy_fsdp
 ```
 
-`--parallelism.dp_shard=-1` shards over however many processes the launcher started.
+A minimal `fsdp.yaml` (FSDP1; shards params/grads/optimizer — ZeRO-3-equivalent):
 
-### Wrap units
-
-FSDP shards the model in units (typically the repeated transformer block) and gathers one unit at a time during forward/backward. Policies declare their wrap units via `_fsdp_wrap_modules` on the policy class. For example, ACT declares `["ACTEncoderLayer", "ACTDecoderLayer"]` and FastWAM declares `["MoTLayer"]`. For a policy without a `_fsdp_wrap_modules` declaration, pass one of the flags below. You can specify the module class name explicitly, or use a size-based policy instead:
-
-```bash
---accelerator.fsdp.wrap_modules='["MyTransformerBlock"]'   # explicit class names
---accelerator.fsdp.min_num_params=1000000                  # or: wrap every submodule above 1M params
+```yaml
+compute_environment: LOCAL_MACHINE
+distributed_type: FSDP
+mixed_precision: bf16
+num_machines: 1
+num_processes: 4
+fsdp_config:
+  fsdp_version: 1
+  fsdp_sharding_strategy: FULL_SHARD # params + grads + optimizer (ZeRO-3)
+  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  fsdp_transformer_layer_cls_to_wrap: <YourTransformerBlock> # repeated block class to shard
+  fsdp_use_orig_params: true # required: optimizer is built pre-prepare
+  fsdp_state_dict_type: FULL_STATE_DICT
 ```
 
-If a policy doesn't declare `_fsdp_wrap_modules` and no `--accelerator.fsdp.wrap_modules` or `--accelerator.fsdp.min_num_params` is passed, the run fails at startup rather than silently wrapping only the root module (which would forfeit all sharding memory savings).
+Set `fsdp_transformer_layer_cls_to_wrap` to your model's repeated transformer-block class so each
+block is sharded as its own unit. `fsdp_use_orig_params: true` is required because LeRobot builds the
+optimizer before `accelerator.prepare()`.
 
-Other sharding settings:
+### FSDP checkpoints
 
-- `--accelerator.fsdp.reshard_after_forward`: whether to keep each unit's parameters resident after forward.
-- `--accelerator.fsdp.cpu_offload`: keeps parameters, gradients and optimizer states on CPU.
-- `--accelerator.fsdp.ignored_modules`: a regex of module paths to keep unsharded.
+LeRobot gathers the full state dict across all ranks and the main process writes it as a single
+`model.safetensors`, loadable as usual with `Policy.from_pretrained(...)`. Two things to look out for:
 
-### HSDP
-
-Hybrid Sharded Data Parallel: parameters, gradients and optimizer states are sharded across `dp_shard` ranks, and that sharding is replicated `dp_replicate` times. Parameter all-gathers and gradient reduce-scatters stay inside a shard group; only the all-reduce that synchronizes the replicas crosses between groups. The two degrees must multiply to the world size:
-
-```bash
-# 16 GPUs = 2 nodes × 8: shard within each node, replicate across nodes
-torchrun --nnodes=2 --nproc-per-node=8 ... $(which lerobot-train) \
-  --parallelism.dp_replicate=2 --parallelism.dp_shard=8 ...
-```
-
-## Checkpoints
-
-Every checkpoint contains a `pretrained_model/` directory and a `training_state/` directory:
-
-```text
-005000/  # the training step at that checkpoint
-├── pretrained_model/
-│   ├── config.json  # policy config
-│   ├── train_config.json  # the full training config
-│   ├── model.safetensors  # full weights (checkpoint_format ∈ {safetensors, safetensors_dcp}, or any non-sharded run)
-│   ├── pytorch_model_fsdp_0/  # DCP weight shards (checkpoint_format ∈ {dcp, safetensors_dcp})
-│   ├── policy_preprocessor.json  # preprocessor config (when the run has a preprocessor)
-│   ├── policy_preprocessor_step_*.safetensors  # state of the stateful preprocessor steps
-│   ├── policy_postprocessor.json  # postprocessor config (when the run has a postprocessor)
-│   └── policy_postprocessor_step_*.safetensors  # state of the stateful postprocessor steps
-└── training_state/
-    ├── training_step.json  # step counter, topology, and batch semantics
-    ├── rng_state.safetensors  # rng states
-    ├── scheduler_state.json  # scheduler state (when the run has a scheduler)
-    ├── optimizer_state.safetensors  # full optimizer state (non-sharded runs)
-    ├── optimizer_param_groups.json  # optimizer param groups (non-sharded runs)
-    └── optimizer_0/  # DCP optimizer shards (sharded runs)
-```
-
-During single-GPU or DDP training, the pipeline serializes each state dict into a single file: `model.safetensors` for the model and `optimizer_state.safetensors` for the optimizer.
-
-During sharded training, the optimizer state is saved as DCP shards under `training_state/optimizer_0/`, and the layout of the model under `pretrained_model/` can be configured through `--checkpoint_format`:
-
-| `--checkpoint_format`     | Weights artifact                             | Use when                                                              |
-| ------------------------- | -------------------------------------------- | --------------------------------------------------------------------- |
-| `safetensors` _(default)_ | single `model.safetensors` only              | you want every checkpoint immediately loadable with `from_pretrained` |
-| `dcp`                     | `pytorch_model_fsdp_0/` shard directory only | gathering the full weights makes saves and resumes too slow           |
-| `safetensors_dcp`         | both                                         | you want fast resume _and_ immediately loadable checkpoints           |
-
-Two things to know about gathered (`safetensors`) checkpoints from sharded runs:
-
-- **They store fp32 weights.** Under mixed precision training, FSDP keeps an fp32 master copy, and the checkpoint saves the master copy to make sure training resumes consistently.
-- The gather is collective (all ranks participate) but only the main process writes.
-
-### Converting DCP checkpoints
-
-`lerobot-convert-dcp` merges a DCP shard directory into a regular `model.safetensors`, offline and without GPUs:
-
-```bash
-lerobot-convert-dcp --checkpoint_dir=outputs/train/run/checkpoints/005000
-lerobot-convert-dcp --checkpoint_dir=... --delete_dcp=true --push_to_hub=${HF_USER}/my_policy
-```
-
-`--push_to_hub` publishes the converted directory as a model repo.
-
-### Resuming
-
-Resume with `--resume=true --config_path=.../checkpoints/last/pretrained_model/train_config.json`. Resuming from a DCP checkpoint supports resharding the model and optimizer state to the _current_ topology, which means you can resume with a different `dp_replicate/dp_shard` split. The data sampler can always resume at the right epoch and offset, but is only _sample-exact_ when the world size and batch size match the original run (a warning is logged otherwise).
-
-> [!NOTE]
-> FSDP checkpoints written by LeRobot 0.6.x and earlier used a different on-disk layout (a gathered full optimizer state) and **cannot be resumed**.
+- **Checkpoints store fp32 weights.** Under mixed precision (`bf16`/`fp16`) FSDP keeps an fp32 master
+  copy, and the checkpoint saves it (~2× the bf16 size on disk) so training can resume consistently
+  with the fp32 optimizer state; `from_pretrained` casts back to the policy dtype on load. FSDP-specific
+  caveat: an fp32 checkpoint is materialized in full precision on the target device _before_ casting,
+  so loading it for inference on a tight GPU can OOM even when the bf16 model would fit — load on CPU
+  first, or cast `model.safetensors` to the deployment dtype offline.
+- The sharded optimizer state is gathered into a full (world-size-independent) state dict and saved
+  alongside the model in the same `optimizer_state.safetensors` / `optimizer_param_groups.json`
+  format as single-GPU training, so **resume-from-checkpoint is supported** with `--resume=true`.
+  Resume reshards both the model and the optimizer state to the _current_ FSDP topology, so you can
+  resume an FSDP checkpoint on a different number of GPUs. Note that the data sampler is only
+  sample-exact when the world size and batch size match the original run (a warning is logged
+  otherwise); the optimizer/model state itself is unaffected.
 
 ## Notes
 
-- Checkpoint saves and end-of-training publishes are collective (every rank enters them). Gathered weights, sidecar files and Hub uploads are written by the main process alone.
-- Metrics are reduced across ranks before logging: losses are averaged, and `samples/s` reports cluster-wide throughput.
-- Learning-rate scheduling is stepped once per training step regardless of the number of processes (`step_scheduler_with_optimizer=False` is baked in).
+- The `--policy.use_amp` flag in `lerobot-train` is only used when **not** running with accelerate. When using accelerate, mixed precision is controlled by accelerate's configuration.
+- Training logs, checkpoints, and hub uploads are only done by the main process to avoid conflicts. Non-main processes have console logging disabled to prevent duplicate output.
+- The effective batch size is `batch_size × num_gpus`. If you use 4 GPUs with `--batch_size=8`, your effective batch size is 32.
+- Learning rate scheduling is handled correctly across multiple processes—LeRobot sets `step_scheduler_with_optimizer=False` to prevent accelerate from adjusting scheduler steps based on the number of processes.
+- When saving or pushing models, LeRobot automatically unwraps the model from accelerate's distributed wrapper to ensure compatibility.
+- WandB integration automatically initializes only on the main process, preventing multiple runs from being created.
 
-For background on the underlying machinery, see the [Accelerate FSDP guide](https://huggingface.co/docs/accelerate/usage_guides/fsdp). To go deeper on large-scale training, check out the [Ultrascale Playbook](https://huggingface.co/spaces/nanotron/ultrascale-playbook).
+For more advanced configurations and troubleshooting, see the [Accelerate documentation](https://huggingface.co/docs/accelerate). If you want to learn more about how to train on a large number of GPUs, checkout this awesome guide: [Ultrascale Playbook](https://huggingface.co/spaces/nanotron/ultrascale-playbook).

docs/source/torch_accelerators.mdx

@@ -40,15 +40,3 @@ lerobot-eval \
 ```
 
 However, in most cases, presence of an accelerator is detected automatically and `policy.device` parameter can be omitted from CLI commands.
-
-## Mixed precision
-
-Training precision is owned by `--accelerator.mixed_precision`, which accepts `no` (default) and `bf16`:
-
-```bash
-lerobot-train \
-    --policy.type=act \
-    --accelerator.mixed_precision=bf16 ...
-```
-
-`bf16` requires an accelerator that supports it.

docs/source/writing_docstrings.mdx

@@ -1,287 +0,0 @@
-# Writing docstrings
-
-LeRobot's API reference is generated directly from the docstrings in `src/lerobot/`. A docstring is not a
-comment — it is the published documentation for that object, and the format below is what the renderer and
-the CI checks parse.
-
-This page is the contract. If you are adding or editing anything public in `src/lerobot/`, follow it.
-
-> [!IMPORTANT]
-> **An undocumented public method is an invisible one.** `[[autodoc]]` silently skips members that have no
-> docstring — no warning, no error, it simply does not appear on the rendered page. Coverage and
-> API-reference completeness are the same problem.
-
-## The format in one example
-
-Google section headers, Hugging Face type formatting. Both, not one or the other.
-
-````python
-def send_action(self, action: RobotAction, rate_hz: float = 30.0) -> RobotAction:
-    """Command the robot to move to a target joint configuration.
-
-    Values are clipped by the configured maximum relative target before reaching the motors, so the
-    returned action may differ from the requested one.
-
-    Args:
-        action (`dict[str, float]`):
-            Target values keyed by motor name, e.g. `{"shoulder_pan.pos": 0.0}`. Keys must match the
-            robot's action features.
-        rate_hz (`float`, *optional*, defaults to `30.0`):
-            Control loop frequency.
-
-    Returns:
-        `dict[str, float]`: The action actually written to the motors after safety clipping.
-
-    Raises:
-        DeviceNotConnectedError: If the robot has not been connected.
-
-    Example:
-        ```python
-        >>> from lerobot.robots.so_follower import SO101Follower, SO101FollowerConfig
-        >>> robot = SO101Follower(SO101FollowerConfig(port="/dev/ttyACM0"))  # doctest: +SKIP
-        >>> robot.connect()  # doctest: +SKIP
-        >>> robot.send_action({"shoulder_pan.pos": 0.0})  # doctest: +SKIP
-        ```
-    """
-````
-
-Cross-references are omitted from the examples on this page — see [Cross-references](#cross-references) for
-their syntax and why they cannot be shown inside a code block.
-
-## Rules
-
-### Sections
-
-`Args:` · `Returns:` · `Raises:` · `Yields:` · `Example:` · `Note:`
-
-In that order. No other section headers. A one-line summary comes first, then an optional free-form
-description, then the sections.
-
-### The `Args:` line is machine-parsed
-
-```
-name (`type`, *optional*, defaults to `X`):
-    Description, indented on its own line.
-```
-
-The `*optional*, defaults to` clause is **checked against the real signature default** by
-`make check-docstrings`. It is not decorative — if you write a default that has drifted from the code, CI
-fails. Omit the clause entirely for required parameters:
-
-```python
-    Args:
-        port (`str`):
-            Serial port the arm is connected to, e.g. `/dev/ttyACM0`.
-        max_relative_target (`float | dict[str, float]`, *optional*):
-            Caps the magnitude of the relative positional target vector. `None` disables clipping.
-        use_degrees (`bool`, *optional*, defaults to `True`):
-            Keep `True` for backward compatibility with existing policies and datasets.
-```
-
-Types go in backticks. Use `*optional*` with no `defaults to` when the default is `None` or is otherwise not
-worth restating.
-
-### `Returns:` is type-first
-
-One indented line, type first, then a colon, then the description:
-
-```python
-    Returns:
-        `dict[str, float]`: The action actually written to the motors after safety clipping.
-```
-
-`Yields:` takes the same shape.
-
-### `**Attributes**:`, never `Attributes:`
-
-doc-builder parses a bare `Attributes:` as a **synonym for `Parameters:`**, so your attributes get rendered
-as constructor arguments. This is silent and wrong. Whenever the attributes differ from the constructor
-parameters, use the bold form with a `--` separator:
-
-```python
-class Robot(abc.ABC):
-    """The base abstract class for all LeRobot-compatible robots.
-
-    **Attributes**:
-        - **config_class** (`type[RobotConfig]`) -- The expected configuration class for this robot.
-        - **name** (`str`) -- The unique robot name used to identify this robot type.
-    """
-```
-
-Note `--`, not `:`.
-
-### Cross-references
-
-Use doc-builder's bracket syntax: a square-bracketed backtick-quoted path. **Sphinx roles (`:pymeth:`,
-`:pyattr:`) are not supported** and render as literal text on the page.
-
-| Want                         | Write                               |
-| ---------------------------- | ----------------------------------- |
-| Class in the main package    | [`Robot`]                   |
-| Method, show the full path   | [`Robot.connect`]           |
-| Method, show the bare name   | [`~Robot.connect`]          |
-| Nested path                  | [`~robots.Robot.connect`]   |
-| Object in another HF library | [`~accelerate.Accelerator`] |
-
-The `~` strips the path from the **link text only**; the link still resolves to the full path.
-
-> [!NOTE]
-> doc-builder resolves this syntax everywhere in a page — including inside fenced code blocks. That is why
-> the docstring examples on this page use plain prose instead of cross-references: a code block containing
-> one would render the resolved link rather than the syntax you need to type. In your own docstrings, use
-> cross-references freely; this restriction only affects documentation _about_ the syntax.
-
-### Callouts
-
-Use GitHub-style blockquotes:
-
-```markdown
-> [!TIP]
-> Call this once at startup — it takes about two seconds.
-
-> [!WARNING]
-> Torque is disabled on disconnect. The arm will drop if it is holding a load.
-```
-
-The `<Tip>` component is legacy per doc-builder; don't add new ones.
-
-### Examples must be fenced
-
-An example lives inside a fenced ` ```python ` block containing `>>> `. The fence is what makes it render
-as a code block, and it is what the doctest preprocessor's regex looks for:
-
-````python
-    Example:
-        ```python
-        >>> from lerobot.robots.so_follower import SO101FollowerConfig
-        >>> cfg = SO101FollowerConfig(port="/dev/ttyACM0")
-        >>> cfg.use_degrees
-        True
-        ```
-````
-
-> [!WARNING]
-> An unfenced `>>>` is still collected — doctest finds prompts anywhere in a docstring. What you lose is the
-> rendering, so it shows up as a wall of prose on the page. Every example needs the fence.
-
-Every example either executes in CI or carries `# doctest: +SKIP`. Anything that touches hardware, a GPU, or
-downloads from the Hub gets `+SKIP`:
-
-````python
-    Example:
-        ```python
-        >>> robot.connect()  # doctest: +SKIP
-        >>> policy = ACTPolicy.from_pretrained("lerobot/act_aloha_sim_transfer_cube_human")  # doctest: +SKIP
-        ```
-````
-
-Add files containing runnable examples to `utils/documentation_tests.txt`.
-
-Put examples on the three to five genuine entry points of a module. Examples on trivial accessors are noise.
-
-## Three patterns you will hit constantly
-
-### Config dataclasses
-
-Configuration fields are historically documented with `#` comments above each field. **doc-builder cannot
-see inline comments** — such a class renders with every field listed and not a single description. Move them
-into an `Args:` block on the class docstring:
-
-```python
-@dataclass
-class SOFollowerConfig:
-    """Configuration for SO-family follower arms.
-
-    Args:
-        port (`str`):
-            Serial port the arm is connected to, e.g. `/dev/ttyACM0`.
-        max_relative_target (`float | dict[str, float]`, *optional*):
-            Caps the magnitude of the relative positional target vector. A scalar applies to all motors;
-            a dict maps motor name to a per-motor cap. `None` disables clipping.
-        use_degrees (`bool`, *optional*, defaults to `True`):
-            Keep `True` for backward compatibility with existing policies and datasets.
-    """
-
-    port: str
-    max_relative_target: float | dict[str, float] | None = None
-    use_degrees: bool = True
-```
-
-> [!IMPORTANT]
-> **doc-builder does not inherit docstrings from base classes.** LeRobot's registered config classes are
-> often thin multiple-inheritance shims:
->
-> ```python
-> @RobotConfig.register_subclass("so101_follower")
-> @dataclass
-> class SOFollowerRobotConfig(RobotConfig, SOFollowerConfig):
->     pass
-> ```
->
-> That class renders **every** field — including the ones it inherits — with no descriptions at all, no
-> matter how well the bases are documented. The `Args:` block must live on the concrete class that
-> `[[autodoc]]` names, and it must cover inherited fields too.
-
-### Base class, then concrete subclass
-
-The abstract base carries the canonical contract. Subclasses document only what deviates — port semantics,
-calibration quirks, motor layout, supported feature keys. Do not copy the base contract into every subclass.
-
-`Robot`, `Teleoperator`, `Camera`, `MotorsBus`, `ProcessorStep`, and `PreTrainedPolicy` all follow this
-shape.
-
-### Module-level aliases
-
-Several public names are aliases rather than distinct classes:
-
-```python
-SO100FollowerConfig = SOFollowerRobotConfig
-SO101FollowerConfig = SOFollowerRobotConfig
-```
-
-`[[autodoc]]` resolves the alias and renders the **canonical** class name, so a `## SO101FollowerConfig`
-heading will show `class lerobot.robots.so_follower.SOFollowerRobotConfig` in the body. Document the
-canonical class once, and mention the aliases in the page's prose rather than giving each alias its own
-autodoc block.
-
-## What not to document
-
-- **Private members.** Anything starting with `_` is not part of the public API.
-- **The type annotation restated as prose.** `port (`str`): A string.` adds nothing. Say what it is for.
-- **Vendored upstream code.** `src/lerobot/policies/molmoact2/molmoact2_hf_model/` is vendored from
-  `transformers` and already carries upstream-style docstrings. Leave it alone — restyling it only creates
-  conflicts on the next sync. It is excluded from the API reference and from the docstring checks.
-
-## How this is enforced
-
-| Check                     | What it catches                                                                                            |
-| ------------------------- | ---------------------------------------------------------------------------------------------------------- |
-| `make check-docstrings`   | An `Args:` entry that doesn't match the signature; a documented default that has drifted from the real one |
-| `make doctest`            | Examples that no longer run                                                                                |
-| `make check-doctest-list` | Stale or unsorted entries in `utils/documentation_tests.txt`                                               |
-| `ruff` (`D` rules)        | Google-convention style violations                                                                         |
-| `interrogate`             | Docstring coverage falling below the current threshold                                                     |
-| doc-builder               | A `[[autodoc]]` path that points at something that doesn't exist — this breaks the docs build              |
-
-Run them together before opening a PR:
-
-```bash
-make check-docstrings && make doctest && pre-commit run --all-files
-```
-
-Then render the page and actually look at it:
-
-```bash
-doc-builder build lerobot docs/source/ --build_dir /tmp/doc-build
-```
-
-## Checklist
-
-- [ ] Every public member you touched has a docstring.
-- [ ] Every `Args:` entry matches the signature, including the `*optional*, defaults to` clause.
-- [ ] `Returns:` is type-first on one indented line.
-- [ ] No bare `Attributes:` — use `**Attributes**:` with `--` separators.
-- [ ] No Sphinx roles — cross-references use &#91;`~module.Class.method`&#93;.
-- [ ] Examples are inside a fenced ` ```python ` block, and either run in CI or carry `# doctest: +SKIP`.
-- [ ] Config dataclass fields are in an `Args:` block on the concrete class, not `#` comments.
-- [ ] The rendered page has been eyeballed.

examples/openarm/README.md

@@ -0,0 +1,95 @@
+# OpenArm — Episode Replay in Simulation
+
+Replay a recorded bimanual-[OpenArm](https://openarm.dev) episode into an mp4 by driving the
+official OpenArm MuJoCo model from a LeRobot dataset's recorded joint states.
+
+By default the replay goes **through end-effector (Cartesian) space**: for every frame and
+every arm it runs forward kinematics (recorded joints → EE pose) and then inverse kinematics
+(EE pose → joints), and drives the simulator with the IK-recovered joints. This exercises the
+exact `RobotKinematics` solver that `OpenArmFollower.make_kinematics()` builds (see the
+[OpenArm docs](../../docs/source/openarm.mdx)), so the video is a visual sanity check of the
+end-effector kinematics — not just of the raw recording. It also prints the FK→IK round-trip
+error (mean joint error in degrees and mean EE-position error in mm). Pass `--joint-space` to
+bypass kinematics and replay the raw recorded joints directly.
+
+## Model provenance
+
+Everything is pulled from Enactic's official, Apache-2.0 OpenArm repositories — nothing is
+vendored into LeRobot:
+
+| Asset                                | Source                                                                          | License    |
+| ------------------------------------ | ------------------------------------------------------------------------------- | ---------- |
+| MuJoCo MJCF (rendering)             | [`enactic/openarm_mujoco`](https://github.com/enactic/openarm_mujoco)           | Apache-2.0 |
+| URDF / xacro (for `RobotKinematics`, the FK/IK round-trip) | [`enactic/openarm_description`](https://github.com/enactic/openarm_description) | Apache-2.0 |
+
+End-effector kinematics use [`placo`](https://github.com/Rhoban/placo) under the hood (install
+LeRobot with the `placo-dep` extra). The script auto-locates the URDF via `--urdf` →
+`$OPENARM_URDF`; set the tip link with `--ee-frame` → `$OPENARM_EE_FRAME`.
+
+Use the **v1** MuJoCo revision (`v1/openarm_bimanual.xml`). v2 is a different wrist hardware
+revision (DM3507) and will look sign-flipped when replaying v1 recordings.
+
+## Setup
+
+```bash
+# LeRobot in your env (see https://huggingface.co/docs/lerobot/installation)
+# Plus the sim/replay deps:
+pip install mujoco av pandas
+
+# Get the OpenArm MuJoCo model (either works):
+pip install openarm-mujoco                                   # installs models under <prefix>/share/openarm_mujoco/
+# or
+git clone https://github.com/enactic/openarm_mujoco.git      # then pass --mjcf .../v1/openarm_bimanual.xml
+```
+
+The script auto-locates the model in this order: `--mjcf` arg → `$OPENARM_MJCF` →
+`<sys.prefix>/share/openarm_mujoco/v1/openarm_bimanual.xml`.
+
+## Dataset layout
+
+`observation.state` must be the 16-D bimanual vector (degrees):
+
+```
+right_joint_1..7, right_gripper, left_joint_1..7, left_gripper
+```
+
+Only the 14 arm joints affect the rendered pose; the two gripper scalars drive the fingers.
+
+## Run
+
+Headless rendering needs `MUJOCO_GL=egl`, and MuJoCo's GL libs on `LD_LIBRARY_PATH`
+(in conda: `$CONDA_PREFIX/lib`).
+
+```bash
+# Replay episode 1 through end-effector kinematics (default)
+LD_LIBRARY_PATH=$CONDA_PREFIX/lib MUJOCO_GL=egl \
+python -m examples.openarm.render_episode \
+    --dataset data/folding_src_meta \
+    --episode 1 \
+    --urdf /path/to/openarm.urdf \
+    --ee-frame openarm_finger_tip_link \
+    --out openarm_ep1.mp4
+
+# Bypass kinematics and replay the raw recorded joints directly
+LD_LIBRARY_PATH=$CONDA_PREFIX/lib MUJOCO_GL=egl \
+python -m examples.openarm.render_episode \
+    --dataset data/folding_src_meta --episode 1 --joint-space --out openarm_ep1_raw.mp4
+
+# No dataset handy? Smoke-test with a synthetic wave (add --urdf to also exercise the kinematics):
+LD_LIBRARY_PATH=$CONDA_PREFIX/lib MUJOCO_GL=egl \
+python -m examples.openarm.render_episode --demo --joint-space --out openarm_demo.mp4
+```
+
+Useful flags: `--fps` (default 30), `--width` / `--height` (default 960×720), `--mjcf` to point
+at an explicit model file, `--urdf` / `--ee-frame` for the kinematics, `--joint-space` to skip it.
+
+## Troubleshooting
+
+| Symptom                               | Fix                                                                    |
+| ------------------------------------- | ---------------------------------------------------------------------- |
+| `Could not find the OpenArm v1 MJCF`  | Pass `--mjcf`, set `$OPENARM_MJCF`, or install/clone `openarm_mujoco`. |
+| `End-effector replay needs the OpenArm URDF` | Pass `--urdf` / set `$OPENARM_URDF`, or use `--joint-space`.     |
+| Large FK→IK round-trip error reported | Wrong `--ee-frame` link name, or the URDF doesn't match the recording. |
+| `libEGL`/`GLEW` / blank window errors | Ensure `MUJOCO_GL=egl` and `LD_LIBRARY_PATH=$CONDA_PREFIX/lib`.        |
+| Wrists look mirrored / flipped        | You are on the v2 model; switch to **v1**.                             |
+| `KeyError: 'observation.state'`       | Dataset isn't in the expected 16-D bimanual layout.                    |

examples/openarm/check_cameras.py

@@ -0,0 +1,68 @@
+#!/usr/bin/env python
+"""Grab one frame from each rollout camera and save labeled PNGs to confirm mapping.
+
+Uses the *same* V4L2 + MJPG + resolution settings as the rollout command so what you
+see is what the policy sees. Run in the lerobot312 env (has cv2):
+
+    python examples/openarm/check_cameras.py
+
+Then open the printed PNG paths and check that left_wrist / right_wrist are correct.
+"""
+
+from __future__ import annotations
+
+import os
+
+import cv2
+
+# label -> (device, width, height) — matches the rollout --robot.cameras block
+CAMERAS = {
+    "left_wrist": ("/dev/video8", 1280, 720),
+    "base": ("/dev/video6", 640, 480),
+    "right_wrist": ("/dev/video4", 1280, 720),
+}
+
+OUT_DIR = os.path.dirname(os.path.abspath(__file__))
+WARMUP_FRAMES = 12  # let auto-exposure/white-balance settle
+
+
+def fourcc_str(cap) -> str:
+    v = int(cap.get(cv2.CAP_PROP_FOURCC))
+    return "".join(chr((v >> (8 * i)) & 0xFF) for i in range(4))
+
+
+def main() -> None:
+    for label, (dev, w, h) in CAMERAS.items():
+        cap = cv2.VideoCapture(dev, cv2.CAP_V4L2)
+        if not cap.isOpened():
+            print(f"[{label}] {dev}: FAILED to open")
+            continue
+        cap.set(cv2.CAP_PROP_FOURCC, cv2.VideoWriter_fourcc(*"MJPG"))
+        cap.set(cv2.CAP_PROP_FRAME_WIDTH, w)
+        cap.set(cv2.CAP_PROP_FRAME_HEIGHT, h)
+        cap.set(cv2.CAP_PROP_FPS, 30)
+
+        frame = None
+        for _ in range(WARMUP_FRAMES):
+            ok, f = cap.read()
+            if ok:
+                frame = f
+        aw = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+        ah = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
+        fc = fourcc_str(cap)
+        cap.release()
+
+        if frame is None:
+            print(f"[{label}] {dev}: opened ({aw}x{ah} {fc}) but no frame read")
+            continue
+
+        # Burn the label into the image so the saved file is self-identifying.
+        cv2.putText(frame, f"{label}  {dev}", (20, 50), cv2.FONT_HERSHEY_SIMPLEX, 1.2,
+                    (0, 255, 0), 3, cv2.LINE_AA)
+        out = os.path.join(OUT_DIR, f"cam_{label}.png")
+        cv2.imwrite(out, frame)
+        print(f"[{label}] {dev}: {aw}x{ah} {fc}  -> {out}")
+
+
+if __name__ == "__main__":
+    main()

examples/openarm/openarm_v1_long.urdf

@@ -0,0 +1,607 @@
+<?xml version="1.0" ?>
+<!-- =================================================================================== -->
+<!-- |    This document was autogenerated by xacro from openarm_v10.urdf.xacro         | -->
+<!-- |    EDITING THIS FILE BY HAND IS NOT RECOMMENDED                                 | -->
+<!-- =================================================================================== -->
+<robot name="openarm">
+  <link name="world"/>
+  <joint name="openarm_body_world_joint" type="fixed">
+    <parent link="world"/>
+    <child link="openarm_body_link0"/>
+    <origin rpy="0 0 0" xyz="0 0 0"/>
+  </joint>
+  <link name="openarm_body_link0">
+    <visual name="openarm_body_link0_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.0 0.0"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/body/visual/body_link0.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_body_link0_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.0 0.0"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/body/collision/body_link0_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.0 0.0"/>
+      <mass value="13.89"/>
+      <inertia ixx="1.653" ixy="0.0" ixz="0.0" iyy="1.653" iyz="0.0" izz="0.051"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_openarm_body_link0_joint" type="fixed">
+    <parent link="openarm_body_link0"/>
+    <child link="openarm_left_link0"/>
+    <origin rpy="-1.5708 0 0" xyz="0 0.0410356 0.746983"/>
+  </joint>
+  <link name="openarm_left_link0">
+    <visual name="openarm_left_link0_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.0 0.0"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link0.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link0_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.0 0.0"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link0_symp.stl" scale="0.001 -0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0009483362816297526 -0.0001580207020448382 0.03076860287587199"/>
+      <mass value="1.1432284943239561"/>
+      <inertia ixx="0.001128" ixy="4e-06" ixz="-3.3e-05" iyy="0.000962" iyz="7e-06" izz="0.00147"/>
+    </inertial>
+  </link>
+  <link name="openarm_left_link1">
+    <visual name="openarm_left_link1_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 0.0 -0.0625"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link1.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link1_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 0.0 -0.0625"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link1_symp.stl" scale="0.001 -0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="0.0011467657911800769 -3.319987657026362e-05 0.05395284380736254"/>
+      <mass value="1.1416684646202298"/>
+      <inertia ixx="0.001567" ixy="1e-06" ixz="-2.9e-05" iyy="0.001273" iyz="-1e-06" izz="0.001016"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_joint1" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0 0.0 0.0625"/>
+    <parent link="openarm_left_link0"/>
+    <child link="openarm_left_link1"/>
+    <axis xyz="0 0 1"/>
+    <limit effort="40" lower="-3.490659" upper="1.3962629999999998" velocity="16.754666"/>
+  </joint>
+  <link name="openarm_left_link2">
+    <visual name="openarm_left_link2_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0301 0.0 -0.1225"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link2.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link2_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0301 0.0 -0.1225"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link2_symp.stl" scale="0.001 -0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="0.00839629182351943 2.0145102027597523e-08 0.03256649300522363"/>
+      <mass value="0.2775092746011571"/>
+      <inertia ixx="0.000359" ixy="-1e-06" ixz="-0.000109" iyy="0.000376" iyz="-1e-06" izz="0.000232"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_joint2" type="revolute">
+    <origin rpy="-1.57079632679 0 0" xyz="-0.0301 0.0 0.06"/>
+    <parent link="openarm_left_link1"/>
+    <child link="openarm_left_link2"/>
+    <axis xyz="-1 0 0"/>
+    <limit effort="40" lower="-3.3161253267948965" upper="0.17453267320510335" velocity="16.754666"/>
+  </joint>
+  <link name="openarm_left_link3">
+    <visual name="openarm_left_link3_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.18875"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link3.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link3_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.18875"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link3_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.002104752099628911 0.0005549085042607548 0.09047470545721961"/>
+      <mass value="1.073863338202347"/>
+      <inertia ixx="0.004372" ixy="1e-06" ixz="1.1e-05" iyy="0.004319" iyz="-3.6e-05" izz="0.000661"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_joint3" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0301 0.0 0.06625"/>
+    <parent link="openarm_left_link2"/>
+    <child link="openarm_left_link3"/>
+    <axis xyz="0 0 1"/>
+    <limit effort="27" lower="-1.570796" upper="1.570796" velocity="5.445426"/>
+  </joint>
+  <link name="openarm_left_link4">
+    <visual name="openarm_left_link4_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0315 -0.3425"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link4.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link4_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0315 -0.3425"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link4_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0029006831074562967 -0.03030575826634669 0.06339637422196209"/>
+      <mass value="0.6348534566833373"/>
+      <inertia ixx="0.000623" ixy="-1e-06" ixz="-1.9e-05" iyy="0.000511" iyz="3.8e-05" izz="0.000334"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_joint4" type="revolute">
+    <origin rpy="0 0 0" xyz="-0 0.0415354 0.202733"/>
+    <parent link="openarm_left_link3"/>
+    <child link="openarm_left_link4"/>
+    <axis xyz="0 1 0"/>
+    <limit effort="27" lower="0.0" upper="2.443461" velocity="5.445426"/>
+  </joint>
+  <link name="openarm_left_link5">
+    <visual name="openarm_left_link5_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.438"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link5.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link5_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.438"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link5_symp.stl" scale="0.001 -0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.003049665024221911 -0.0008866902457326625 0.043079803024980934"/>
+      <mass value="0.6156588026168502"/>
+      <inertia ixx="0.000423" ixy="8e-06" ixz="6e-06" iyy="0.000445" iyz="6e-06" izz="0.000324"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_joint5" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0 -0.0315 0.0955"/>
+    <parent link="openarm_left_link4"/>
+    <child link="openarm_left_link5"/>
+    <axis xyz="0 0 1"/>
+    <limit effort="7" lower="-1.570796" upper="1.570796" velocity="20.943946"/>
+  </joint>
+  <link name="openarm_left_link6">
+    <visual name="openarm_left_link6_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0375 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link6.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link6_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0375 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link6_symp.stl" scale="0.001 -0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.037136587005447405 -0.00033230528343419053 -9.498374522309838e-05"/>
+      <mass value="0.475202773187987"/>
+      <inertia ixx="0.000143" ixy="-1e-06" ixz="1e-06" iyy="0.000157" iyz="-1e-06" izz="0.000159"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_joint6" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0375 0.0 0.1205"/>
+    <parent link="openarm_left_link5"/>
+    <child link="openarm_left_link6"/>
+    <axis xyz="1 0 0"/>
+    <limit effort="7" lower="-0.785398" upper="0.785398" velocity="20.943946"/>
+  </joint>
+  <link name="openarm_left_link7">
+    <visual name="openarm_left_link7_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link7.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_link7_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link7_symp.stl" scale="0.001 -0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="6.875510271106056e-05 -0.01266175250761268 0.06951945409987448"/>
+      <mass value="0.4659771327380578"/>
+      <inertia ixx="0.000639" ixy="-1e-06" ixz="1e-06" iyy="0.000497" iyz="-8.9e-05" izz="0.000342"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_joint7" type="revolute">
+    <origin rpy="0 0 0" xyz="-0.0375 0.0 0.0"/>
+    <parent link="openarm_left_link6"/>
+    <child link="openarm_left_link7"/>
+    <axis xyz="0 -1 0"/>
+    <limit effort="7" lower="-1.570796" upper="1.570796" velocity="20.943946"/>
+  </joint>
+  <joint name="openarm_right_openarm_body_link0_joint" type="fixed">
+    <parent link="openarm_body_link0"/>
+    <child link="openarm_right_link0"/>
+    <origin rpy="1.5708 0 0" xyz="0 -0.0209647 0.746983"/>
+  </joint>
+  <link name="openarm_right_link0">
+    <visual name="openarm_right_link0_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.0 0.0"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link0.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link0_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.0 0.0"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link0_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0009483362816297526 0.0001580207020448382 0.03076860287587199"/>
+      <mass value="1.1432284943239561"/>
+      <inertia ixx="0.001128" ixy="-4e-06" ixz="-3.3e-05" iyy="0.000962" iyz="-7e-06" izz="0.00147"/>
+    </inertial>
+  </link>
+  <link name="openarm_right_link1">
+    <visual name="openarm_right_link1_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 0.0 -0.0625"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link1.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link1_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 0.0 -0.0625"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link1_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="0.0011467657911800769 3.319987657026362e-05 0.05395284380736254"/>
+      <mass value="1.1416684646202298"/>
+      <inertia ixx="0.001567" ixy="-1e-06" ixz="-2.9e-05" iyy="0.001273" iyz="1e-06" izz="0.001016"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_joint1" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0 0.0 0.0625"/>
+    <parent link="openarm_right_link0"/>
+    <child link="openarm_right_link1"/>
+    <axis xyz="0 0 1"/>
+    <limit effort="40" lower="-1.396263" upper="3.490659" velocity="16.754666"/>
+  </joint>
+  <link name="openarm_right_link2">
+    <visual name="openarm_right_link2_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0301 0.0 -0.1225"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link2.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link2_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0301 0.0 -0.1225"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link2_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="0.00839629182351943 -2.0145102027597523e-08 0.03256649300522363"/>
+      <mass value="0.2775092746011571"/>
+      <inertia ixx="0.000359" ixy="1e-06" ixz="-0.000109" iyy="0.000376" iyz="1e-06" izz="0.000232"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_joint2" type="revolute">
+    <origin rpy="1.57079632679 0 0" xyz="-0.0301 0.0 0.06"/>
+    <parent link="openarm_right_link1"/>
+    <child link="openarm_right_link2"/>
+    <axis xyz="-1 0 0"/>
+    <limit effort="40" lower="-0.17453267320510335" upper="3.3161253267948965" velocity="16.754666"/>
+  </joint>
+  <link name="openarm_right_link3">
+    <visual name="openarm_right_link3_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.18875"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link3.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link3_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.18875"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link3_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.002104752099628911 0.0005549085042607548 0.09047470545721961"/>
+      <mass value="1.073863338202347"/>
+      <inertia ixx="0.004372" ixy="1e-06" ixz="1.1e-05" iyy="0.004319" iyz="-3.6e-05" izz="0.000661"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_joint3" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0301 0.0 0.06625"/>
+    <parent link="openarm_right_link2"/>
+    <child link="openarm_right_link3"/>
+    <axis xyz="0 0 1"/>
+    <limit effort="27" lower="-1.570796" upper="1.570796" velocity="5.445426"/>
+  </joint>
+  <link name="openarm_right_link4">
+    <visual name="openarm_right_link4_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0315 -0.3425"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link4.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link4_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0315 -0.3425"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link4_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0029006831074562967 -0.03030575826634669 0.06339637422196209"/>
+      <mass value="0.6348534566833373"/>
+      <inertia ixx="0.000623" ixy="-1e-06" ixz="-1.9e-05" iyy="0.000511" iyz="3.8e-05" izz="0.000334"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_joint4" type="revolute">
+    <origin rpy="0 0 0" xyz="-0 0.0415354 0.202733"/>
+    <parent link="openarm_right_link3"/>
+    <child link="openarm_right_link4"/>
+    <axis xyz="0 1 0"/>
+    <limit effort="27" lower="0.0" upper="2.443461" velocity="5.445426"/>
+  </joint>
+  <link name="openarm_right_link5">
+    <visual name="openarm_right_link5_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.438"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link5.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link5_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0 -0.0 -0.438"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link5_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.003049665024221911 0.0008866902457326625 0.043079803024980934"/>
+      <mass value="0.6156588026168502"/>
+      <inertia ixx="0.000423" ixy="-8e-06" ixz="6e-06" iyy="0.000445" iyz="-6e-06" izz="0.000324"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_joint5" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0 -0.0315 0.0955"/>
+    <parent link="openarm_right_link4"/>
+    <child link="openarm_right_link5"/>
+    <axis xyz="0 0 1"/>
+    <limit effort="7" lower="-1.570796" upper="1.570796" velocity="20.943946"/>
+  </joint>
+  <link name="openarm_right_link6">
+    <visual name="openarm_right_link6_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0375 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link6.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link6_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="-0.0375 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link6_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="-0.037136587005447405 0.00033230528343419053 -9.498374522309838e-05"/>
+      <mass value="0.475202773187987"/>
+      <inertia ixx="0.000143" ixy="1e-06" ixz="1e-06" iyy="0.000157" iyz="1e-06" izz="0.000159"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_joint6" type="revolute">
+    <origin rpy="0 0 0" xyz="0.0375 0.0 0.1205"/>
+    <parent link="openarm_right_link5"/>
+    <child link="openarm_right_link6"/>
+    <axis xyz="1 0 0"/>
+    <limit effort="7" lower="-0.785398" upper="0.785398" velocity="20.943946"/>
+  </joint>
+  <link name="openarm_right_link7">
+    <visual name="openarm_right_link7_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/visual/link7.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_link7_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.0 -0.5585"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/robot/openarm_v1.0/mesh/arm/collision/link7_symp.stl" scale="0.001 0.001 0.001"/>
+        <!-- <mesh filename="package://openarm_description/meshes/arm/${arm_type}/collision/${name}.stl" scale="0.001 0.001 0.001" /> -->
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0.0 0.0 0.0" xyz="6.875510271106056e-05 0.01266175250761268 0.06951945409987448"/>
+      <mass value="0.4659771327380578"/>
+      <inertia ixx="0.000639" ixy="1e-06" ixz="1e-06" iyy="0.000497" iyz="8.9e-05" izz="0.000342"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_joint7" type="revolute">
+    <origin rpy="0 0 0" xyz="-0.0375 0.0 0.0"/>
+    <parent link="openarm_right_link6"/>
+    <child link="openarm_right_link7"/>
+    <axis xyz="0 1 0"/>
+    <limit effort="7" lower="-1.570796" upper="1.570796" velocity="20.943946"/>
+  </joint>
+  <link name="openarm_left_hand_tcp"/>
+  <joint name="openarm_left_hand_tcp_joint" type="fixed">
+    <origin rpy="0 0 0" xyz="0 0 0"/>
+    <parent link="openarm_left_link7"/>
+    <child link="openarm_left_hand_tcp"/>
+  </joint>
+  <link name="openarm_left_left_finger">
+    <visual name="openarm_left_left_finger_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/visual/finger.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_left_finger_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/collision/finger.stl" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0 0 0" xyz="0.0064528 0.01702 0.0219685"/>
+      <mass value="0.03602545343277134"/>
+      <inertia ixx="2.3749999999999997e-06" ixy="1e-06" ixz="1e-06" iyy="2.3749999999999997e-06" iyz="1e-06" izz="7.5e-07"/>
+    </inertial>
+  </link>
+  <link name="openarm_left_right_finger">
+    <visual name="openarm_left_right_finger_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/visual/finger.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_left_right_finger_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/collision/finger.stl" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0 0 0" xyz="0.0064528 -0.01702 0.0219685"/>
+      <mass value="0.03602545343277134"/>
+      <inertia ixx="2.3749999999999997e-06" ixy="1e-06" ixz="1e-06" iyy="2.3749999999999997e-06" iyz="1e-06" izz="7.5e-07"/>
+    </inertial>
+  </link>
+  <joint name="openarm_left_finger_joint1" type="prismatic">
+    <parent link="openarm_left_link7"/>
+    <child link="openarm_left_right_finger"/>
+    <origin rpy="0 0 0" xyz="0.0 0.0 0.1025"/>
+    <axis xyz="0 -1 0"/>
+    <limit effort="333" lower="0.0" upper="0.044" velocity="10.0"/>
+  </joint>
+  <joint name="openarm_left_finger_joint2" type="prismatic">
+    <parent link="openarm_left_link7"/>
+    <child link="openarm_left_left_finger"/>
+    <origin rpy="0 0 0" xyz="0.0 -0.0 0.1025"/>
+    <axis xyz="0 1 0"/>
+    <limit effort="333" lower="0.0" upper="0.044" velocity="10.0"/>
+    <mimic joint="openarm_left_finger_joint1"/>
+  </joint>
+  <!-- <joint name="${ee_prefix}finger_joint1" type="prismatic">
+      <parent link="${connected_to}" /> <child link="${ee_prefix}right_finger" />
+      <origin xyz="0 -0.006 0.115" rpy="0 0 0" />
+      <axis xyz="0 -1 0" />
+      <limit effort="333" lower="0.0" upper="0.044" velocity="10.0" />
+    </joint>
+
+    <joint name="${ee_prefix}finger_joint2" type="prismatic">
+      <parent link="${connected_to}" /> <child link="${ee_prefix}left_finger" />
+      <origin xyz="0 0.006 0.115" rpy="0 0 0" />
+      <axis xyz="0 1 0" />
+      <limit effort="333" lower="0.0" upper="0.044" velocity="10.0" />
+      <mimic joint="${ee_prefix}finger_joint1" />
+    </joint> -->
+  <link name="openarm_right_hand_tcp"/>
+  <joint name="openarm_right_hand_tcp_joint" type="fixed">
+    <origin rpy="0 0 0" xyz="0 0 0"/>
+    <parent link="openarm_right_link7"/>
+    <child link="openarm_right_hand_tcp"/>
+  </joint>
+  <link name="openarm_right_left_finger">
+    <visual name="openarm_right_left_finger_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/visual/finger.dae" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_left_finger_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 -0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/collision/finger.stl" scale="0.001 0.001 0.001"/>
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0 0 0" xyz="0.0064528 0.01702 0.0219685"/>
+      <mass value="0.03602545343277134"/>
+      <inertia ixx="2.3749999999999997e-06" ixy="1e-06" ixz="1e-06" iyy="2.3749999999999997e-06" iyz="1e-06" izz="7.5e-07"/>
+    </inertial>
+  </link>
+  <link name="openarm_right_right_finger">
+    <visual name="openarm_right_right_finger_visual">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/visual/finger.dae" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </visual>
+    <collision name="openarm_right_right_finger_collision">
+      <origin rpy="0.0 0.0 0.0" xyz="0.0 0.05 -0.673001"/>
+      <geometry>
+        <mesh filename="package://openarm_description/assets/end_effector/parallel_link/meshes/collision/finger.stl" scale="0.001 -0.001 0.001"/>
+      </geometry>
+    </collision>
+    <inertial>
+      <origin rpy="0 0 0" xyz="0.0064528 -0.01702 0.0219685"/>
+      <mass value="0.03602545343277134"/>
+      <inertia ixx="2.3749999999999997e-06" ixy="1e-06" ixz="1e-06" iyy="2.3749999999999997e-06" iyz="1e-06" izz="7.5e-07"/>
+    </inertial>
+  </link>
+  <joint name="openarm_right_finger_joint1" type="prismatic">
+    <parent link="openarm_right_link7"/>
+    <child link="openarm_right_right_finger"/>
+    <origin rpy="0 0 0" xyz="0.0 0.0 0.1025"/>
+    <axis xyz="0 -1 0"/>
+    <limit effort="333" lower="0.0" upper="0.044" velocity="10.0"/>
+  </joint>
+  <joint name="openarm_right_finger_joint2" type="prismatic">
+    <parent link="openarm_right_link7"/>
+    <child link="openarm_right_left_finger"/>
+    <origin rpy="0 0 0" xyz="0.0 -0.0 0.1025"/>
+    <axis xyz="0 1 0"/>
+    <limit effort="333" lower="0.0" upper="0.044" velocity="10.0"/>
+    <mimic joint="openarm_right_finger_joint1"/>
+  </joint>
+  <!-- <joint name="${ee_prefix}finger_joint1" type="prismatic">
+      <parent link="${connected_to}" /> <child link="${ee_prefix}right_finger" />
+      <origin xyz="0 -0.006 0.115" rpy="0 0 0" />
+      <axis xyz="0 -1 0" />
+      <limit effort="333" lower="0.0" upper="0.044" velocity="10.0" />
+    </joint>
+
+    <joint name="${ee_prefix}finger_joint2" type="prismatic">
+      <parent link="${connected_to}" /> <child link="${ee_prefix}left_finger" />
+      <origin xyz="0 0.006 0.115" rpy="0 0 0" />
+      <axis xyz="0 1 0" />
+      <limit effort="333" lower="0.0" upper="0.044" velocity="10.0" />
+      <mimic joint="${ee_prefix}finger_joint1" />
+    </joint> -->
+</robot>

examples/openarm/render_episode.py

@@ -0,0 +1,344 @@
+#!/usr/bin/env python
+
+# Copyright 2025 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Replay a recorded bimanual-OpenArm episode into an mp4, in simulation.
+
+This drives the official OpenArm MuJoCo model (``enactic/openarm_mujoco``, v1) from a
+LeRobot dataset's recorded ``observation.state`` and renders a headless video.
+
+By default the replay goes *through end-effector (Cartesian) space* rather than pushing the
+recorded joint angles straight into the simulator: for every frame and every arm it runs
+forward kinematics (recorded joints -> EE pose) and then inverse kinematics (EE pose ->
+joints), and drives MuJoCo with the IK-recovered joints. This exercises the exact
+``lerobot.model.RobotKinematics`` solver that ``OpenArmFollower.make_kinematics()`` builds,
+so the rendered video is a visual sanity check of the OpenArm end-effector kinematics -- not
+just of the raw recording. Pass ``--joint-space`` to bypass kinematics and replay the raw
+recorded joints directly (the previous behaviour).
+
+The ``observation.state`` is expected in the 16-D bimanual layout (degrees):
+    right_joint_1..7, right_gripper, left_joint_1..7, left_gripper
+
+The single published OpenArm URDF is *bimanual*, so end-effector kinematics build one solver
+per arm (right/left), each keyed to that arm's joints (``openarm_<side>_joint1..7``) and
+tool-center frame (``openarm_<side>_hand_tcp``). Solving in the full bimanual frame is what
+keeps both arms at their correct, matching heights.
+
+Assets (both Apache-2.0, nothing vendored into LeRobot):
+    MuJoCo MJCF: https://github.com/enactic/openarm_mujoco  (use the **v1** revision; v2 is a
+        different wrist hardware revision and will look sign-flipped on v1 recordings).
+    URDF (for RobotKinematics): https://github.com/enactic/openarm_description
+
+Examples:
+    # replay episode 1 of a local LeRobot v3.0 dataset through EE kinematics (bimanual URDF)
+    LD_LIBRARY_PATH=$CONDA_PREFIX/lib MUJOCO_GL=egl python -m examples.openarm.render_episode \
+        --dataset data/folding_src_meta --episode 1 \
+        --urdf /path/to/openarm_bimanual.urdf \
+        --out openarm_ep1.mp4
+
+    # smoke test with a synthetic wave (no dataset; add --urdf to also exercise the kinematics)
+    LD_LIBRARY_PATH=$CONDA_PREFIX/lib MUJOCO_GL=egl python -m examples.openarm.render_episode \
+        --demo --out openarm_demo.mp4
+
+    # bypass kinematics and replay the raw recorded joints directly
+    LD_LIBRARY_PATH=$CONDA_PREFIX/lib MUJOCO_GL=egl python -m examples.openarm.render_episode \
+        --dataset data/folding_src_meta --episode 1 --joint-space --out openarm_ep1_raw.mp4
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+import sys
+from pathlib import Path
+
+os.environ.setdefault("MUJOCO_GL", "egl")  # headless GPU rendering
+
+import numpy as np
+
+# Policy/dataset state layout: 16-D, degrees.
+POLICY_ORDER = [
+    *(f"right_joint_{i}" for i in range(1, 8)),
+    "right_gripper",
+    *(f"left_joint_{i}" for i in range(1, 8)),
+    "left_gripper",
+]
+# MuJoCo joint name for each of the 14 arm entries (grippers handled separately).
+ARM_MAP = {
+    **{f"right_joint_{i}": f"openarm_right_joint{i}" for i in range(1, 8)},
+    **{f"left_joint_{i}": f"openarm_left_joint{i}" for i in range(1, 8)},
+}
+RIGHT_GRIPPER_IDX, LEFT_GRIPPER_IDX = 7, 15
+GRIP_FULL_DEG = 65.0  # follower gripper limit magnitude -> fully open
+
+# Per-arm slices into the 16-D state (7 arm joints each; grippers handled separately).
+ARM_JOINT_SLICES = {"right": slice(0, 7), "left": slice(8, 15)}
+# The only published OpenArm URDF is *bimanual*, so each arm has its own joint names and
+# end-effector (tool-center-point) frame. Using the bimanual URDF -- rather than a single
+# arm placed at the origin -- is what makes FK/IK return correct world poses for each arm
+# (it encodes the shoulder-mount transform), so the two arms land at the right heights.
+ARM_SIDE_JOINTS = {
+    side: [f"openarm_{side}_joint{i}" for i in range(1, 8)] for side in ("right", "left")
+}
+DEFAULT_EE_FRAMES = {side: f"openarm_{side}_hand_tcp" for side in ("right", "left")}
+
+
+def locate_mjcf(explicit: str | None) -> str:
+    """Resolve the OpenArm v1 bimanual MJCF path.
+
+    Priority: --mjcf arg, then $OPENARM_MJCF, then the file installed by the
+    ``openarm_mujoco`` pip package under ``<prefix>/share/openarm_mujoco/v1``.
+    """
+    if explicit:
+        return explicit
+    if os.environ.get("OPENARM_MJCF"):
+        return os.environ["OPENARM_MJCF"]
+    for prefix in (sys.prefix, os.environ.get("CONDA_PREFIX", "")):
+        if not prefix:
+            continue
+        cand = Path(prefix) / "share" / "openarm_mujoco" / "v1" / "openarm_bimanual.xml"
+        if cand.exists():
+            return str(cand)
+    raise SystemExit(
+        "Could not find the OpenArm v1 MJCF. Pass --mjcf /path/to/v1/openarm_bimanual.xml, "
+        "set $OPENARM_MJCF, or clone https://github.com/enactic/openarm_mujoco."
+    )
+
+
+def locate_urdf(explicit: str | None) -> str | None:
+    """Resolve the single-arm OpenArm URDF path for ``RobotKinematics`` (may be None)."""
+    if explicit:
+        return explicit
+    if os.environ.get("OPENARM_URDF"):
+        return os.environ["OPENARM_URDF"]
+    return None
+
+
+def make_kinematics(urdf_path: str, ee_frames: dict[str, str]) -> dict:
+    """Build one ``RobotKinematics`` EE solver per arm from the bimanual OpenArm URDF.
+
+    This mirrors what ``OpenArmFollower.make_kinematics()`` does per arm, but we construct
+    ``RobotKinematics`` directly (instead of instantiating an ``OpenArmFollower``) so this
+    render-only example does not pull in the CAN/motor hardware stack. Each side gets its own
+    solver keyed to that arm's URDF joint names (``openarm_<side>_joint1..7``) and tool-center
+    frame, so forward/inverse kinematics resolve in the full bimanual (world) frame -- that is
+    what keeps the two arms at their correct, matching heights.
+    """
+    from lerobot.model import RobotKinematics
+
+    return {
+        side: RobotKinematics(
+            urdf_path=urdf_path,
+            target_frame_name=ee_frames[side],
+            joint_names=ARM_SIDE_JOINTS[side],
+        )
+        for side in ("right", "left")
+    }
+
+
+def ee_roundtrip(kins: dict, traj: np.ndarray) -> tuple[np.ndarray, float, float]:
+    """Route a recorded joint trajectory through end-effector space, per arm.
+
+    For each frame and each arm, run forward kinematics (recorded joints -> EE pose) then
+    inverse kinematics (EE pose -> joints) with that arm's own solver, replacing the arm joints
+    with the IK-recovered ones. Grippers pass through unchanged. Returns the new trajectory plus
+    the mean joint and EE-position round-trip errors (a sanity check that the solver tracks the
+    recording).
+    """
+    out = traj.copy()
+    joint_err = []
+    pos_err = []
+    n_arm = len(ARM_SIDE_JOINTS["right"])
+    for t in range(traj.shape[0]):
+        for side, sl in ARM_JOINT_SLICES.items():
+            kin = kins[side]
+            recorded = traj[t, sl].astype(np.float64)
+            ee_pose = kin.forward_kinematics(recorded)
+            recovered = kin.inverse_kinematics(recorded, ee_pose)[:n_arm]
+            out[t, sl] = recovered
+            joint_err.append(np.abs(recovered - recorded).mean())
+            # EE position after IK vs. the FK target, to catch non-converged solves.
+            pos_err.append(np.linalg.norm(kin.forward_kinematics(recovered)[:3, 3] - ee_pose[:3, 3]))
+    return out, float(np.mean(joint_err)), float(np.mean(pos_err))
+
+
+def load_state_from_dataset(root: str, ep: int) -> np.ndarray:
+    """Read one episode's recorded observation.state (N, 16; degrees) from a LeRobot v3.0 root."""
+    import pandas as pd
+
+    root = Path(root)
+    ep_meta = pd.read_parquet(root / "meta" / "episodes" / "chunk-000" / "file-000.parquet")
+    row = ep_meta[ep_meta["episode_index"] == ep].iloc[0]
+    a, b = int(row["dataset_from_index"]), int(row["dataset_to_index"])
+    dchunk, dfile = int(row["data/chunk_index"]), int(row["data/file_index"])
+    df = pd.read_parquet(root / "data" / f"chunk-{dchunk:03d}" / f"file-{dfile:03d}.parquet")
+    df = df[(df["index"] >= a) & (df["index"] < b)].sort_values("frame_index")
+    return np.stack(df["observation.state"].to_numpy()).astype(np.float32)
+
+
+def encode_mp4(frames: list[np.ndarray], path: str, fps: int) -> None:
+    import av
+
+    h, w = frames[0].shape[:2]
+    container = av.open(path, mode="w")
+    stream = container.add_stream("libx264", rate=fps)
+    stream.width, stream.height, stream.pix_fmt = w, h, "yuv420p"
+    for f in frames:
+        frame = av.VideoFrame.from_ndarray(np.ascontiguousarray(f), format="rgb24")
+        for pkt in stream.encode(frame):
+            container.mux(pkt)
+    for pkt in stream.encode():
+        container.mux(pkt)
+    container.close()
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawTextHelpFormatter)
+    ap.add_argument("--dataset", default=None, help="LeRobot v3.0 dataset root (meta/ + data/)")
+    ap.add_argument("--episode", type=int, default=0)
+    ap.add_argument("--demo", action="store_true", help="drive a synthetic wave (no dataset)")
+    ap.add_argument("--mjcf", default=None, help="path to v1/openarm_bimanual.xml (see locate_mjcf)")
+    ap.add_argument(
+        "--urdf",
+        default=None,
+        help="bimanual OpenArm URDF for end-effector kinematics (or set $OPENARM_URDF)",
+    )
+    ap.add_argument(
+        "--right-ee-frame",
+        default=os.environ.get("OPENARM_RIGHT_EE_FRAME", DEFAULT_EE_FRAMES["right"]),
+        help="right-arm end-effector link name in the URDF (or set $OPENARM_RIGHT_EE_FRAME)",
+    )
+    ap.add_argument(
+        "--left-ee-frame",
+        default=os.environ.get("OPENARM_LEFT_EE_FRAME", DEFAULT_EE_FRAMES["left"]),
+        help="left-arm end-effector link name in the URDF (or set $OPENARM_LEFT_EE_FRAME)",
+    )
+    ap.add_argument(
+        "--joint-space",
+        action="store_true",
+        help="bypass kinematics and replay the raw recorded joints directly",
+    )
+    ap.add_argument("--out", default="openarm_episode.mp4")
+    ap.add_argument("--fps", type=int, default=30)
+    ap.add_argument("--width", type=int, default=960)
+    ap.add_argument("--height", type=int, default=720)
+    args = ap.parse_args()
+
+    import mujoco
+
+    model = mujoco.MjModel.from_xml_path(locate_mjcf(args.mjcf))
+    model.vis.global_.offwidth = max(model.vis.global_.offwidth, args.width)
+    model.vis.global_.offheight = max(model.vis.global_.offheight, args.height)
+    data = mujoco.MjData(model)
+
+    qadr = {
+        pk: int(model.jnt_qposadr[mujoco.mj_name2id(model, mujoco.mjtObj.mjOBJ_JOINT, mj)])
+        for pk, mj in ARM_MAP.items()
+    }
+
+    def finger_adr(names):
+        out = []
+        for nm in names:
+            jid = mujoco.mj_name2id(model, mujoco.mjtObj.mjOBJ_JOINT, nm)
+            out.append((int(model.jnt_qposadr[jid]), model.jnt_range[jid].copy(), int(model.jnt_type[jid])))
+        return out
+
+    right_fingers = finger_adr(["openarm_right_finger_joint1", "openarm_right_finger_joint2"])
+    left_fingers = finger_adr(["openarm_left_finger_joint1", "openarm_left_finger_joint2"])
+    hinge_type = int(mujoco.mjtJoint.mjJNT_HINGE)
+
+    def finger_target(gripper_deg, rng, jtype):
+        opening = min(1.0, abs(gripper_deg) / GRIP_FULL_DEG)  # 0=closed .. 1=open
+        if jtype == hinge_type:  # hinge in radians; sign encodes side via range direction
+            lo, hi = rng
+            mag = np.deg2rad(min(abs(gripper_deg), GRIP_FULL_DEG))
+            return np.clip(-mag if lo < 0 else mag, lo, hi)
+        return rng[0] + opening * (rng[1] - rng[0])  # slide (v1): lo=closed .. hi=open
+
+    if args.demo:
+        n_frames = 120
+        traj = np.zeros((n_frames, 16), np.float32)
+        wave = 40.0 * np.sin(np.linspace(0, 2 * np.pi, n_frames))
+        for i, pk in enumerate(POLICY_ORDER):
+            if pk in qadr:
+                traj[:, i] = wave * (0.5 + 0.5 * (i % 3))
+    elif args.dataset:
+        traj = load_state_from_dataset(args.dataset, args.episode)
+    else:
+        raise SystemExit("provide --dataset <root> (with --episode) or --demo")
+    n_frames = traj.shape[0]
+
+    # Route the recorded joints through end-effector (Cartesian) space via the same solver
+    # OpenArmFollower.make_kinematics() builds, unless the user opted for raw joint replay.
+    urdf_path = None if args.joint_space else locate_urdf(args.urdf)
+    if args.joint_space:
+        print(f"driving {n_frames} frames (raw joint space)")
+    elif urdf_path is None:
+        raise SystemExit(
+            "End-effector replay needs the OpenArm URDF: pass --urdf /path/to/openarm.urdf "
+            "(and --ee-frame if the tip link differs), set $OPENARM_URDF, or use --joint-space "
+            "to replay the raw recorded joints. URDF: https://github.com/enactic/openarm_description"
+        )
+    else:
+        ee_frames = {"right": args.right_ee_frame, "left": args.left_ee_frame}
+        kins = make_kinematics(urdf_path, ee_frames)
+        traj, joint_err, pos_err = ee_roundtrip(kins, traj)
+        print(
+            f"driving {n_frames} frames (end-effector kinematics via frames "
+            f"right='{ee_frames['right']}', left='{ee_frames['left']}'; "
+            f"FK->IK round-trip: {joint_err:.3f}° mean joint error, {pos_err * 1e3:.2f} mm mean EE error)"
+        )
+
+    # Auto-frame the arms (exclude pedestal/world) from body positions at the mid pose.
+    mid = n_frames // 2
+    for i, pk in enumerate(POLICY_ORDER):
+        if pk in qadr:
+            data.qpos[qadr[pk]] = np.deg2rad(traj[mid, i])
+    mujoco.mj_forward(model, data)
+    arm_pts = [
+        data.xpos[b].copy()
+        for b in range(model.nbody)
+        if (mujoco.mj_id2name(model, mujoco.mjtObj.mjOBJ_BODY, b) or "").startswith("openarm")
+        and "base" not in (mujoco.mj_id2name(model, mujoco.mjtObj.mjOBJ_BODY, b) or "")
+    ]
+    arm_pts = np.array(arm_pts) if arm_pts else data.xpos[1:]
+    lo, hi = arm_pts.min(0), arm_pts.max(0)
+    cam = mujoco.MjvCamera()
+    mujoco.mjv_defaultCamera(cam)
+    cam.azimuth, cam.elevation = 150.0, -20.0
+    cam.distance = max(0.8, float(np.linalg.norm(hi - lo)) * 1.3)
+    cam.lookat[:] = (lo + hi) / 2.0
+
+    renderer = mujoco.Renderer(model, height=args.height, width=args.width)
+    frames = []
+    for t in range(n_frames):
+        for i, pk in enumerate(POLICY_ORDER):
+            if pk in qadr:
+                data.qpos[qadr[pk]] = np.deg2rad(traj[t, i])
+        for adr, rng, jt in right_fingers:
+            data.qpos[adr] = finger_target(float(traj[t, RIGHT_GRIPPER_IDX]), rng, jt)
+        for adr, rng, jt in left_fingers:
+            data.qpos[adr] = finger_target(float(traj[t, LEFT_GRIPPER_IDX]), rng, jt)
+        mujoco.mj_forward(model, data)
+        renderer.update_scene(data, camera=cam)
+        frames.append(renderer.render())
+    renderer.close()
+
+    encode_mp4(frames, args.out, args.fps)
+    print(f"wrote {args.out}  ({n_frames} frames @ {args.fps} fps, {args.width}x{args.height})")
+
+
+if __name__ == "__main__":
+    main()

examples/openarm/retarget_replay.py

@@ -0,0 +1,127 @@
+#!/usr/bin/env python
+"""Replay a precomputed retargeted SHORT-arm trajectory on the real bimanual OpenArm.
+
+The trajectory (N,16 degrees, POLICY order: right_joint_1..7,right_gripper,left_joint_1..7,
+left_gripper) is produced offline by .precompute_short_traj.py as:
+    FK on the LONG model (recorded data) -> gripper-tip pose -> IK on the SHORT model,
+clamped to the real per-arm joint limits. This is the deterministic, no-policy validation of
+the morphology retarget before running the live policy rollout.
+
+Run in the deployment env (lerobot312), with CAN up (can0/can1):
+
+    python examples/openarm/retarget_replay.py \
+        --traj openarm_ep1_short_retargeted.npy \
+        --left-port can1 --right-port can0 --id openarms \
+        --max-relative-target 8.0 --fps 30 --ramp-seconds 4.0
+
+Use --dry-run first to print the ramp/first/last targets without touching the robot.
+"""
+from __future__ import annotations
+
+import argparse
+import time
+
+import numpy as np
+
+# POLICY column -> robot action key (bimanual, ".pos", degrees)
+JMAP: list[tuple[str, int]] = (
+    [(f"right_joint_{i}.pos", i - 1) for i in range(1, 8)]
+    + [("right_gripper.pos", 7)]
+    + [(f"left_joint_{i}.pos", 8 + i - 1) for i in range(1, 8)]
+    + [("left_gripper.pos", 15)]
+)
+
+
+def action_of(row: np.ndarray) -> dict[str, float]:
+    return {k: float(row[c]) for k, c in JMAP}
+
+
+def present_row(obs: dict) -> np.ndarray:
+    row = np.zeros(16, dtype=np.float64)
+    for k, c in JMAP:
+        row[c] = float(obs[k])
+    return row
+
+
+def build_robot(args):
+    from lerobot.robots.bi_openarm_follower import BiOpenArmFollower, BiOpenArmFollowerConfig
+    from lerobot.robots.openarm_follower import OpenArmFollowerConfigBase
+
+    common = dict(
+        can_interface="socketcan",
+        disable_torque_on_disconnect=True,
+        max_relative_target=args.max_relative_target,
+        cameras={},
+    )
+    cfg = BiOpenArmFollowerConfig(
+        id=args.id,
+        left_arm_config=OpenArmFollowerConfigBase(port=args.left_port, side="left", **common),
+        right_arm_config=OpenArmFollowerConfigBase(port=args.right_port, side="right", **common),
+        cameras={},
+    )
+    return BiOpenArmFollower(cfg)
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--traj", required=True, help="(N,16) npy in degrees, POLICY order")
+    ap.add_argument("--left-port", default="can1")
+    ap.add_argument("--right-port", default="can0")
+    ap.add_argument("--id", default="openarms")
+    ap.add_argument("--max-relative-target", type=float, default=8.0)
+    ap.add_argument("--fps", type=float, default=30.0)
+    ap.add_argument("--ramp-seconds", type=float, default=4.0,
+                    help="time to smoothly move from current pose to the first frame")
+    ap.add_argument("--start-index", type=int, default=0)
+    ap.add_argument("--end-index", type=int, default=-1)
+    ap.add_argument("--dry-run", action="store_true", help="print plan, do not connect")
+    args = ap.parse_args()
+
+    traj = np.load(args.traj).astype(np.float64)
+    assert traj.ndim == 2 and traj.shape[1] == 16, f"expected (N,16), got {traj.shape}"
+    end = traj.shape[0] if args.end_index < 0 else args.end_index
+    traj = traj[args.start_index:end]
+    n = traj.shape[0]
+    dt = 1.0 / args.fps
+    print(f"loaded {args.traj}: {n} frames @ {args.fps} fps (~{n*dt:.1f}s)")
+    print("first target:", np.round(traj[0], 1))
+    print("last  target:", np.round(traj[-1], 1))
+
+    if args.dry_run:
+        print("[dry-run] not connecting.")
+        return
+
+    robot = build_robot(args)
+    print("connecting... (ensure CAN is up and arms are clear)")
+    robot.connect()
+    try:
+        # --- gentle ramp from current pose to first frame ---
+        present = present_row(robot.get_observation())
+        n_ramp = max(1, int(round(args.ramp_seconds * args.fps)))
+        print(f"ramping to first frame over {args.ramp_seconds:.1f}s ({n_ramp} steps)...")
+        for k in range(1, n_ramp + 1):
+            a = k / n_ramp
+            row = (1.0 - a) * present + a * traj[0]
+            t0 = time.perf_counter()
+            robot.send_action(action_of(row))
+            time.sleep(max(0.0, dt - (time.perf_counter() - t0)))
+
+        # --- stream the trajectory ---
+        print("replaying...")
+        start = time.perf_counter()
+        for t in range(n):
+            robot.send_action(action_of(traj[t]))
+            target = start + (t + 1) * dt
+            time.sleep(max(0.0, target - time.perf_counter()))
+            if t % 60 == 0:
+                print(f"  frame {t}/{n}  ({t/args.fps:.1f}s)")
+        print("done.")
+    except KeyboardInterrupt:
+        print("\ninterrupted by user.")
+    finally:
+        print("disconnecting (torque off)...")
+        robot.disconnect()
+
+
+if __name__ == "__main__":
+    main()

examples/openarm/rollout_retarget.py

@@ -0,0 +1,615 @@
+#!/usr/bin/env python
+"""lerobot-rollout with gold-standard MuJoCo EE retargeting at the robot boundary.
+
+The policy was trained on the LONG arm (upper arm +5 cm). The real robot is the
+SHORT (stock) arm. We bridge the morphology gap with the *exact* MuJoCo FK/IK that
+produced the validated cyan/red overlay video, wrapping the raw robot so that:
+
+  * get_observation():  SHORT joints --FK(short)--> gripper-tip pose --IK(long)-->
+                        LONG joints   (the state the policy expects)
+  * send_action():      LONG joint targets --FK(long)--> pose --IK(short)-->
+                        SHORT joint targets  (clamped to the real per-arm limits)
+
+Between those two boundaries everything (state, relative-action anchor, policy
+output) lives consistently in LONG space, so no other part of the rollout stack
+needs to change. We inject the wrapper by patching
+``lerobot.rollout.context.make_robot_from_config`` and then hand off to the normal
+``lerobot-rollout`` entry point, so every CLI flag behaves identically.
+
+Usage: same args as ``lerobot-rollout``, e.g.
+
+    python examples/openarm/rollout_retarget.py \
+        --policy.path=/home/yope/Documents/sonic/data/folding_latest \
+        --robot.type=bi_openarm_follower --robot.id=openarms \
+        --robot.cameras='{ ... }' \
+        --robot.left_arm_config.port=can1 ... --task="Fold the T-shirt properly" \
+        --fps=30 --duration=2000 --device=cuda --display_data=true
+
+Env toggles (optional):
+    RETARGET_OBS=0        disable observation retargeting (short->long)
+    RETARGET_ACT=0        disable action retargeting (long->short)
+    RETARGET_ITERS=25     IK iterations per tick (warm-started)
+    RETARGET_ITERS0=80    IK iterations on the very first tick (cold seed)
+    RETARGET_NULL_GAIN=0.3   nullspace bias: pull short joints toward long joints (0 = off,
+                             EE-only). Higher keeps the elbow closer to the long pose.
+
+  Joint-space smoothing streamer (decouples motor rate from the slow control loop):
+    STREAM=1              enable the background smoothing streamer
+    STREAM_HZ=60          motor command rate of the streamer thread (Hz)
+    STREAM_SMOOTH_TIME=0.10   SmoothDamp time constant (s); larger = smoother/laggier
+    STREAM_MAX_SPEED=150  per-joint speed cap (deg/s); 0 disables the cap
+"""
+
+from __future__ import annotations
+
+import importlib.util
+import logging
+import os
+import threading
+import time
+
+os.environ.setdefault("MUJOCO_GL", "egl")
+
+import mujoco
+import numpy as np
+
+logger = logging.getLogger("rollout_retarget")
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+_REPO = os.path.abspath(os.path.join(_HERE, "..", ".."))
+
+
+def _load(name: str, path: str):
+    spec = importlib.util.spec_from_file_location(name, path)
+    mod = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)
+    return mod
+
+
+# Reuse the *validated* geometry + FK/IK helpers (same code that made the overlay video).
+ov = _load("ov", os.path.join(_REPO, ".overlay_rest.py"))
+rt = _load("rt", os.path.join(_REPO, ".roundtrip_overlay.py"))
+
+from lerobot.robots.openarm_follower.config_openarm_follower import (  # noqa: E402
+    LEFT_DEFAULT_JOINTS_LIMITS,
+    RIGHT_DEFAULT_JOINTS_LIMITS,
+)
+
+SIDES = ("right", "left")
+LIMITS = {"right": RIGHT_DEFAULT_JOINTS_LIMITS, "left": LEFT_DEFAULT_JOINTS_LIMITS}
+
+# POLICY_ORDER index of each gripper in the 16-vector (right block 0..7, left block 8..15).
+GRIPPER_IDX = {"right": 7, "left": 15}
+
+
+def smooth_damp(
+    current: np.ndarray,
+    target: np.ndarray,
+    velocity: np.ndarray,
+    smooth_time: float,
+    dt: float,
+    max_speed: float = 0.0,
+) -> tuple[np.ndarray, np.ndarray]:
+    """Critically-damped 2nd-order smoothing toward ``target`` (vectorized SmoothDamp).
+
+    Gives C1-continuous position + velocity with no overshoot, and re-plans every tick,
+    so it degrades gracefully when targets arrive irregularly or late. Returns the new
+    (position, velocity). ``max_speed <= 0`` disables the per-joint speed cap.
+    """
+    smooth_time = max(1e-4, smooth_time)
+    omega = 2.0 / smooth_time
+    x = omega * dt
+    exp = 1.0 / (1.0 + x + 0.48 * x * x + 0.235 * x * x * x)
+    change = current - target
+    original_to = target.copy()
+    if max_speed and max_speed > 0.0:
+        max_change = max_speed * smooth_time
+        change = np.clip(change, -max_change, max_change)
+    shifted_target = current - change
+    temp = (velocity + omega * change) * dt
+    velocity = (velocity - omega * temp) * exp
+    output = shifted_target + (change + temp) * exp
+    # Kill overshoot: if we crossed the original target, snap to it and match velocity.
+    overshoot = (original_to - current > 0.0) == (output > original_to)
+    output = np.where(overshoot, original_to, output)
+    velocity = np.where(overshoot, (output - original_to) / dt, velocity)
+    return output, velocity
+
+
+class RetargetRobot:
+    """Boundary wrapper that retargets state (short->long) and actions (long->short).
+
+    All non-overridden attributes/methods proxy transparently to the wrapped robot
+    (``connect``, ``disconnect``, ``observation_features``, ``cameras``, ...), so the
+    rollout stack treats it exactly like the underlying ``bi_openarm_follower``.
+    """
+
+    def __init__(
+        self,
+        robot,
+        obs_iters: int = 25,
+        act_iters: int = 25,
+        iters0: int = 80,
+        retarget_obs: bool = True,
+        retarget_act: bool = True,
+        null_gain: float = 0.3,
+        stream: bool = False,
+        stream_hz: float = 60.0,
+        smooth_time: float = 0.10,
+        max_speed: float = 150.0,
+    ) -> None:
+        self._robot = robot
+        self._lock = threading.Lock()
+        # Serialises *all* real-robot bus I/O (obs reads + streamer writes) so the
+        # observation thread and the streamer thread never touch the CAN bus at once.
+        self._io_lock = threading.Lock()
+        self._obs_iters = obs_iters
+        self._act_iters = act_iters
+        self._iters0 = iters0
+        self._do_obs = retarget_obs
+        self._do_act = retarget_act
+        # Nullspace secondary-task gain: pull the redundant DOF toward the reference
+        # (other-arm) joints so the two arms match in joint space without moving the EE.
+        self._null_gain = null_gain
+        # EE-priority guard for the nullspace bias (metres of 6-vector residual norm).
+        self._ee_tol = 0.01  # only consider a fallback when biased residual exceeds this
+        self._ee_slack = 0.005  # ...and EE-only beats it by more than this
+        # Task-only iterations appended after the biased iterations (re-tightens EE a bit
+        # without fully washing out the elbow bias; the fallback is the real EE guarantee).
+        self._final_task_iters = int(os.environ.get("RETARGET_FINAL_TASK_ITERS", "3"))
+
+        # --- joint-space smoothing streamer ---
+        self._stream = stream
+        self._stream_hz = stream_hz
+        self._smooth_time = smooth_time
+        self._max_speed = max_speed
+        self._goal_lock = threading.Lock()
+        self._goal: np.ndarray | None = None  # latest short target (16-vec, deg), POLICY order
+        self._stream_current: np.ndarray | None = None  # smoothed setpoint (16-vec, deg)
+        self._stream_vel = np.zeros(16)
+        self._stream_stop = threading.Event()
+        self._stream_thread: threading.Thread | None = None
+        # Full last real short pose (16-vec, deg) incl grippers, for streamer seeding.
+        self._last_short_full: np.ndarray | None = None
+        # Present short pose captured at connect() -> exact home to return to on shutdown.
+        self._home: np.ndarray | None = None
+
+        m_short = mujoco.MjModel.from_xml_path(ov.MJCF)
+        m_long, _ = ov.make_long(m_short)
+        self.m_short, self.m_long = m_short, m_long
+        # Separate MjData per direction so obs- and action-side solves never share buffers
+        # (the ThreadSafeRobot lock already serialises calls, but this is belt-and-braces).
+        self.d_short_o = mujoco.MjData(m_short)  # obs: FK on short
+        self.d_long_o = mujoco.MjData(m_long)  # obs: IK on long
+        self.d_long_a = mujoco.MjData(m_long)  # action: FK on long
+        self.d_short_a = mujoco.MjData(m_short)  # action: IK on short
+
+        self.qadr_s = rt.joint_adr(m_short)
+        self.qadr_l = rt.joint_adr(m_long)
+        self.tcp_s = {s: mujoco.mj_name2id(m_short, mujoco.mjtObj.mjOBJ_BODY, rt.TCP[s]) for s in SIDES}
+        self.tcp_l = {s: mujoco.mj_name2id(m_long, mujoco.mjtObj.mjOBJ_BODY, rt.TCP[s]) for s in SIDES}
+        self.dofs_s = {s: rt.arm_dofs(m_short, s) for s in SIDES}  # (qposadr, dofadr, range)
+        self.dofs_l = {s: rt.arm_dofs(m_long, s) for s in SIDES}
+        # Real per-arm limits (radians) so retargeted commands are executable without clipping.
+        self.rng_real = {
+            s: np.deg2rad(np.array([LIMITS[s][f"joint_{i}"] for i in range(1, 8)], float)) for s in SIDES
+        }
+
+        self._obs_seed: dict[str, np.ndarray | None] = {s: None for s in SIDES}
+        self._act_seed: dict[str, np.ndarray | None] = {s: None for s in SIDES}
+        # Last *real* short-arm joints (rad) seen in get_observation. Used to seed the
+        # first action-side IK so the initial command lands in the null-space branch
+        # nearest the robot's actual pose (the 7-DOF arm is redundant, so the same EE
+        # admits many joint configs; without this the first tick could command a large
+        # elbow-swivel reconfiguration toward an arbitrary branch).
+        self._last_short: dict[str, np.ndarray] = {}
+        self._first_obs = True
+        self._first_act = True
+
+        if self._stream:
+            logger.info(
+                "Smoothing streamer ENABLED (hz=%.0f, smooth_time=%.3fs, max_speed=%.0f deg/s)",
+                self._stream_hz,
+                self._smooth_time,
+                self._max_speed,
+            )
+        logger.info(
+            "RetargetRobot ready (obs=%s, act=%s, iters=%d, iters0=%d)",
+            self._do_obs,
+            self._do_act,
+            self._obs_iters,
+            self._iters0,
+        )
+
+    # -- transparent proxy for everything else (connect/disconnect/features/...) --
+    def __getattr__(self, name):
+        return getattr(object.__getattribute__(self, "_robot"), name)
+
+    # -- helpers -----------------------------------------------------------------
+    @staticmethod
+    def _read_arm_deg(d: dict, side: str) -> np.ndarray:
+        return np.array([d[f"{side}_joint_{i}.pos"] for i in range(1, 8)], float)
+
+    def _has_full_arms(self, d: dict) -> bool:
+        return all(f"{s}_joint_{i}.pos" in d for s in SIDES for i in range(1, 8))
+
+    def _ik(
+        self,
+        m,
+        d,
+        tcp_id: int,
+        dofadr: np.ndarray,
+        qadr7: np.ndarray,
+        rng7: np.ndarray,
+        pt: np.ndarray,
+        Rt: np.ndarray,
+        seed: np.ndarray,
+        q_ref: np.ndarray,
+        iters: int,
+        use_null: bool = True,
+        lam: float = 0.06,
+        step: float = 0.25,
+        null_step: float = 0.08,
+    ) -> tuple[np.ndarray, float]:
+        """Damped least-squares IK for the 6-DOF EE pose with a nullspace bias toward ``q_ref``.
+
+        Primary task: reach (pt, Rt). Secondary task (projected into the task nullspace so it
+        never disturbs the EE): minimise ||q - q_ref||, using the 7-DOF arm's 1 redundant DOF
+        to keep the elbow close to the reference (long/short) configuration.
+        """
+        q = seed.copy()
+        jacp = np.zeros((3, m.nv))
+        jacr = np.zeros((3, m.nv))
+        eye6 = np.eye(6)
+        eye7 = np.eye(7)
+        last = 1e9
+        for it in range(iters):
+            d.qpos[qadr7] = q
+            mujoco.mj_kinematics(m, d)
+            mujoco.mj_comPos(m, d)  # required by mj_jac
+            p = d.xpos[tcp_id].copy()
+            R = d.xmat[tcp_id].reshape(3, 3)
+            e = np.concatenate([pt - p, rt.rot_err(R, Rt)])
+            last = float(np.linalg.norm(e))
+            mujoco.mj_jac(m, d, jacp, jacr, p, tcp_id)
+            J = np.vstack([jacp[:, dofadr], jacr[:, dofadr]])  # 6x7
+            Jt = J.T
+            # Task step: DAMPED pinv for stability near singularities. Clipped on its own so
+            # the primary always gets its full authority.
+            dq = np.clip(Jt @ np.linalg.solve(J @ Jt + (lam**2) * eye6, e), -step, step)
+            # Apply the nullspace bias only in the early iterations; the last
+            # ``final_task_iters`` are task-only so the EE is always re-tightened from the
+            # biased configuration (strict EE priority even if q_ref is far/infeasible).
+            null_active = use_null and self._null_gain > 0.0 and (iters - it) > self._final_task_iters
+            if null_active:
+                # Nullspace projector from the TRUE pinv so J @ N == 0 exactly: the secondary
+                # (elbow-toward-q_ref) task lives purely in the redundant DOF and never moves
+                # the EE. Clipped small so it stays strictly secondary to the task step.
+                Jpinv_true = np.linalg.pinv(J)  # 7x6
+                nullproj = eye7 - Jpinv_true @ J  # 7x7
+                dq_null = np.clip(nullproj @ (self._null_gain * (q_ref - q)), -null_step, null_step)
+                dq = dq + dq_null
+            q = q + dq
+            q = np.clip(q, rng7[:, 0], rng7[:, 1])
+        return q, last
+
+    def _solve(self, m, d, tcp_id, dofadr, qadr7, rng7, pt, Rt, seed, q_ref, iters):
+        """Nullspace-biased IK with strict EE priority.
+
+        Solve with the elbow-toward-``q_ref`` bias; if that leaves an EE residual that a
+        pure EE-only solve would beat by more than ``_ee_slack``, fall back to EE-only. So the
+        bias is applied only when it is (nearly) free in EE terms — a weird/infeasible target
+        can never trade away end-effector accuracy for elbow matching.
+        """
+        q, res = self._ik(m, d, tcp_id, dofadr, qadr7, rng7, pt, Rt, seed, q_ref, iters)
+        if self._null_gain > 0.0 and res > self._ee_tol:
+            q0, res0 = self._ik(
+                m, d, tcp_id, dofadr, qadr7, rng7, pt, Rt, seed, q_ref, iters, use_null=False
+            )
+            if res0 + self._ee_slack < res:
+                return q0, res0
+        return q, res
+
+    # -- observation: SHORT joints -> LONG joints (FK short, IK long) ------------
+    def get_observation(self) -> dict:
+        with self._io_lock:
+            obs = self._robot.get_observation()
+        # Record the full real short pose (arms + grippers) for streamer seeding.
+        if self._has_full_arms(obs):
+            full = np.zeros(16)
+            for s in SIDES:
+                full[rt.ARM_JOINT_SLICES[s]] = self._read_arm_deg(obs, s)
+                gk = f"{s}_gripper.pos"
+                if gk in obs:
+                    full[GRIPPER_IDX[s]] = float(obs[gk])
+            self._last_short_full = full
+        if not self._do_obs or not self._has_full_arms(obs):
+            return obs
+        with self._lock:
+            state = np.zeros(16)
+            for s in SIDES:
+                arm = self._read_arm_deg(obs, s)
+                state[rt.ARM_JOINT_SLICES[s]] = arm
+                self._last_short[s] = np.deg2rad(arm)
+            rt.set_arms(self.m_short, self.d_short_o, self.qadr_s, state)
+            mujoco.mj_forward(self.m_short, self.d_short_o)
+            out = dict(obs)
+            for s in SIDES:
+                pt = self.d_short_o.xpos[self.tcp_s[s]].copy()
+                Rt = self.d_short_o.xmat[self.tcp_s[s]].reshape(3, 3)
+                q_ref = np.deg2rad(state[rt.ARM_JOINT_SLICES[s]])  # bias long pose toward short obs
+                seed = self._obs_seed[s]
+                if seed is None:
+                    seed = q_ref.copy()
+                iters = self._iters0 if self._first_obs else self._obs_iters
+                q7, _ = self._solve(
+                    self.m_long,
+                    self.d_long_o,
+                    self.tcp_l[s],
+                    self.dofs_l[s][1],
+                    self.dofs_l[s][0],
+                    self.dofs_l[s][2],  # clamp to LONG model ranges (policy's training space)
+                    pt,
+                    Rt,
+                    seed,
+                    q_ref,
+                    iters,
+                )
+                self._obs_seed[s] = q7
+                deg = np.rad2deg(q7)
+                for i in range(1, 8):
+                    out[f"{s}_joint_{i}.pos"] = float(deg[i - 1])
+            self._first_obs = False
+            return out
+
+    # -- action: LONG joint targets -> SHORT joint targets (FK long, IK short) ---
+    def _retarget_action(self, action: dict) -> dict:
+        """Return the short-arm action dict (IK-retargeted if enabled, else a copy)."""
+        if not self._do_act or not isinstance(action, dict) or not self._has_full_arms(action):
+            return action
+        with self._lock:
+            long_q = np.zeros(16)
+            for s in SIDES:
+                long_q[rt.ARM_JOINT_SLICES[s]] = self._read_arm_deg(action, s)
+            rt.set_arms(self.m_long, self.d_long_a, self.qadr_l, long_q)
+            mujoco.mj_forward(self.m_long, self.d_long_a)
+            out = dict(action)
+            for s in SIDES:
+                pt = self.d_long_a.xpos[self.tcp_l[s]].copy()
+                Rt = self.d_long_a.xmat[self.tcp_l[s]].reshape(3, 3)
+                q_ref = np.deg2rad(long_q[rt.ARM_JOINT_SLICES[s]])  # bias short pose toward long target
+                seed = self._act_seed[s]
+                if seed is None:
+                    # Prefer the robot's real current short pose (nearest branch, minimal
+                    # startup motion); fall back to the long target angles if unseen.
+                    seed = (
+                        self._last_short[s].copy()
+                        if s in self._last_short
+                        else q_ref.copy()
+                    )
+                iters = self._iters0 if self._first_act else self._act_iters
+                q7, _ = self._solve(
+                    self.m_short,
+                    self.d_short_a,
+                    self.tcp_s[s],
+                    self.dofs_s[s][1],
+                    self.dofs_s[s][0],
+                    self.rng_real[s],  # clamp to REAL limits -> safe on hardware
+                    pt,
+                    Rt,
+                    seed,
+                    q_ref,
+                    iters,
+                )
+                self._act_seed[s] = q7
+                deg = np.rad2deg(q7)
+                for i in range(1, 8):
+                    out[f"{s}_joint_{i}.pos"] = float(deg[i - 1])
+            self._first_act = False
+            return out
+
+    def send_action(self, action: dict):
+        short = self._retarget_action(action)
+        # Streaming path: just publish the target; the streamer thread writes to the bus.
+        if self._stream and isinstance(short, dict) and self._has_full_arms(short):
+            self._set_goal(short)
+            return short
+        with self._io_lock:
+            return self._robot.send_action(short)
+
+    # -- streamer: joint-space smoothing at a fixed high rate ---------------------
+    @staticmethod
+    def _dict_to_vec(d: dict, fallback: np.ndarray | None = None) -> np.ndarray:
+        vec = np.zeros(16) if fallback is None else fallback.copy()
+        for s in SIDES:
+            base = rt.ARM_JOINT_SLICES[s].start
+            for i in range(1, 8):
+                vec[base + i - 1] = float(d[f"{s}_joint_{i}.pos"])
+            gk = f"{s}_gripper.pos"
+            if gk in d:
+                vec[GRIPPER_IDX[s]] = float(d[gk])
+        return vec
+
+    @staticmethod
+    def _vec_to_action(vec: np.ndarray) -> dict:
+        out = {}
+        for s in SIDES:
+            base = rt.ARM_JOINT_SLICES[s].start
+            for i in range(1, 8):
+                out[f"{s}_joint_{i}.pos"] = float(vec[base + i - 1])
+            out[f"{s}_gripper.pos"] = float(vec[GRIPPER_IDX[s]])
+        return out
+
+    def _set_goal(self, short: dict) -> None:
+        with self._goal_lock:
+            base = self._goal if self._goal is not None else self._last_short_full
+            self._goal = self._dict_to_vec(short, fallback=base)
+
+    def _seed_current(self) -> None:
+        """Seed the streamer setpoint from the robot's actual present pose (ramp start)."""
+        try:
+            with self._io_lock:
+                obs = self._robot.get_observation()
+            if self._has_full_arms(obs):
+                self._stream_current = self._dict_to_vec(obs)
+                self._stream_vel = np.zeros(16)
+                if self._home is None:
+                    self._home = self._stream_current.copy()
+                logger.info("Streamer seeded from present robot pose")
+                return
+        except Exception as e:  # noqa: BLE001
+            logger.warning("Streamer seed failed (%s); will seed from first goal", e)
+        self._stream_current = None
+
+    def _stream_loop(self) -> None:
+        dt = 1.0 / self._stream_hz
+        pos_eps = 0.05  # deg: below this distance to goal we consider the axis settled
+        vel_eps = 0.5  # deg/s: below this speed we consider motion stopped
+        while not self._stream_stop.is_set():
+            t0 = time.perf_counter()
+            with self._goal_lock:
+                goal = None if self._goal is None else self._goal.copy()
+            if goal is None:
+                time.sleep(dt)
+                continue
+            if self._stream_current is None:
+                self._stream_current = goal.copy()
+                self._stream_vel = np.zeros(16)
+            self._stream_current, self._stream_vel = smooth_damp(
+                self._stream_current, goal, self._stream_vel, self._smooth_time, dt, self._max_speed
+            )
+            # Damiao MIT mode holds the last command, so when we've converged on the
+            # goal and stopped moving we skip the CAN write entirely. This keeps the
+            # bus from saturating (Errno 105) during the pauses when the policy goal
+            # is constant, and frees the _io_lock so the main loop's reads stay fast.
+            settled = (
+                float(np.max(np.abs(self._stream_current - goal))) < pos_eps
+                and float(np.max(np.abs(self._stream_vel))) < vel_eps
+            )
+            if not settled:
+                act = self._vec_to_action(self._stream_current)
+                try:
+                    with self._io_lock:
+                        self._robot.send_action(act)
+                except Exception as e:  # noqa: BLE001
+                    logger.warning("Streamer send_action failed: %s", e)
+            sleep_t = dt - (time.perf_counter() - t0)
+            if sleep_t > 0:
+                time.sleep(sleep_t)
+
+    # -- lifecycle (start/stop the streamer around the real connect/disconnect) ---
+    def connect(self, *args, **kwargs):
+        result = self._robot.connect(*args, **kwargs)
+        if self._stream and self._stream_thread is None:
+            self._seed_current()
+            self._stream_stop.clear()
+            self._stream_thread = threading.Thread(
+                target=self._stream_loop, name="RetargetStreamer", daemon=True
+            )
+            self._stream_thread.start()
+            logger.info("Smoothing streamer thread started")
+        return result
+
+    def _drain_home(self, timeout_s: float = 6.0, tol_deg: float = 0.7) -> None:
+        """Command the captured home pose and keep the streamer running until it
+        actually converges (or times out), so the arms fully reach home before we
+        cut the streamer thread. Only meaningful when streaming is enabled."""
+        if self._home is None or self._stream_thread is None:
+            return
+        logger.info("Returning arms to home pose (draining streamer)...")
+        with self._goal_lock:
+            self._goal = self._home.copy()
+        t_start = time.perf_counter()
+        while time.perf_counter() - t_start < timeout_s:
+            cur = self._stream_current
+            if cur is not None and float(np.max(np.abs(cur - self._home))) < tol_deg:
+                logger.info("Home pose reached")
+                return
+            time.sleep(0.05)
+        logger.warning("Home drain timed out after %.1fs; stopping streamer anyway", timeout_s)
+
+    def disconnect(self, *args, **kwargs):
+        if self._stream_thread is not None:
+            try:
+                self._drain_home()
+            except Exception as e:  # noqa: BLE001
+                logger.warning("Home drain failed: %s", e)
+            self._stream_stop.set()
+            self._stream_thread.join(timeout=2.0)
+            self._stream_thread = None
+            logger.info("Smoothing streamer thread stopped")
+        return self._robot.disconnect(*args, **kwargs)
+
+
+def _wrap_factory(orig):
+    obs_iters = int(os.environ.get("RETARGET_ITERS", "25"))
+    act_iters = int(os.environ.get("RETARGET_ITERS", "25"))
+    iters0 = int(os.environ.get("RETARGET_ITERS0", "80"))
+    do_obs = os.environ.get("RETARGET_OBS", "1") != "0"
+    do_act = os.environ.get("RETARGET_ACT", "1") != "0"
+    null_gain = float(os.environ.get("RETARGET_NULL_GAIN", "0.3"))
+    stream = os.environ.get("STREAM", "0") != "0"
+    stream_hz = float(os.environ.get("STREAM_HZ", "40"))
+    smooth_time = float(os.environ.get("STREAM_SMOOTH_TIME", "0.10"))
+    max_speed = float(os.environ.get("STREAM_MAX_SPEED", "150"))
+
+    def factory(cfg):
+        real = orig(cfg)
+        logger.info("Wrapping %s with RetargetRobot (MuJoCo long<->short EE retarget)", type(real).__name__)
+        return RetargetRobot(
+            real,
+            obs_iters=obs_iters,
+            act_iters=act_iters,
+            iters0=iters0,
+            retarget_obs=do_obs,
+            retarget_act=do_act,
+            null_gain=null_gain,
+            stream=stream,
+            stream_hz=stream_hz,
+            smooth_time=smooth_time,
+            max_speed=max_speed,
+        )
+
+    return factory
+
+
+def _patch_rtc_realtime():
+    """Make RTC re-anchor new chunks on the *actual* number of actions consumed
+    during inference instead of the fps-based estimate.
+
+    The stock ActionQueue computes ``real_delay = ceil(latency * fps)`` and discards
+    that many actions from every new chunk. When the control loop runs below --fps
+    (e.g. 20 Hz while fps=30), real_delay (18) overshoots the actions that were truly
+    consumed (indexes_diff=12), so it skips ~6 extra actions per inference and the
+    trajectory plays faster than real time. ``indexes_diff = last_index - idx_before``
+    is the ground-truth count of what the robot actually executed during inference, so
+    skipping exactly that many gives real-time playback independent of the loop rate.
+    """
+    try:
+        from lerobot.policies.rtc.action_queue import ActionQueue
+    except Exception as e:  # noqa: BLE001
+        logger.warning("Could not patch RTC action queue for real-time playback: %s", e)
+        return
+
+    def _resolve(self, real_delay, action_index_before_inference=None):
+        if action_index_before_inference is not None:
+            return max(0, self.last_index - action_index_before_inference)
+        return max(0, real_delay)
+
+    ActionQueue._check_and_resolve_delays = _resolve
+    logger.info("Patched RTC ActionQueue: re-anchor on real consumed index (no over-skip)")
+
+
+def main():
+    import lerobot.rollout.context as context
+
+    context.make_robot_from_config = _wrap_factory(context.make_robot_from_config)
+    _patch_rtc_realtime()
+
+    from lerobot.scripts.lerobot_rollout import main as rollout_main
+
+    rollout_main()
+
+
+if __name__ == "__main__":
+    main()

pyproject.toml

@@ -25,7 +25,7 @@ discord = "https://discord.gg/s3KuuzsPFb"
 
 [project]
 name = "lerobot"
-version = "0.6.2"
+version = "0.6.1"
 description = "🤗 LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch"
 dynamic = ["readme"]
 license = { text = "Apache-2.0" }
@@ -346,7 +346,6 @@ lerobot-record="lerobot.scripts.lerobot_record:main"
 lerobot-replay="lerobot.scripts.lerobot_replay:main"
 lerobot-setup-motors="lerobot.scripts.lerobot_setup_motors:main"
 lerobot-teleoperate="lerobot.scripts.lerobot_teleoperate:main"
-lerobot-convert-dcp="lerobot.scripts.lerobot_convert_dcp:main"
 lerobot-eval="lerobot.scripts.lerobot_eval:main"
 lerobot-train="lerobot.scripts.lerobot_train:main"
 lerobot-train-tokenizer="lerobot.scripts.lerobot_train_tokenizer:main"
@@ -401,63 +400,19 @@ exclude = ["tests/artifacts/**/*.safetensors", "*_pb2.py", "*_pb2_grpc.py"]
 # N: pep8-naming
 # TODO: Uncomment rules when ready to use
 select = [
-    "E", "W", "F", "I", "B", "C4", "T20", "N", "UP", "SIM", "D" #, "A", "S", "RUF"
+    "E", "W", "F", "I", "B", "C4", "T20", "N", "UP", "SIM" #, "A", "S", "D", "RUF"
 ]
 ignore = [
     "E501", # Line too long
     "T201", # Print statement found
     "T203", # Pprint statement found
     "B008", # Perform function call in argument defaults
-    # D100/D104: module- and package-level docstrings. The API reference is generated from class and
-    # function docstrings; a banner at the top of every file and every __init__.py would not appear on any
-    # rendered page. Coverage of the things that do get rendered is enforced by interrogate instead.
-    "D100",
-    "D104",
 ]
 
 [tool.ruff.lint.per-file-ignores]
-"__init__.py" = ["F401", "F403", "E402", "D104"]
+"__init__.py" = ["F401", "F403", "E402"]
 # E402: conditional-import guards (TYPE_CHECKING / is_package_available) must precede the imports they protect
 "src/lerobot/scripts/convert_dataset_v21_to_v30.py" = ["E402"]
-
-# D (pydocstyle) is enabled globally, but only holds for code that has been converted to the docstring
-# standard in docs/source/writing_docstrings.mdx. Every module below is still on the old style; each entry
-# is deleted as that module is converted, and this block can be removed once it is empty.
-#
-# Not part of the API reference and not planned for conversion: tests, examples, benchmarks, templates,
-# CI helper scripts and the packaging shim.
-"tests/**" = ["D"]
-"examples/**" = ["D"]
-"benchmarks/**" = ["D"]
-"scripts/**" = ["D"]
-"setup.py" = ["D"]
-"src/lerobot/templates/**" = ["D"]
-# Vendored from transformers; keeps its upstream docstring style so syncs stay clean.
-"src/lerobot/policies/molmoact2/molmoact2_hf_model/**" = ["D"]
-# Awaiting conversion, one PR per module.
-"src/lerobot/annotations/**" = ["D"]
-"src/lerobot/async_inference/**" = ["D"]
-"src/lerobot/cameras/**" = ["D"]
-"src/lerobot/common/**" = ["D"]
-"src/lerobot/configs/**" = ["D"]
-"src/lerobot/data_processing/**" = ["D"]
-"src/lerobot/distributed/**" = ["D"]
-"src/lerobot/envs/**" = ["D"]
-"src/lerobot/jobs/**" = ["D"]
-"src/lerobot/model/**" = ["D"]
-"src/lerobot/motors/**" = ["D"]
-"src/lerobot/optim/**" = ["D"]
-"src/lerobot/policies/**" = ["D"]
-"src/lerobot/processor/**" = ["D"]
-"src/lerobot/rewards/**" = ["D"]
-"src/lerobot/rl/**" = ["D"]
-"src/lerobot/rollout/**" = ["D"]
-"src/lerobot/scripts/**" = ["D"]
-"src/lerobot/teleoperators/**" = ["D"]
-"src/lerobot/transforms/**" = ["D"]
-"src/lerobot/transport/**" = ["D"]
-"src/lerobot/utils/**" = ["D"]
-"src/lerobot/lerobot_types.py" = ["D"]
 [tool.ruff.lint.isort]
 combine-as-imports = true
 known-first-party = ["lerobot"]
@@ -501,34 +456,25 @@ default.extend-ignore-identifiers-re = [
     "seperated_timestep",
 ]
 
-# Docstring coverage gate. `fail-under` is a RATCHET, not a target: it is set just below the currently
-# measured coverage so it passes today, and is raised in the same PR that documents a module. Never set it
-# to a value that fails on main. The destination is 100; see docs/source/writing_docstrings.mdx.
-[tool.interrogate]
-ignore-init-module = true
-ignore-init-method = true
-ignore-nested-functions = false
-ignore-magic = false
-ignore-semiprivate = false
-ignore-private = false
-ignore-property-decorators = false
-ignore-module = false
-ignore-setters = false
-fail-under = 55
-output-format = "term-missing"
-color = true
-paths = ["src/lerobot"]
-exclude = ["src/lerobot/policies/molmoact2/molmoact2_hf_model"]
+# TODO: Uncomment when ready to use
+# [tool.interrogate]
+# ignore-init-module = true
+# ignore-init-method = true
+# ignore-nested-functions = false
+# ignore-magic = false
+# ignore-semiprivate = false
+# ignore-private = false
+# ignore-property-decorators = false
+# ignore-module = false
+# ignore-setters = false
+# fail-under = 80
+# output-format = "term-missing"
+# color = true
+# paths = ["src/lerobot"]
 
 # TODO: Enable mypy gradually module by module across multiple PRs
 # Uncomment [tool.mypy] first, then uncomment individual module overrides as they get proper type annotations
 
-[tool.pytest.ini_options]
-markers = [
-    "multigpu: distributed tests needing 2-4 GPUs (CI: docker_publish.yml lane)",
-    "multigpu_heavy: 8-GPU sweeps and soak tests; never run in CI",
-]
-
 [tool.mypy]
 python_version = "3.12"
 ignore_missing_imports = true
@@ -575,15 +521,6 @@ disallow_untyped_defs = true
 disallow_incomplete_defs = true
 check_untyped_defs = true
 
-[[tool.mypy.overrides]]
-module = "lerobot.distributed.*"
-ignore_errors = false
-
-# extra strictness for the distributed engine
-disallow_untyped_defs = true
-disallow_incomplete_defs = true
-check_untyped_defs = true
-
 [[tool.mypy.overrides]]
 module = "lerobot.optim.*"
 ignore_errors = false

src/lerobot/__init__.py

@@ -14,7 +14,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""LeRobot -- PyTorch library for real-world robotics.
+"""
+LeRobot -- PyTorch library for real-world robotics.
 
 Provides datasets, pretrained policies, and tools for training, evaluation,
 data collection, and robot control. Integrates with Hugging Face Hub for

src/lerobot/__version__.py

@@ -13,7 +13,7 @@
 # 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.
-"""To enable `lerobot.__version__`."""
+"""To enable `lerobot.__version__`"""
 
 from importlib.metadata import PackageNotFoundError, version
 

src/lerobot/cameras/camera.py

@@ -33,10 +33,10 @@ class Camera(abc.ABC):
     - Connection/disconnection
     - Frame capture (sync/async/latest)
 
-    **Attributes**:
-        - **fps** (`int | None`) -- Configured frames per second.
-        - **width** (`int | None`) -- Frame width in pixels.
-        - **height** (`int | None`) -- Frame height in pixels.
+    Attributes:
+        fps (int | None): Configured frames per second
+        width (int | None): Frame width in pixels
+        height (int | None): Frame height in pixels
     """
 
     def __init__(self, config: CameraConfig):

src/lerobot/cameras/opencv/configuration_opencv.py

@@ -40,20 +40,17 @@ class OpenCVCameraConfig(CameraConfig):
     OpenCVCameraConfig(0, 30, 1280, 720, fourcc="YUYV")     # With YUYV format
     ```
 
-    **Attributes**:
-        - **index_or_path** (`int | Path`) -- Either an integer representing the camera device index, or a
-          Path object pointing to a video file.
-        - **fps** -- Requested frames per second for the color stream.
-        - **width** -- Requested frame width in pixels for the color stream.
-        - **height** -- Requested frame height in pixels for the color stream.
-        - **color_mode** (`ColorMode`) -- Color mode for image output (RGB or BGR). Defaults to RGB.
-        - **rotation** (`Cv2Rotation`) -- Image rotation setting (0°, 90°, 180°, or 270°). Defaults to no
-          rotation.
-        - **warmup_s** (`int`) -- Time reading frames before returning from connect (in seconds)
-        - **fourcc** (`str | None`) -- FOURCC code for video format (e.g., "MJPG", "YUYV", "I420"). Defaults
-          to None (auto-detect).
-        - **backend** (`Cv2Backends`) -- OpenCV backend identifier
-          (https://docs.opencv.org/3.4/d4/d15/group__videoio__flags__base.html). Defaults to ANY.
+    Attributes:
+        index_or_path: Either an integer representing the camera device index,
+                      or a Path object pointing to a video file.
+        fps: Requested frames per second for the color stream.
+        width: Requested frame width in pixels for the color stream.
+        height: Requested frame height in pixels for the color stream.
+        color_mode: Color mode for image output (RGB or BGR). Defaults to RGB.
+        rotation: Image rotation setting (0°, 90°, 180°, or 270°). Defaults to no rotation.
+        warmup_s: Time reading frames before returning from connect (in seconds)
+        fourcc: FOURCC code for video format (e.g., "MJPG", "YUYV", "I420"). Defaults to None (auto-detect).
+        backend: OpenCV backend identifier (https://docs.opencv.org/3.4/d4/d15/group__videoio__flags__base.html). Defaults to ANY.
 
     Note:
         - Only 3-channel color output (RGB/BGR) is currently supported.

src/lerobot/cameras/reachy2_camera/configuration_reachy2_camera.py

@@ -43,16 +43,16 @@ class Reachy2CameraConfig(CameraConfig):
     )  # Left teleop camera, 640x480 @ 30FPS
     ```
 
-    **Attributes**:
-        - **name** (`str`) -- Name of the camera device. Can be "teleop" or "depth".
-        - **image_type** (`str`) -- Type of image stream. For "teleop" camera, can be "left" or "right". For
-          "depth" camera, can be "rgb" or "depth". (depth is not supported yet)
-        - **fps** -- Requested frames per second for the color stream. Not configurable for Reachy 2 cameras.
-        - **width** -- Requested frame width in pixels for the color stream.
-        - **height** -- Requested frame height in pixels for the color stream.
-        - **color_mode** (`ColorMode`) -- Color mode for image output (RGB or BGR). Defaults to RGB.
-        - **ip_address** (`str | None`) -- IP address of the robot. Defaults to "localhost".
-        - **port** (`int`) -- Port number for the camera server. Defaults to 50065.
+    Attributes:
+        name: Name of the camera device. Can be "teleop" or "depth".
+        image_type: Type of image stream. For "teleop" camera, can be "left" or "right".
+                    For "depth" camera, can be "rgb" or "depth". (depth is not supported yet)
+        fps: Requested frames per second for the color stream. Not configurable for Reachy 2 cameras.
+        width: Requested frame width in pixels for the color stream.
+        height: Requested frame height in pixels for the color stream.
+        color_mode: Color mode for image output (RGB or BGR). Defaults to RGB.
+        ip_address: IP address of the robot. Defaults to "localhost".
+        port: Port number for the camera server. Defaults to 50065.
 
     Note:
         - Only 3-channel color output (RGB/BGR) is currently supported.

src/lerobot/cameras/realsense/configuration_realsense.py

@@ -36,28 +36,27 @@ class RealSenseCameraConfig(CameraConfig):
     RealSenseCameraConfig("0123456789", 30, 640, 480, rotation=Cv2Rotation.ROTATE_90)  # With 90° rotation
     ```
 
-    **Attributes**:
-        - **fps** -- Requested frames per second for the color stream.
-        - **width** -- Requested frame width in pixels for the color stream.
-        - **height** -- Requested frame height in pixels for the color stream.
-        - **serial_number_or_name** (`str`) -- Unique serial number or human-readable name to identify the
-          camera.
-        - **color_mode** (`ColorMode`) -- Color mode for image output (RGB or BGR). Defaults to RGB.
-        - **use_rgb** (`bool`) -- Whether to enable the color stream. Defaults to True.
-        - **use_depth** (`bool`) -- Whether to enable depth stream. Defaults to False.
-        - **rotation** (`Cv2Rotation`) -- Image rotation setting (0°, 90°, 180°, or 270°). Defaults to no
-          rotation.
-        - **warmup_s** (`int`) -- Time reading frames before returning from connect (in seconds)
-        - **exposure** (`int | None`) -- Manual exposure value for the color sensor. When set, auto-exposure
-          is disabled and this fixed value is used. Valid ranges are camera-model specific and reported if the
-          value is rejected. Defaults to None (leave unchanged).
-        - **gain** (`int | None`) -- Manual gain value for the color sensor. When set, auto-exposure is
-          disabled and this fixed gain is used, which also freezes exposure at its current value when no
-          exposure is configured. Valid ranges are camera-model specific and reported if the value is
-          rejected. Defaults to None (leave unchanged).
-        - **white_balance** (`int | None`) -- Manual white balance value for the color sensor. When set, auto
-          white balance is disabled and this fixed value is used. Valid ranges are camera-model specific and
-          reported if the value is rejected. Defaults to None (leave unchanged).
+    Attributes:
+        fps: Requested frames per second for the color stream.
+        width: Requested frame width in pixels for the color stream.
+        height: Requested frame height in pixels for the color stream.
+        serial_number_or_name: Unique serial number or human-readable name to identify the camera.
+        color_mode: Color mode for image output (RGB or BGR). Defaults to RGB.
+        use_rgb: Whether to enable the color stream. Defaults to True.
+        use_depth: Whether to enable depth stream. Defaults to False.
+        rotation: Image rotation setting (0°, 90°, 180°, or 270°). Defaults to no rotation.
+        warmup_s: Time reading frames before returning from connect (in seconds)
+        exposure: Manual exposure value for the color sensor. When set, auto-exposure is
+            disabled and this fixed value is used. Valid ranges are camera-model specific
+            and reported if the value is rejected. Defaults to None (leave unchanged).
+        gain: Manual gain value for the color sensor. When set, auto-exposure is disabled
+            and this fixed gain is used, which also freezes exposure at its current value
+            when no exposure is configured. Valid ranges are camera-model specific and
+            reported if the value is rejected. Defaults to None (leave unchanged).
+        white_balance: Manual white balance value for the color sensor. When set, auto
+            white balance is disabled and this fixed value is used. Valid ranges are
+            camera-model specific and reported if the value is rejected. Defaults to None
+            (leave unchanged).
 
     Note:
         - Either name or serial_number must be specified.

src/lerobot/common/train_utils.py

@@ -13,41 +13,16 @@
 # 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.
-"""Training-output persistence: checkpoints, two-phase resume, and hub publishing.
-
-Rank discipline: every function here that can
-contain a collective is documented as such and must run on ALL ranks; rank-0-only file writes
-sit under one grouped ``is_main_process()`` gate per contiguous region, placed below all
-collectives. The leaf save/load helpers carry no rank gates of their own — the exception is
-``PreTrainedPolicy._save_pretrained``, whose gate is internal because its collective gather and
-its writes live in the same method.
-"""
-
-import logging
-from importlib.resources import files
 from pathlib import Path
-from tempfile import TemporaryDirectory
-from typing import TYPE_CHECKING, Any
 
-import torch.distributed as dist
-from huggingface_hub import HfApi, ModelCard, ModelCardData, snapshot_download
+from huggingface_hub import HfApi, snapshot_download
 from torch.optim import Optimizer
 from torch.optim.lr_scheduler import LRScheduler
 
-from lerobot.__version__ import __version__
-from lerobot.configs.policies import PreTrainedConfig
-from lerobot.configs.rewards import RewardModelConfig
 from lerobot.configs.train import TrainPipelineConfig
-from lerobot.distributed.checkpoint import (
-    is_sharded_module,
-    load_sharded_model,
-    load_sharded_optimizer,
-    save_sharded_model,
-    save_sharded_optimizer,
-)
-from lerobot.distributed.utils import is_main_process
 from lerobot.optim import (
     load_optimizer_state,
+    load_optimizer_state_dict,
     load_scheduler_state,
     save_optimizer_state,
     save_scheduler_state,
@@ -65,39 +40,14 @@ from lerobot.utils.hub import find_latest_hub_checkpoint
 from lerobot.utils.io_utils import load_json, write_json
 from lerobot.utils.random_utils import load_rng_state, save_rng_state
 
-if TYPE_CHECKING:
-    from accelerate import Accelerator
-
-    from lerobot.datasets.dataset_metadata import LeRobotDatasetMetadata
-    from lerobot.rewards.pretrained import PreTrainedRewardModel
-
 
 def get_step_identifier(step: int, total_steps: int) -> str:
-    """Format a step number as the zero-padded identifier used for checkpoint directory names.
-
-    Args:
-        step (int): The training step to format.
-        total_steps (int): The total number of training steps; sets the padding width
-            (minimum 6 digits).
-
-    Returns:
-        str: The zero-padded step identifier, e.g. `"005000"`.
-    """
     num_digits = max(6, len(str(total_steps)))
     return f"{step:0{num_digits}d}"
 
 
 def get_step_checkpoint_dir(output_dir: Path, total_steps: int, step: int) -> Path:
-    """Returns the checkpoint sub-directory corresponding to the step number.
-
-    Args:
-        output_dir (Path): The training run's output directory.
-        total_steps (int): The total number of training steps; sets the identifier padding.
-        step (int): The training step of the checkpoint.
-
-    Returns:
-        Path: The checkpoint step directory, `output_dir/checkpoints/<step-identifier>`.
-    """
+    """Returns the checkpoint sub-directory corresponding to the step number."""
     step_identifier = get_step_identifier(step, total_steps)
     return output_dir / CHECKPOINTS_DIR / step_identifier
 
@@ -113,15 +63,37 @@ def should_save_checkpoint(step: int, save_freq: int, total_steps: int) -> bool:
     return (save_freq > 0 and step % save_freq == 0) or step == total_steps
 
 
-def update_last_checkpoint(checkpoint_dir: Path) -> None:
-    """Point the `last` symlink in the checkpoints directory at the given checkpoint.
+def save_training_step(
+    step: int, save_dir: Path, num_processes: int | None = None, batch_size: int | None = None
+) -> None:
+    state: dict = {"step": step}
+    # num_processes and batch_size are recorded so a resumed run can detect a changed world size or
+    # batch size: the sampler's resume offset is computed from the (num_processes, batch_size) that
+    # produced `step`, since both scale how many sampler positions a step consumes (see
+    # compute_sampler_state).
+    if num_processes is not None:
+        state["num_processes"] = num_processes
+    if batch_size is not None:
+        state["batch_size"] = batch_size
+    write_json(state, save_dir / TRAINING_STEP)
 
-    Any existing `last` symlink is replaced. The link target is relative to the checkpoints
-    directory, so the tree stays valid when the run directory is moved.
 
-    Args:
-        checkpoint_dir (Path): The checkpoint step directory the `last` link should target.
-    """
+def load_training_step(save_dir: Path) -> int:
+    training_step = load_json(save_dir / TRAINING_STEP)
+    return training_step["step"]
+
+
+def load_training_num_processes(checkpoint_dir: Path) -> int | None:
+    """World size recorded at checkpoint time, or None for checkpoints written before it was stored."""
+    return load_json(checkpoint_dir / TRAINING_STATE_DIR / TRAINING_STEP).get("num_processes")
+
+
+def load_training_batch_size(checkpoint_dir: Path) -> int | None:
+    """Per-process batch size recorded at checkpoint time, or None for older checkpoints."""
+    return load_json(checkpoint_dir / TRAINING_STATE_DIR / TRAINING_STEP).get("batch_size")
+
+
+def update_last_checkpoint(checkpoint_dir: Path) -> Path:
     last_checkpoint_dir = checkpoint_dir.parent / LAST_CHECKPOINT_LINK
     if last_checkpoint_dir.is_symlink():
         last_checkpoint_dir.unlink()
@@ -129,68 +101,6 @@ def update_last_checkpoint(checkpoint_dir: Path) -> None:
     last_checkpoint_dir.symlink_to(relative_target)
 
 
-# ---------------------------------------------------------------------------------------------
-# training_step.json
-# ---------------------------------------------------------------------------------------------
-
-
-def save_training_metadata(step: int, save_dir: Path, cfg: TrainPipelineConfig) -> None:
-    """Record the step counter plus everything a resume needs to reason about topology changes.
-
-    `step` counts loop iterations (= micro-batches), so
-    the sampler resume offset is `step x batch_size x dp_world_size` with no grad-accum factor.
-    `grad_accum_steps` and the parallelism snapshot are recorded so a resume can warn precisely
-    when the optimizer-update cadence or the sharding topology changed.
-
-    Args:
-        step (int): The training step (micro-batch counter) to record.
-        save_dir (Path): The `training_state/` directory to write `training_step.json` into.
-        cfg (TrainPipelineConfig): The training config whose batch size, gradient-accumulation,
-            and parallelism settings are snapshotted alongside the step.
-    """
-    state: dict[str, Any] = {
-        "step": step,
-        "dp_world_size": cfg.parallelism.dp_world_size,
-        "batch_size": cfg.batch_size,
-        "grad_accum_steps": cfg.accelerator.gradient_accumulation.steps,
-        "parallelism": {
-            "dp_replicate": cfg.parallelism.dp_replicate,
-            "dp_shard": cfg.parallelism.dp_shard,
-            "ring_degree": cfg.parallelism.context_parallel.ring_degree,
-            "ulysses_degree": cfg.parallelism.context_parallel.ulysses_degree,
-        },
-    }
-    write_json(state, save_dir / TRAINING_STEP)
-
-
-def load_training_metadata(training_state_dir: Path) -> dict[str, Any]:
-    """Read everything `save_training_metadata` recorded, in a single pass.
-
-    Every key is always present: fields a checkpoint predates come back as None, so a caller
-    reading `metadata["batch_size"]` gets a KeyError on a typo rather than a silent None.
-
-    Args:
-        training_state_dir (Path): The checkpoint's `training_state/` directory.
-
-    Returns:
-        dict[str, Any]: `step` plus the `dp_world_size`, `batch_size`, `grad_accum_steps` and
-            `parallelism` snapshot recorded alongside it (None where not recorded).
-    """
-    state = load_json(training_state_dir / TRAINING_STEP)
-    return {
-        "step": int(state["step"]),
-        "dp_world_size": state.get("dp_world_size", state.get("num_processes")),
-        "batch_size": state.get("batch_size"),
-        "grad_accum_steps": state.get("grad_accum_steps"),
-        "parallelism": state.get("parallelism"),
-    }
-
-
-# ---------------------------------------------------------------------------------------------
-# Checkpoint save
-# ---------------------------------------------------------------------------------------------
-
-
 def save_checkpoint(
     checkpoint_dir: Path,
     step: int,
@@ -200,301 +110,192 @@ def save_checkpoint(
     scheduler: LRScheduler | None = None,
     preprocessor: PolicyProcessorPipeline | None = None,
     postprocessor: PolicyProcessorPipeline | None = None,
-    accelerator: "Accelerator | None" = None,
+    num_processes: int | None = None,
+    batch_size: int | None = None,
+    model_state_dict: dict | None = None,
+    optim_state_dict: dict | None = None,
 ) -> None:
     """This function creates the following directory structure:
 
     005000/  #  training step at checkpoint
     ├── pretrained_model/
     │   ├── config.json  # policy config
-    │   ├── model.safetensors  # policy weights (checkpoint_format ∈ {safetensors, safetensors_dcp}, or any non-sharded run)
-    │   ├── pytorch_model_fsdp_0/  # DCP model shards (checkpoint_format ∈ {dcp, safetensors_dcp})
+    │   ├── model.safetensors  # policy weights
     │   ├── train_config.json  # train config
-    │   ├── policy_preprocessor.json  # preprocessor config (if preprocessor provided)
-    │   ├── policy_preprocessor_step_*.safetensors  # state of the stateful preprocessor steps
-    │   ├── policy_postprocessor.json  # postprocessor config (if postprocessor provided)
-    │   └── policy_postprocessor_step_*.safetensors  # state of the stateful postprocessor steps
+    │   ├── processor.json  # processor config (if preprocessor provided)
+    │   └── step_*.safetensors  # processor state files (if any)
     └── training_state/
-        ├── optimizer_param_groups.json  # optimizer param groups (non-sharded runs)
-        ├── optimizer_state.safetensors  # optimizer state (non-sharded runs)
-        ├── optimizer_0/  # DCP optimizer shards (sharded runs)
+        ├── optimizer_param_groups.json  #  optimizer param groups
+        ├── optimizer_state.safetensors  # optimizer state
         ├── rng_state.safetensors  # rng states
-        ├── scheduler_state.json  # scheduler state (if scheduler provided)
-        └── training_step.json  # training step + dp_world_size/batch_size/grad_accum + topology
-
-    Collective: MUST be called on every rank. Rank-0-only writes are gated internally, so the
-    call site needs no rank branches.
+        ├── scheduler_state.json  # scheduler state
+        └── training_step.json  # training step
 
     Args:
-        checkpoint_dir (Path): The checkpoint step directory to write (e.g. `.../checkpoints/005000`).
-        step (int): The training step at that checkpoint.
         cfg (TrainPipelineConfig): The training config used for this run.
+        step (int): The training step at that checkpoint.
         policy (PreTrainedPolicy): The policy to save.
-        optimizer (Optimizer): The optimizer to save the state from.
+        optimizer (Optimizer | None, optional): The optimizer to save the state from. Defaults to None.
         scheduler (LRScheduler | None, optional): The scheduler to save the state from. Defaults to None.
-        preprocessor (PolicyProcessorPipeline | None, optional): The preprocessor/pipeline to save.
+        preprocessor: The preprocessor/pipeline to save. Defaults to None.
+        postprocessor: The postprocessor/pipeline to save. Defaults to None.
+        num_processes (int | None, optional): Distributed world size to record for sample-exact
+            resume. Defaults to None (not recorded).
+        batch_size (int | None, optional): Per-process batch size to record for sample-exact
+            resume. Defaults to None (not recorded).
+        model_state_dict: Pre-gathered full (unsharded) model state dict. Required under FSDP,
+            where `policy.state_dict()` would return sharded tensors; the caller gathers it via a
+            cross-rank collective and passes it here so rank 0 can write it directly. It holds
+            FSDP's fp32 master weights and is saved as-is (the loader casts to the policy dtype on
+            read). When None (DDP / single-GPU), the model is saved the normal way. Defaults to None.
+        optim_state_dict: Pre-gathered full (unsharded) optimizer state dict. Required under FSDP
+            (gathered alongside `model_state_dict` via `gather_fsdp_state_dicts`); saved in the same
+            safetensors format as the single-GPU path. When None, `optimizer.state_dict()` is used.
             Defaults to None.
-        postprocessor (PolicyProcessorPipeline | None, optional): The postprocessor/pipeline to save.
-            Defaults to None.
-        accelerator (Accelerator | None, optional): The accelerator the policy was prepared with;
-            used to unwrap the model and required on sharded runs, where it owns the DCP save
-            channels. Defaults to None (plain single-process saves).
     """
     pretrained_dir = checkpoint_dir / PRETRAINED_MODEL_DIR
-    fmt = cfg.checkpoint_format
-    policy_to_save = accelerator.unwrap_model(policy) if accelerator is not None else policy
-    sharded = is_sharded_module(policy_to_save)
-
-    # -- model artifact(s): the two collective-capable calls ----------------------------------
+    policy.save_pretrained(pretrained_dir, state_dict=model_state_dict)
+    cfg.save_pretrained(pretrained_dir)
     if cfg.peft is not None:
-        # PeftModel.save_pretrained is an external API with no internal rank gate, and the
-        # adapters are replicated (PEFT x sharded is rejected at validation): main rank writes.
-        if is_main_process():
-            policy_to_save.save_pretrained(pretrained_dir)
-    elif fmt.wants_safetensors or not sharded:
-        # Collective when sharded (full gather); writes happen on the main process only in all
-        # multi-rank layouts (the gate lives inside _save_pretrained, next to its collective gather).
-        policy_to_save.save_pretrained(pretrained_dir)
-    if fmt.wants_dcp and sharded:
-        save_sharded_model(accelerator, policy_to_save, pretrained_dir)
-
-    # -- sidecar configs: ONE gate for the whole contiguous rank-0-only region ----------------
-    if is_main_process():
-        if fmt.wants_dcp and not fmt.wants_safetensors:
-            # save_pretrained did not run: keep the DCP-only checkpoint self-describing.
-            policy_to_save.config.save_pretrained(pretrained_dir)
-        cfg.save_pretrained(pretrained_dir)
-        if cfg.peft is not None:
-            # PEFT's save_pretrained writes only adapter weights + config; the policy config
-            # needed to reload the base model is written explicitly.
-            policy_to_save.config.save_pretrained(pretrained_dir)
-        if preprocessor is not None:
-            preprocessor.save_pretrained(pretrained_dir)
-        if postprocessor is not None:
-            postprocessor.save_pretrained(pretrained_dir)
-
+        # When using PEFT, policy.save_pretrained will only write the adapter weights + config, not the
+        # policy config which we need for loading the model. In this case we'll write it ourselves.
+        policy.config.save_pretrained(pretrained_dir)
+    if preprocessor is not None:
+        preprocessor.save_pretrained(pretrained_dir)
+    if postprocessor is not None:
+        postprocessor.save_pretrained(pretrained_dir)
     save_training_state(
-        checkpoint_dir, step, cfg, optimizer, scheduler, accelerator, sharded=sharded, model=policy_to_save
+        checkpoint_dir,
+        step,
+        optimizer,
+        scheduler,
+        num_processes=num_processes,
+        batch_size=batch_size,
+        optim_state_dict=optim_state_dict,
     )
-    if accelerator is not None:
-        accelerator.wait_for_everyone()
 
 
 def save_training_state(
     checkpoint_dir: Path,
-    step: int,
-    cfg: TrainPipelineConfig,
-    optimizer: Optimizer | dict[str, Optimizer] | None = None,
+    train_step: int,
+    optimizer: Optimizer | None = None,
     scheduler: LRScheduler | None = None,
-    accelerator: "Accelerator | None" = None,
-    *,
-    sharded: bool = False,
-    model: PreTrainedPolicy | None = None,
+    num_processes: int | None = None,
+    batch_size: int | None = None,
+    optim_state_dict: dict | None = None,
 ) -> None:
-    """Write training_state/. Collective under sharding: call on every rank.
+    """
+    Saves the training step, optimizer state, scheduler state, and rng state.
 
     Args:
-        checkpoint_dir (Path): The checkpoint step directory; `training_state/` is created inside it.
-        step (int): The training step at that checkpoint.
-        cfg (TrainPipelineConfig): The training config used for this run (its topology and
-            accumulation settings are recorded in `training_step.json`).
-        optimizer (Optimizer | dict[str, Optimizer] | None, optional): The optimizer(s) to save
-            the state from. Defaults to None.
-        scheduler (LRScheduler | None, optional): The scheduler to save the state from.
+        save_dir (Path): The directory to save artifacts to.
+        train_step (int): Current training step.
+        optimizer (Optimizer | None, optional): The optimizer from which to save the state_dict.
             Defaults to None.
-        accelerator (Accelerator | None, optional): Required when `sharded` is True — it owns
-            the DCP optimizer save channel. Defaults to None.
-        sharded (bool): The model's sharding state, computed once in `save_checkpoint` and
-            threaded here so the two sites cannot disagree. Defaults to False.
-        model (PreTrainedPolicy | None, optional): Required only for the sharded optimizer
-            channel: torch's optimizer DCP APIs are model-coupled (the state dict is keyed by
-            model FQNs), so accelerate's `save_fsdp_optimizer` needs the sharded module
-            alongside the optimizer. Defaults to None.
+        scheduler (LRScheduler | None, optional): The scheduler from which to save the state_dict.
+            Defaults to None.
+        num_processes (int | None, optional): Distributed world size to record. Defaults to None.
+        batch_size (int | None, optional): Per-process batch size to record. Defaults to None.
+        optim_state_dict: Pre-gathered full optimizer state dict (for FSDP). Saved instead of
+            `optimizer.state_dict()` when provided. Defaults to None.
     """
     save_dir = checkpoint_dir / TRAINING_STATE_DIR
-    # All ranks: the directory must exist before the DCP optimizer collective writes into it
-    # (exist_ok makes the concurrent mkdir race-free on shared filesystems).
     save_dir.mkdir(parents=True, exist_ok=True)
-
-    if optimizer is not None and sharded:
-        if accelerator is None or model is None:
-            raise ValueError("Saving a sharded optimizer state requires the accelerator and model.")
-        # Collective — all ranks write their DCP shards into optimizer_0/.
-        save_sharded_optimizer(accelerator, optimizer, model, save_dir)
-
-    if is_main_process():  # ONE grouped gate for the whole rank-0-only region
-        save_training_metadata(step, save_dir, cfg)
-        save_rng_state(save_dir)
-        if scheduler is not None:
-            save_scheduler_state(scheduler, save_dir)
-        if optimizer is not None and not sharded:
-            save_optimizer_state(optimizer, save_dir)
+    save_training_step(train_step, save_dir, num_processes=num_processes, batch_size=batch_size)
+    save_rng_state(save_dir)
+    if optimizer is not None:
+        save_optimizer_state(optimizer, save_dir, optim_state_dict=optim_state_dict)
+    if scheduler is not None:
+        save_scheduler_state(scheduler, save_dir)
 
 
-# ---------------------------------------------------------------------------------------------
-# Two-phase resume
-# ---------------------------------------------------------------------------------------------
-
-
-def resume_before_prepare(cfg: TrainPipelineConfig) -> int:
-    """Phase 1 — before `accelerator.prepare()`: restore RNG and return the step counter.
-
-    Pure loaders only. The sampler resume offset is *derived* from the returned step inside the
-    dataloader factory, and everything bound to sharded objects (model DCP shards, optimizer,
-    scheduler) loads in `resume_after_prepare`.
+def load_training_state(
+    checkpoint_dir: Path, optimizer: Optimizer, scheduler: LRScheduler | None, load_optimizer: bool = True
+) -> tuple[int, Optimizer, LRScheduler | None]:
+    """
+    Loads the training step, optimizer state, scheduler state, and rng state.
+    This is used to resume a training run.
 
     Args:
-        cfg (TrainPipelineConfig): The resumed training config; `cfg.checkpoint_path` locates
-            the checkpoint to restore from.
+        checkpoint_dir (Path): The checkpoint directory. Should contain a 'training_state' dir.
+        optimizer (Optimizer): The optimizer to load the state_dict to.
+        scheduler (LRScheduler | None): The scheduler to load the state_dict to (can be None).
+        load_optimizer (bool, optional): Whether to load the optimizer state from disk. Defaults to
+            True. Set to False under FSDP, where the sharded optimizer state must be loaded after
+            `accelerator.prepare()` via `load_fsdp_optimizer_state` (the optimizer is returned
+            untouched here).
+
+    Raises:
+        NotADirectoryError: If 'checkpoint_dir' doesn't contain a 'training_state' dir
 
     Returns:
-        int: The training step recorded in the checkpoint (micro-batch counter).
-
-    Raises:
-        NotADirectoryError: If the checkpoint has no `training_state/` directory.
-        ValueError: If the resumed topology crosses the sharded/non-sharded boundary relative
-            to the one recorded in the checkpoint.
+        tuple[int, Optimizer, LRScheduler | None]: training step, optimizer and scheduler with their
+            state_dict loaded.
     """
-    training_state_dir = cfg.checkpoint_path / TRAINING_STATE_DIR
+    training_state_dir = checkpoint_dir / TRAINING_STATE_DIR
     if not training_state_dir.is_dir():
         raise NotADirectoryError(training_state_dir)
-    metadata = load_training_metadata(training_state_dir)
-    _guard_resume_changes(cfg, metadata)
+
     load_rng_state(training_state_dir)
-    return metadata["step"]
-
-
-def _guard_resume_changes(cfg: TrainPipelineConfig, metadata: dict[str, Any]) -> None:
-    """Check the resumed run settings against the ones recorded in the checkpoint.
-
-    Two tiers, both driven by the checkpoint's recorded parallelism snapshot:
-
-    - **Hard error** when the resume crosses the sharded/non-sharded boundary in either
-      direction: the checkpoint's training-state artifacts only support resuming on the same
-      kind of topology (resharding works across sizes, not across kinds). Checkpoints without
-      a recorded snapshot skip this check.
-    - **One warning** naming every other recorded setting that differs — those changes are
-      legal (DCP reshards weights and optimizer state across topologies and the sampler offset
-      adapts), but a changed ``grad_accum_steps`` shifts the optimizer-update cadence, so the
-      resume says precisely what differs. The sampler-exactness warnings
-      (``dp_world_size``/``batch_size``) live with the sampler math in the dataloader factory.
-
-    Args:
-        cfg (TrainPipelineConfig): The resumed training config, compared against the settings
-            recorded in the checkpoint.
-        metadata (dict[str, Any]): The checkpoint's recorded training metadata, as returned by
-            `load_training_metadata`.
-
-    Raises:
-        ValueError: If the checkpoint records a sharded topology and the resumed run is
-            non-sharded, or vice versa.
-    """
-    snapshot = metadata["parallelism"]
-
-    if snapshot is not None:
-        recorded_sharded = (
-            snapshot.get("dp_shard", 1) != 1
-            or snapshot.get("ring_degree", 1) * snapshot.get("ulysses_degree", 1) > 1
-        )
-        if recorded_sharded != cfg.parallelism.is_sharded:
-            raise ValueError(
-                f"Cannot resume: the checkpoint was written with a "
-                f"{'sharded' if recorded_sharded else 'non-sharded'} topology "
-                f"(dp_replicate={snapshot.get('dp_replicate')}, dp_shard={snapshot.get('dp_shard')}) "
-                f"but this run is {'sharded' if cfg.parallelism.is_sharded else 'non-sharded'} "
-                f"(dp_replicate={cfg.parallelism.dp_replicate}, dp_shard={cfg.parallelism.dp_shard})."
-            )
-
-    recorded = {
-        "grad_accum_steps": (
-            metadata["grad_accum_steps"],
-            cfg.accelerator.gradient_accumulation.steps,
-        ),
-    }
-    if snapshot is not None:
-        recorded.update(
-            {
-                "dp_replicate": (snapshot.get("dp_replicate"), cfg.parallelism.dp_replicate),
-                "dp_shard": (snapshot.get("dp_shard"), cfg.parallelism.dp_shard),
-                "ring_degree": (
-                    snapshot.get("ring_degree"),
-                    cfg.parallelism.context_parallel.ring_degree,
-                ),
-                "ulysses_degree": (
-                    snapshot.get("ulysses_degree"),
-                    cfg.parallelism.context_parallel.ulysses_degree,
-                ),
-            }
-        )
-    changed = [f"{key}: {was} -> {now}" for key, (was, now) in recorded.items() if was not in (None, now)]
-    if changed and is_main_process():
-        logging.warning(
-            "Resuming with settings that differ from the checkpoint: " + "; ".join(changed) + ". "
-            "Topology changes reshard safely via DCP; a changed grad_accum_steps shifts the "
-            "optimizer-update cadence (the step counter keeps counting micro-batches)."
-        )
-
-
-def resume_after_prepare(
-    cfg: TrainPipelineConfig,
-    accelerator: "Accelerator",
-    policy: PreTrainedPolicy,
-    optimizer: Optimizer | dict[str, Optimizer],
-    scheduler: LRScheduler | None,
-) -> None:
-    """Phase 2 — after `accelerator.prepare()`: model (DCP) -> optimizer -> scheduler.
-
-    Collective under sharding: call on every rank. The model-weight source follows the
-    checkpoint's own recorded `checkpoint_format` (on resume, `cfg` was parsed from the
-    checkpoint's train_config.json): DCP-bearing formats load shards here into the prepared
-    model (whose construction skipped the safetensors load); the safetensors format was already
-    loaded by `from_pretrained` before sharding — no model step here.
-
-    Args:
-        cfg (TrainPipelineConfig): The resumed training config; `cfg.checkpoint_path` locates
-            the checkpoint and `cfg.checkpoint_format` selects the model-weight source.
-        accelerator (Accelerator): The accelerator the policy was prepared with; it unwraps the
-            model and owns the DCP load channels.
-        policy (PreTrainedPolicy): The prepared (possibly sharded) policy to load weights into.
-        optimizer (Optimizer | dict[str, Optimizer]): The prepared optimizer(s) to restore.
-        scheduler (LRScheduler | None): The scheduler to restore, or None if the run has none.
-
-    Raises:
-        FileNotFoundError: If the checkpoint format declares DCP model shards but the shard
-            directory is missing (e.g. it was pruned before upload).
-    """
-    checkpoint_dir = cfg.checkpoint_path
-    pretrained_dir = checkpoint_dir / PRETRAINED_MODEL_DIR
-    training_state_dir = checkpoint_dir / TRAINING_STATE_DIR
-    unwrapped = accelerator.unwrap_model(policy)
-    sharded = is_sharded_module(unwrapped)
-
-    if cfg.checkpoint_format.wants_dcp:
-        from accelerate.utils.constants import FSDP_MODEL_NAME
-
-        dcp_dir = pretrained_dir / f"{FSDP_MODEL_NAME}_0"
-        if not dcp_dir.is_dir():
-            raise FileNotFoundError(
-                f"checkpoint_format={cfg.checkpoint_format.value} declares DCP model shards, "
-                f"but {dcp_dir} is missing. If the shards were pruned, convert what remains "
-                "with `lerobot-convert-dcp` or resume from a safetensors checkpoint."
-            )
-        load_sharded_model(accelerator, unwrapped, pretrained_dir)
-
-    if sharded:
-        # Requires the prepared optimizer: FSDP2's prepare rebinds param groups to DTensors but
-        # never migrates optimizer.state — DCP reshards it here (works across topology changes).
-        load_sharded_optimizer(accelerator, optimizer, unwrapped, training_state_dir)
-    else:
-        load_optimizer_state(optimizer, training_state_dir)
-
+    step = load_training_step(training_state_dir)
+    if load_optimizer:
+        optimizer = load_optimizer_state(optimizer, training_state_dir)
     if scheduler is not None:
-        load_scheduler_state(scheduler, training_state_dir)
+        scheduler = load_scheduler_state(scheduler, training_state_dir)
+
+    return step, optimizer, scheduler
 
 
-# ---------------------------------------------------------------------------------------------
-# Hub: checkpoint push (resume artifact) and publishing (distribution artifact)
-# ---------------------------------------------------------------------------------------------
+def gather_fsdp_state_dicts(model, optimizer) -> tuple[dict, dict]:
+    """Gather the full (unsharded) model and optimizer state dicts under FSDP.
+
+    `model.state_dict()` and `FSDP.optim_state_dict(...)` are cross-rank collectives, so this must be
+    called on *every* rank with the prepared (FSDP-wrapped) `model` and `optimizer`. With
+    `rank0_only=True` and `offload_to_cpu=True`, every rank runs the all-gather but only rank 0
+    materializes the full dicts (the others get empty dicts) and they are kept on CPU to bound GPU
+    memory. The returned optimizer state dict is keyed by parameter FQNs and is world-size
+    independent; `load_fsdp_optimizer_state` reshards it on resume.
+
+    Returns:
+        (model_state_dict, optim_state_dict): full dicts on rank 0, empty dicts on other ranks.
+    """
+    from torch.distributed.fsdp import (
+        FullOptimStateDictConfig,
+        FullStateDictConfig,
+        FullyShardedDataParallel as FSDP,  # noqa F401
+        StateDictType,
+    )
+
+    state_cfg = FullStateDictConfig(offload_to_cpu=True, rank0_only=True)
+    optim_cfg = FullOptimStateDictConfig(offload_to_cpu=True, rank0_only=True)
+    with FSDP.state_dict_type(model, StateDictType.FULL_STATE_DICT, state_cfg, optim_cfg):
+        model_state_dict = model.state_dict()
+        optim_state_dict = FSDP.optim_state_dict(model, optimizer)
+    return model_state_dict, optim_state_dict
+
+
+def load_fsdp_optimizer_state(model, optimizer, checkpoint_dir: Path) -> None:
+    """Load the FSDP optimizer state (saved as safetensors) and reshard it into the optimizer.
+
+    This is a cross-rank collective and must be called on every rank *after* `accelerator.prepare()`
+    with the prepared (FSDP-wrapped) `model` and `optimizer`. The saved state is the full,
+    world-size-independent optimizer state (keyed by parameter FQNs); `FSDP.optim_state_dict_to_load`
+    reshards it to the current FSDP topology, so resume on a different number of GPUs works.
+    """
+    from torch.distributed.fsdp import (
+        FullOptimStateDictConfig,
+        FullStateDictConfig,
+        FullyShardedDataParallel as FSDP,  # noqa F401
+        StateDictType,
+    )
+
+    # Every rank reads the same full state from the (shared) checkpoint dir, so rank0_only=False.
+    full_osd = load_optimizer_state_dict(checkpoint_dir / TRAINING_STATE_DIR)
+    state_cfg = FullStateDictConfig(rank0_only=False)
+    optim_cfg = FullOptimStateDictConfig(rank0_only=False)
+    with FSDP.state_dict_type(model, StateDictType.FULL_STATE_DICT, state_cfg, optim_cfg):
+        sharded_osd = FSDP.optim_state_dict_to_load(model=model, optim=optimizer, optim_state_dict=full_osd)
+    optimizer.load_state_dict(sharded_osd)
 
 
 def push_checkpoint_to_hub(
@@ -510,16 +311,6 @@ def push_checkpoint_to_hub(
     The model repo is created idempotently, and the commit is tagged with the
     checkpoint step so a checkpoint can be recovered with
     --policy.pretrained_revision=<step> instead of a commit sha.
-
-    The directory is uploaded verbatim — including DCP shards under the DCP formats: this tree
-    exists for *resume*, not distribution, and `resolve_resume_checkpoint` downloads it back
-    symmetrically.
-
-    Args:
-        checkpoint_dir (Path): The local checkpoint step directory to upload.
-        repo_id (str): The Hub model repo to push to (created idempotently if missing).
-        private (bool | None): Whether a newly created repo should be private. Defaults to
-            None (public unless the organization's default is private).
     """
     api = HfApi()
     api.create_repo(repo_id=repo_id, repo_type="model", private=private, exist_ok=True)
@@ -547,16 +338,6 @@ def resolve_resume_checkpoint(repo_id: str, output_dir: Path) -> Path:
     into `output_dir/checkpoints/<step>/`, recreate the local `last` symlink, and return that local
     checkpoint dir. Used to resume training from the Hub on a machine (or HF Jobs pod) that does not
     have the original local run dir.
-
-    Args:
-        repo_id (str): The Hub model repo holding `checkpoints/<step>/` subtrees.
-        output_dir (Path): The local run directory to download the checkpoint into.
-
-    Returns:
-        Path: The local checkpoint step directory, `output_dir/checkpoints/<step>`.
-
-    Raises:
-        FileNotFoundError: If the repo contains no checkpoints under `checkpoints/`.
     """
     latest = find_latest_hub_checkpoint(repo_id)
     if latest is None:
@@ -573,214 +354,3 @@ def resolve_resume_checkpoint(repo_id: str, output_dir: Path) -> Path:
     checkpoint_dir = output_dir / latest
     update_last_checkpoint(checkpoint_dir)
     return checkpoint_dir
-
-
-def publish_trained_model(
-    cfg: TrainPipelineConfig,
-    model: "PreTrainedPolicy | PreTrainedRewardModel",
-    preprocessor: PolicyProcessorPipeline | None,
-    postprocessor: PolicyProcessorPipeline | None,
-    dataset_meta: "LeRobotDatasetMetadata | None",
-    *,
-    peft_model: Any | None = None,
-) -> None:
-    """Publish the complete training bundle as a distributable model repo.
-
-    Collective-safe: call on ALL ranks — the model commit gathers sharded weights through
-    `save_pretrained`; uploads happen on the main process only (gated inside
-    `HubMixin.push_to_hub` and here). Commits, in order: (1) the model (skipped for PEFT —
-    adapters replace full weights), (2) the preprocessor, (3) the postprocessor, (4) the bundle
-    sidecar: README.md model card + train_config.json (+ adapter weights and the wrapped
-    policy's config in the PEFT case). Every commit uploads a freshly assembled directory, so
-    a published repo carries only the distributable artifacts.
-
-    Args:
-        cfg (TrainPipelineConfig): The training config; saved as `train_config.json` and used
-            to render the model card.
-        model (PreTrainedPolicy | PreTrainedRewardModel): The trained model to publish; its
-            config supplies the target repo id, visibility, license, and tags.
-        preprocessor (PolicyProcessorPipeline | None): The preprocessor pipeline to publish
-            alongside the model, if any.
-        postprocessor (PolicyProcessorPipeline | None): The postprocessor pipeline to publish
-            alongside the model, if any.
-        dataset_meta (LeRobotDatasetMetadata | None): Dataset metadata for the model card, if
-            available.
-        peft_model (Any | None): The PEFT wrapper when training adapters; its adapter weights
-            replace the full model weights in the published repo. Defaults to None.
-
-    Raises:
-        ValueError: If the model config carries no repo id (`--policy.repo_id`).
-    """
-    model_cfg = model.config
-    repo_id = model_cfg.repo_id
-    if not repo_id:
-        raise ValueError("Publishing requires a repo id (--policy.repo_id).")
-    ignore = ["*.tmp", "*.log"]
-
-    if peft_model is None:
-        # Calls are made on the exact objects that own each method (never through PEFT's
-        # attribute forwarding), so the peft branch below never touches this path.
-        model.push_to_hub(repo_id, private=model_cfg.private, ignore_patterns=ignore)
-    if preprocessor is not None:
-        preprocessor.push_to_hub(repo_id, private=model_cfg.private)
-    if postprocessor is not None:
-        postprocessor.push_to_hub(repo_id, private=model_cfg.private)
-
-    if is_main_process():
-        api = HfApi()
-        repo_id = api.create_repo(repo_id=repo_id, private=model_cfg.private, exist_ok=True).repo_id
-        with TemporaryDirectory(ignore_cleanup_errors=True) as tmp:
-            saved_path = Path(tmp) / repo_id
-            saved_path.mkdir(parents=True, exist_ok=True)
-            if peft_model is not None:
-                peft_model.save_pretrained(saved_path)  # adapter weights + adapter config
-                model.config.save_pretrained(saved_path)  # PEFT cannot write the policy config
-            card = generate_model_card(model_cfg, cfg=cfg, dataset_meta=dataset_meta)
-            card.save(str(saved_path / "README.md"))
-            cfg.save_pretrained(saved_path)  # train_config.json
-            commit_info = api.upload_folder(
-                repo_id=repo_id,
-                repo_type="model",
-                folder_path=saved_path,
-                commit_message="Upload model card and train config",
-                allow_patterns=["*.safetensors", "*.json", "*.yaml", "*.md"],
-                ignore_patterns=ignore,
-            )
-        # Contract: lerobot.jobs.hf.submit_to_hf watches for this exact "Model pushed to <url>"
-        # line to end a remote run early. Keep the wording and URL format in sync.
-        logging.info(f"Model pushed to {commit_info.repo_url.url}")
-
-    if dist.is_initialized():
-        dist.barrier()
-
-
-# ---------------------------------------------------------------------------------------------
-# Model card
-# ---------------------------------------------------------------------------------------------
-
-_BASE_MODEL_MAPPING = {
-    "smolvla": "lerobot/smolvla_base",
-    "pi0": "lerobot/pi0_base",
-    "pi05": "lerobot/pi05_base",
-    "pi0_fast": "lerobot/pi0fast-base",
-    "xvla": "lerobot/xvla-base",
-}
-
-
-def build_card_context(
-    cfg: TrainPipelineConfig | None,
-    dataset_meta: "LeRobotDatasetMetadata | None",
-    input_features: dict | None,
-    output_features: dict | None,
-) -> dict:
-    """Collect optional data for the model-card template.
-
-    Returns plain values only (no Markdown) — the template in
-    ``lerobot/templates/lerobot_modelcard_template.md`` decides how and whether to show
-    each one. Everything is best-effort: anything unavailable is left empty/None and the
-    template simply skips that section, so this never breaks a Hub push.
-
-    Args:
-        cfg (TrainPipelineConfig | None): The training config supplying the training section,
-            if available.
-        dataset_meta (LeRobotDatasetMetadata | None): Dataset metadata supplying the dataset,
-            robot-type, and camera sections, if available.
-        input_features (dict | None): The policy's input feature declarations, if any.
-        output_features (dict | None): The policy's output feature declarations, if any.
-
-    Returns:
-        dict: Template context with `training`, `input_features`, `output_features`,
-            `dataset`, `robot_type`, and `cameras` entries; unavailable pieces stay
-            empty/None.
-    """
-    context = {
-        "training": None,
-        "input_features": input_features or {},
-        "output_features": output_features or {},
-        "dataset": None,
-        "robot_type": None,
-        "cameras": [],
-    }
-
-    if cfg is not None:
-        optimizer = getattr(cfg, "optimizer", None)
-        context["training"] = {
-            "steps": cfg.steps,
-            "batch_size": cfg.batch_size,
-            "seed": cfg.seed,
-            "optimizer": getattr(optimizer, "type", None) if optimizer else None,
-            "lr": getattr(optimizer, "lr", None) if optimizer else None,
-            "lerobot_version": __version__,
-        }
-
-    if dataset_meta is not None:
-        context["dataset"] = {
-            "repo_id": dataset_meta.repo_id,
-            "episodes": dataset_meta.total_episodes,
-            "frames": dataset_meta.total_frames,
-            "fps": dataset_meta.fps,
-            "tasks": [str(task) for task in dataset_meta.tasks.index],
-        }
-        context["robot_type"] = dataset_meta.robot_type
-        context["cameras"] = [key.split(".")[-1] for key in dataset_meta.camera_keys]
-
-    return context
-
-
-def generate_model_card(
-    model_cfg: PreTrainedConfig | RewardModelConfig,
-    cfg: TrainPipelineConfig | None = None,
-    dataset_meta: "LeRobotDatasetMetadata | None" = None,
-) -> ModelCard:
-    """Render the LeRobot model card for a trained policy or reward model.
-
-    A free function on purpose: every template variable comes from arguments — the model
-    config, the training config, and the dataset metadata — none from a live model, so a card
-    can also be rendered from a checkpoint's `config.json` alone (see `lerobot-convert-dcp`).
-    The config type selects the template: reward models get the reward-model card, policies the
-    policy card with the training/dataset sections.
-
-    Args:
-        model_cfg (PreTrainedConfig | RewardModelConfig): The model config providing type,
-            license, tags, repo id, and — for policies — the feature declarations.
-        cfg (TrainPipelineConfig | None, optional): The training config for the training and
-            dataset card sections. Defaults to None.
-        dataset_meta (LeRobotDatasetMetadata | None, optional): Dataset metadata for the
-            dataset card sections. Defaults to None.
-
-    Returns:
-        ModelCard: The rendered and validated LeRobot model card.
-    """
-    model_type = model_cfg.type
-    base_model = _BASE_MODEL_MAPPING.get(model_type)
-
-    if isinstance(model_cfg, RewardModelConfig):
-        tags = {"robotics", "lerobot", "reward-model", model_type}
-        template_card = (
-            files("lerobot.templates")
-            .joinpath("lerobot_rewardmodel_modelcard_template.md")
-            .read_text("utf-8")
-        )
-        context: dict[str, Any] = {}  # the reward template renders from card_data alone
-    else:
-        tags = {"robotics", "lerobot", model_type}
-        template_card = (
-            files("lerobot.templates").joinpath("lerobot_modelcard_template.md").read_text("utf-8")
-        )
-        context = build_card_context(cfg, dataset_meta, model_cfg.input_features, model_cfg.output_features)
-        # Used by the template to pre-fill commands and the "Fine-tuned from" line.
-        context["policy_repo_id"] = model_cfg.repo_id
-        context["base_model"] = base_model
-
-    card_data = ModelCardData(
-        license=model_cfg.license or "apache-2.0",
-        library_name="lerobot",
-        pipeline_tag="robotics",
-        tags=list(tags.union(model_cfg.tags or [])),
-        model_name=model_type,
-        datasets=cfg.dataset.repo_id if cfg is not None else None,
-        base_model=base_model,
-    )
-    card = ModelCard.from_template(card_data, template_str=template_card, **context)
-    card.validate()
-    return card

src/lerobot/configs/accelerator.py

@@ -1,273 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Execution-runtime configuration: everything handed to (or applied by) the `Accelerator`.
-
-Each sub-config mirrors the plain-typed subset of the corresponding accelerate object and
-builds it at runtime (the way ``OptimizerConfig.build()`` constructs a ``torch.optim.Optimizer``),
-so the whole tree round-trips through the CLI and ``train_config.json`` and parsing a config
-never imports accelerate.
-"""
-
-from dataclasses import dataclass, field
-from enum import Enum
-from typing import TYPE_CHECKING
-
-from lerobot.configs.parallelism import ParallelismConfig
-
-if TYPE_CHECKING:
-    from accelerate import Accelerator
-    from accelerate.utils import (
-        DistributedDataParallelKwargs,
-        FullyShardedDataParallelPlugin,
-        GradientAccumulationPlugin,
-    )
-
-
-@dataclass
-class FSDPConfig:
-    """Mirror of the `FullyShardedDataParallelPlugin` subset LeRobot supports (FSDP2 only).
-
-    Exactly one wrap policy applies: `wrap_modules` (module *class names* forming the FSDP
-    units — and, later, the activation-checkpointing units) or `min_num_params` (size-based).
-    When both are None, the policy's own `_fsdp_wrap_modules` declaration is used; a run where
-    no wrap source exists at all fails loudly rather than silently wrapping only the root.
-    """
-
-    reshard_after_forward: bool = True
-    wrap_modules: list[str] | None = None
-    min_num_params: int | None = None
-    cpu_offload: bool = False
-    # Regex matched against module FQNs to exclude their parameters from sharding.
-    ignored_modules: str | None = None
-
-    def __post_init__(self) -> None:
-        """Validate the wrap-policy fields.
-
-        Raises:
-            ValueError: If both ``wrap_modules`` and ``min_num_params`` are set (they are
-                mutually exclusive wrap policies), or if ``min_num_params`` is < 1.
-        """
-        if self.wrap_modules is not None and self.min_num_params is not None:
-            raise ValueError(
-                "fsdp.wrap_modules and fsdp.min_num_params are mutually exclusive wrap policies."
-            )
-        if self.min_num_params is not None and self.min_num_params < 1:
-            raise ValueError(f"fsdp.min_num_params must be >= 1, got {self.min_num_params}.")
-
-    def build_plugin(self) -> "FullyShardedDataParallelPlugin":
-        """Build the FSDP2 plugin for `Accelerator(fsdp_plugin=...)`.
-
-        Returns:
-            FullyShardedDataParallelPlugin: FSDP2 (`fsdp_version=2`) plugin carrying the
-                mirrored wrap policy, resharding, CPU-offload, and ignored-modules settings.
-        """
-        from accelerate.utils import FullyShardedDataParallelPlugin
-
-        use_size_policy = self.min_num_params is not None
-        return FullyShardedDataParallelPlugin(
-            fsdp_version=2,
-            reshard_after_forward=self.reshard_after_forward,
-            auto_wrap_policy="size_based_wrap" if use_size_policy else "transformer_based_wrap",
-            # May legitimately still be None here: the policy-declared default is applied right
-            # before `accelerator.prepare()` (see lerobot.distributed.factory.set_fsdp_wrap_modules).
-            transformer_cls_names_to_wrap=list(self.wrap_modules) if self.wrap_modules else None,
-            min_num_params=self.min_num_params,
-            cpu_offload=self.cpu_offload,
-            ignored_modules=self.ignored_modules,
-            # state_dict_type stays at the FSDP2 default (SHARDED_STATE_DICT) and is never
-            # switched: full gathers go through torch's state-dict API, which does not consult
-            # the plugin. activation_checkpointing stays False: AC is LeRobot-owned.
-        )
-
-
-@dataclass
-class DDPConfig:
-    """Mirror of the `DistributedDataParallelKwargs` subset LeRobot exposes."""
-
-    # Today's in-script default, kept for models with conditional computation.
-    find_unused_parameters: bool = True
-    gradient_as_bucket_view: bool = False
-    static_graph: bool = False
-
-    def build_kwargs_handler(self) -> "DistributedDataParallelKwargs":
-        """Build the DDP kwargs handler for `Accelerator(kwargs_handlers=[...])`.
-
-        Returns:
-            DistributedDataParallelKwargs: Handler carrying the mirrored DDP fields, applied
-                by accelerate when it wraps the model in `DistributedDataParallel`.
-        """
-        from accelerate.utils import DistributedDataParallelKwargs
-
-        return DistributedDataParallelKwargs(
-            find_unused_parameters=self.find_unused_parameters,
-            gradient_as_bucket_view=self.gradient_as_bucket_view,
-            static_graph=self.static_graph,
-        )
-
-
-@dataclass
-class GradientAccumulationConfig:
-    """Mirror of the `GradientAccumulationPlugin` subset LeRobot supports.
-
-    Only the step count is a knob. ``sync_with_dataloader`` is pinned to False by
-    :meth:`build_plugin`: the training loop cycles a finite dataloader, so accelerate's default
-    of syncing at every dataloader end would force an optimizer step at every dataset epoch
-    boundary instead of every ``steps`` micro-batches.
-    """
-
-    steps: int = 1
-
-    def __post_init__(self) -> None:
-        """Validate the accumulation step count.
-
-        Raises:
-            ValueError: If ``steps`` is < 1.
-        """
-        if self.steps < 1:
-            raise ValueError(f"gradient_accumulation.steps must be >= 1, got {self.steps}.")
-
-    def build_plugin(self) -> "GradientAccumulationPlugin":
-        """Build the plugin for `Accelerator(gradient_accumulation_plugin=...)`.
-
-        A named plugin argument, not a `kwargs_handlers` entry: accelerate consumes this object
-        through its dedicated constructor parameter — the `KwargsHandler` base class only lends
-        it `to_kwargs()`, so the consumption site, not the inheritance, decides its role.
-
-        Returns:
-            GradientAccumulationPlugin: Carrying the mirrored step count, with
-                ``sync_with_dataloader=False`` pinned (see the class docstring).
-        """
-        from accelerate.utils import GradientAccumulationPlugin
-
-        return GradientAccumulationPlugin(num_steps=self.steps, sync_with_dataloader=False)
-
-
-@dataclass
-class CompileConfig:
-    """torch.compile knobs — a configured placeholder: wiring lands in a later round.
-
-    The setup-order contract it will follow is already fixed: compile applies
-    after CP dispatch install and activation checkpointing, before `fully_shard`, regionally
-    (per wrap unit) — the only combination proven with FSDP2.
-    """
-
-    enabled: bool = False
-    backend: str = "inductor"
-    mode: str | None = None
-    regional: bool = True
-
-
-class ActivationCheckpointingMode(str, Enum):
-    NONE = "none"
-    FULL = "full"
-
-
-@dataclass
-class ActivationCheckpointingConfig:
-    """Activation-checkpointing knobs — a configured placeholder: wiring lands in a later round.
-
-    AC units will coincide with the FSDP wrap units (one declaration drives both), applied
-    before torch.compile and `fully_shard` (the same ordering contract as CompileConfig).
-    """
-
-    mode: ActivationCheckpointingMode = ActivationCheckpointingMode.NONE
-
-
-@dataclass
-class AcceleratorConfig:
-    """Builds the `Accelerator` — the runtime counterpart of the `parallelism` topology.
-
-    `mixed_precision` selects accelerate-native AMP for DDP/single-GPU runs and the FSDP2
-    `MixedPrecisionPolicy` for sharded runs (accelerate derives it). Sharded runs support
-    "no" and "bf16" only; fp16's GradScaler-over-DTensor path is unverified and fails fast
-    at config validation.
-    """
-
-    mixed_precision: str = "no"
-    gradient_accumulation: GradientAccumulationConfig = field(default_factory=GradientAccumulationConfig)
-    fsdp: FSDPConfig = field(default_factory=FSDPConfig)
-    ddp: DDPConfig = field(default_factory=DDPConfig)
-    compile: CompileConfig = field(default_factory=CompileConfig)
-    activation_checkpointing: ActivationCheckpointingConfig = field(
-        default_factory=ActivationCheckpointingConfig
-    )
-
-    def __post_init__(self) -> None:
-        """Validate the accelerate-facing scalar fields.
-
-        Raises:
-            ValueError: If ``mixed_precision`` is not one of ``"no"``, ``"fp16"``, ``"bf16"``.
-        """
-        if self.mixed_precision not in ("no", "fp16", "bf16"):
-            raise ValueError(
-                f"mixed_precision must be one of 'no', 'fp16', 'bf16', got {self.mixed_precision!r}."
-            )
-
-    def build(self, parallelism: ParallelismConfig, *, cpu: bool = False) -> "Accelerator":
-        """Translate the mirrored fields into a ready `Accelerator` (call once per process).
-
-        `parallelism` must already be resolved against the world size. The degradation matrix
-        is encoded here and nowhere else: sharded -> FSDP2 (+HSDP via the accelerate
-        `ParallelismConfig` mesh), replicated-only -> DDP kwargs, single process -> plain.
-
-        Args:
-            parallelism (ParallelismConfig): The resolved process topology; selects which
-                accelerate path (FSDP2 mesh, DDP kwargs handler, or plain) is configured.
-            cpu (bool): Force CPU execution even when CUDA is available. Defaults to False.
-
-        Returns:
-            Accelerator: The configured accelerate entry point for this process.
-        """
-        from accelerate import Accelerator
-
-        kwargs: dict = {
-            # LeRobot steps its scheduler manually once per training step; accelerate must not
-            # rescale scheduler stepping by num_processes.
-            "step_scheduler_with_optimizer": False,
-            "gradient_accumulation_plugin": self.gradient_accumulation.build_plugin(),
-            "mixed_precision": self.mixed_precision,
-            "cpu": cpu,
-        }
-        if parallelism.is_sharded:
-            kwargs["fsdp_plugin"] = self.fsdp.build_plugin()
-            kwargs["parallelism_config"] = _accelerate_parallelism_config(parallelism)
-        elif parallelism.is_replicated_only:
-            kwargs["kwargs_handlers"] = [self.ddp.build_kwargs_handler()]
-        return Accelerator(**kwargs)
-
-
-def _accelerate_parallelism_config(parallelism: ParallelismConfig) -> object:
-    """LeRobot topology -> accelerate `ParallelismConfig`.
-
-    CP is declared honestly (`cp_size = ring x ulysses`) so accelerate builds the canonical
-    mesh, folds CP into the FSDP shard group (`dp_shard_cp`), and duplicates batches within CP
-    groups. The ring/ulysses sub-structure stays private to `lerobot.distributed.ParallelDims`.
-
-    Args:
-        parallelism (ParallelismConfig): The resolved LeRobot topology to translate.
-
-    Returns:
-        object: The accelerate `ParallelismConfig` mirroring `dp_replicate`, `dp_shard`, and
-            the collapsed `cp_size` (annotated as `object` so importing this module never
-            imports accelerate).
-    """
-    from accelerate.parallelism_config import ParallelismConfig as AccelerateParallelismConfig
-
-    return AccelerateParallelismConfig(
-        dp_replicate_size=parallelism.dp_replicate,
-        dp_shard_size=parallelism.dp_shard,
-        cp_size=parallelism.cp_size,
-    )

src/lerobot/configs/default.py

@@ -14,7 +14,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import logging
 from dataclasses import dataclass, field
 
 from lerobot.transforms import ImageTransformsConfig
@@ -22,8 +21,6 @@ from lerobot.utils.import_utils import get_safe_default_video_backend
 
 from .video import DEFAULT_DEPTH_UNIT, DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT
 
-logger = logging.getLogger(__name__)
-
 
 @dataclass
 class DatasetConfig:
@@ -32,15 +29,10 @@ class DatasetConfig:
     # "dataset_index" into the returned item. The index mapping is made according to the order in which the
     # datasets are provided.
     repo_id: str
-    # Hub repository type: "dataset" (default) or "bucket" for an HF Storage Bucket streamed over
-    # hf://buckets/. Buckets are streaming-only, so "bucket" requires streaming=true.
-    repo_type: str = "dataset"
     # Root directory for a concrete local dataset tree (e.g. 'dataset/path'). If None, local datasets are
     # looked up under $HF_LEROBOT_HOME/repo_id and Hub downloads use a revision-safe cache under $HF_LEROBOT_HOME/hub.
     root: str | None = None
     episodes: list[int] | None = None
-    # Episode indices to drop (e.g. corrupt or heterogeneous ones). Applied on top of `episodes`.
-    exclude_episodes: list[int] | None = None
     image_transforms: ImageTransformsConfig = field(default_factory=ImageTransformsConfig)
     revision: str | None = None
     use_imagenet_stats: bool = True
@@ -56,16 +48,6 @@ class DatasetConfig:
     eval_split: float = 0.0
 
     def __post_init__(self) -> None:
-        if self.repo_type not in ("dataset", "bucket"):
-            raise ValueError(f"repo_type must be 'dataset' or 'bucket', got {self.repo_type!r}")
-        if self.repo_type == "bucket" and not self.streaming:
-            raise ValueError(
-                "repo_type='bucket' is streaming-only: set streaming=true to train from an HF Storage Bucket."
-            )
-        if self.repo_type == "bucket" and self.eval_split != 0.0:
-            raise ValueError(
-                "eval_split requires map-style datasets and is not supported with repo_type='bucket'."
-            )
         if self.depth_output_unit not in (DEPTH_METER_UNIT, DEPTH_MILLIMETER_UNIT):
             raise ValueError(
                 f"depth_output_unit must be '{DEPTH_METER_UNIT}' or '{DEPTH_MILLIMETER_UNIT}', got {self.depth_output_unit!r}"
@@ -80,14 +62,6 @@ class DatasetConfig:
             if len(self.episodes) != len(set(self.episodes)):
                 duplicates = sorted({ep for ep in self.episodes if self.episodes.count(ep) > 1})
                 raise ValueError(f"Episode indices contain duplicates: {duplicates}")
-        if self.exclude_episodes is not None:
-            negative_episodes = [episode for episode in self.exclude_episodes if episode < 0]
-            if negative_episodes:
-                logger.warning(
-                    "Ignoring negative exclude_episodes entries: %s",
-                    negative_episodes,
-                )
-                self.exclude_episodes = [episode for episode in self.exclude_episodes if episode >= 0]
 
 
 @dataclass

src/lerobot/configs/parallelism.py

@@ -1,190 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Declarative process topology for distributed training and inference.
-
-The mesh convention (canonical row-major rank layout, outermost first)::
-
-    (dp_replicate, dp_shard, ring, ulysses)
-
-- ``dp_replicate x dp_shard`` is the data-parallel world: HSDP replicates over
-  ``dp_replicate`` and shards parameters over ``dp_shard``. FSDP2's actual shard
-  group folds context parallelism in (``dp_shard x ring x ulysses``), matching
-  accelerate's ``dp_shard_cp`` flattening and torchtitan's ``fsdp`` axis.
-- ``ring`` is the outer and ``ulysses`` the inner context-parallel dim
-  (diffusers convention: ulysses all-to-all exchanges run over adjacent, typically
-  NVLink-connected ranks).
-- ``cfg_parallel`` (classifier-free-guidance parallelism) is a branch-parallel,
-  inference-only dim that sits between dp and the sequence dims. It never
-  affects weight sharding or checkpoints.
-
-This module is pure configuration: plain-typed dataclasses that draccus can
-round-trip through the CLI and ``train_config.json``. Runtime objects (device
-meshes, process groups) live in :mod:`lerobot.distributed`.
-"""
-
-import os
-from dataclasses import dataclass, field
-
-
-@dataclass
-class ContextParallelConfig:
-    """Ring x Ulysses context parallelism (sequence parallelism for attention).
-
-    Both degrees are configured placeholders in this release: the CP engine is not implemented
-    yet, and enabling either degree > 1 fails fast at config validation. The fields exist now so
-    that the CLI surface, checkpoint metadata, and mesh math are stable when the engine lands.
-    """
-
-    ring_degree: int = 1
-    ulysses_degree: int = 1
-
-    def __post_init__(self) -> None:
-        """Validate the declared context-parallel degrees.
-
-        Raises:
-            ValueError: If ``ring_degree`` or ``ulysses_degree`` is < 1.
-        """
-        if self.ring_degree < 1 or self.ulysses_degree < 1:
-            raise ValueError(
-                f"Context-parallel degrees must be >= 1, got ring_degree={self.ring_degree}, "
-                f"ulysses_degree={self.ulysses_degree}."
-            )
-
-    @property
-    def size(self) -> int:
-        """Total number of ranks a full sequence is sharded across."""
-        return self.ring_degree * self.ulysses_degree
-
-
-@dataclass
-class ParallelismConfig:
-    """Degrees of every parallelism dim. Invariant: their product equals the world size.
-
-    Degradations are expressed purely through the degrees (no mode flags):
-
-    - single process: all degrees 1;
-    - DDP: ``dp_replicate == world_size`` (auto-filled when every sharding field is left at its
-      default — plain ``torchrun`` keeps today's out-of-the-box behavior);
-    - FSDP: ``dp_shard > 1`` (or ``-1`` to fill the remaining world into the shard dim);
-    - HSDP: ``dp_replicate > 1`` and ``dp_shard > 1``.
-
-    ``resolve()`` turns the declared degrees into concrete ones once the world size is known and
-    is the single place the world-size equation is enforced. It is called by
-    :func:`lerobot.distributed.factory.make_accelerator`; the config is inert until then.
-    """
-
-    dp_replicate: int = 1
-    # -1 is an explicit opt-in sentinel: shard over world_size // (dp_replicate * cp).
-    dp_shard: int = 1
-    context_parallel: ContextParallelConfig = field(default_factory=ContextParallelConfig)
-    # Classifier-free-guidance parallelism — inference-only (cosmos/vllm-omni precedent:
-    # cond/uncond branches on different ranks). Reserved for the serving round; training
-    # validates it to 1. Meaningful values are 1 or 2 (Cosmos3 has two CFG branches).
-    cfg_parallel: int = 1
-
-    def __post_init__(self) -> None:
-        """Validate the declared degrees (world-size-independent checks only).
-
-        Raises:
-            ValueError: If ``dp_replicate`` is < 1, ``dp_shard`` is neither >= 1 nor the
-                ``-1`` infer sentinel, or ``cfg_parallel`` is not 1 or 2.
-        """
-        if self.dp_replicate < 1:
-            raise ValueError(f"dp_replicate must be >= 1, got {self.dp_replicate}.")
-        if self.dp_shard < 1 and self.dp_shard != -1:
-            raise ValueError(f"dp_shard must be >= 1, or -1 to infer, got {self.dp_shard}.")
-        if self.cfg_parallel not in (1, 2):
-            raise ValueError(f"cfg_parallel must be 1 or 2, got {self.cfg_parallel}.")
-
-    @property
-    def cp_size(self) -> int:
-        """Total context-parallel size (``ring_degree * ulysses_degree``)."""
-        return self.context_parallel.size
-
-    @property
-    def is_sharded(self) -> bool:
-        """True when the run uses FSDP2 (parameters sharded); selects the sharded engine path."""
-        return self.dp_shard != 1 or self.cp_size > 1
-
-    @property
-    def is_replicated_only(self) -> bool:
-        """True for plain DDP (weights replicated, no sharding)."""
-        return not self.is_sharded and self.dp_replicate > 1
-
-    @property
-    def dp_world_size(self) -> int:
-        """Number of distinct data-parallel workers (batches are sharded this many ways).
-
-        Returns:
-            int: ``dp_replicate * dp_shard``.
-
-        Raises:
-            RuntimeError: If accessed while ``dp_shard`` is still the ``-1`` sentinel, i.e.
-                before :meth:`resolve` has bound the degrees to a world size.
-        """
-        if self.dp_shard == -1:
-            raise RuntimeError("dp_world_size is undefined before resolve() fills dp_shard=-1.")
-        return self.dp_replicate * self.dp_shard
-
-    def resolve(self, world_size: int) -> None:
-        """Bind the declared degrees to a concrete world size (idempotent).
-
-        Fills the ``dp_shard=-1`` sentinel, auto-fills ``dp_replicate`` for the DDP degradation,
-        and enforces ``dp_replicate * dp_shard * cp == world_size`` with every degree echoed on
-        failure.
-
-        Args:
-            world_size (int): Total number of launched processes (torchrun's ``WORLD_SIZE``).
-
-        Raises:
-            ValueError: If a context-parallel degree is > 1 (the CP engine is not implemented
-                yet), if ``dp_shard=-1`` cannot be inferred because ``world_size`` is not
-                divisible by ``dp_replicate * cp``, or if the resolved degrees do not multiply
-                to ``world_size``.
-        """
-        if self.cp_size > 1:
-            raise ValueError(
-                "Context parallelism is not implemented yet: ring_degree and ulysses_degree "
-                "must be 1. The fields are reserved for the CP engine round."
-            )
-        if self.is_sharded:
-            if self.dp_shard == -1:
-                self.dp_shard, remainder = divmod(world_size, self.dp_replicate * self.cp_size)
-                if remainder or self.dp_shard < 1:
-                    raise ValueError(
-                        f"Cannot infer dp_shard: world_size={world_size} is not divisible by "
-                        f"dp_replicate={self.dp_replicate} * cp={self.cp_size}."
-                    )
-        elif self.dp_replicate == 1:
-            # Untouched config on a multi-process launch: fill the DDP degradation.
-            self.dp_replicate = world_size
-        total = self.dp_replicate * self.dp_shard * self.cp_size
-        if total != world_size:
-            raise ValueError(
-                f"Parallelism degrees do not multiply to the world size: dp_replicate="
-                f"{self.dp_replicate} * dp_shard={self.dp_shard} * ring="
-                f"{self.context_parallel.ring_degree} * ulysses="
-                f"{self.context_parallel.ulysses_degree} = {total} != WORLD_SIZE={world_size}."
-            )
-
-
-def world_size_from_env() -> int:
-    """World size as set by torchrun (or 1 outside distributed launches).
-
-    Returns:
-        int: The ``WORLD_SIZE`` environment variable, or 1 when unset.
-    """
-    return int(os.environ.get("WORLD_SIZE", "1"))

src/lerobot/configs/recipe.py

@@ -23,7 +23,6 @@ from typing import Any, Literal, get_args
 
 MessageRole = Literal["user", "assistant", "system", "tool"]
 MessageStream = Literal["high_level", "low_level"]
-RecipeRoute = Literal["vqa"]
 
 DEFAULT_BINDINGS = {
     "subtask": "active_at(t, style=subtask)",
@@ -41,7 +40,6 @@ discovery (here) and rendered-message substitution (in ``language_render``)."""
 
 _VALID_ROLES = frozenset(get_args(MessageRole))
 _VALID_STREAMS = frozenset(get_args(MessageStream))
-_VALID_ROUTES = frozenset(get_args(RecipeRoute))
 
 
 @dataclass
@@ -80,7 +78,7 @@ class MessageTurn:
             raise ValueError(f"Unsupported message stream: {self.stream!r}")
         if self.content is None and self.tool_calls_from is None:
             raise ValueError("MessageTurn.content is required unless tool_calls_from is set.")
-        if self.content is not None and not isinstance(self.content, str | list):
+        if self.content is not None and not isinstance(self.content, (str, list)):
             raise TypeError("MessageTurn.content must be a string, a list of HF-style blocks, or None.")
         if isinstance(self.content, list):
             for block in self.content:
@@ -101,16 +99,13 @@ class TrainingRecipe:
 
     A recipe is either a *message recipe* (``messages`` plus optional
     ``bindings``) or a *blend recipe* (``blend`` mapping names to weighted
-    sub-recipes). ``weight`` and ``route`` are only meaningful inside a blend;
-    ``route: vqa`` gives sparse VQA annotations priority over normal weighted
-    selection.
+    sub-recipes). ``weight`` is only meaningful inside a blend.
     """
 
     messages: list[MessageTurn] | None = None
     bindings: dict[str, str] | None = None
     blend: dict[str, TrainingRecipe] | None = None
     weight: float | None = None
-    route: RecipeRoute | None = None
 
     def __post_init__(self) -> None:
         """Validate that exactly one of ``messages`` or ``blend`` is set."""
@@ -118,10 +113,6 @@ class TrainingRecipe:
             raise ValueError("TrainingRecipe must set only one of messages or blend.")
         if self.messages is None and self.blend is None:
             raise ValueError("TrainingRecipe must set one of messages or blend.")
-        if self.route is not None and self.route not in _VALID_ROUTES:
-            raise ValueError(f"Unsupported recipe route: {self.route!r}")
-        if self.blend is not None and self.route is not None:
-            raise ValueError("TrainingRecipe.route may only be set on a message recipe inside a blend.")
 
         if self.messages is not None:
             self._validate_message_recipe()
@@ -156,9 +147,8 @@ class TrainingRecipe:
         return cls.from_dict(data)
 
     def _validate_message_recipe(self) -> None:
-        """Validate bindings and require text or low-level action supervision."""
-        if self.messages is None:
-            raise ValueError("Cannot validate a message recipe without messages.")
+        """Ensure every templated binding is known and at least one turn is a target."""
+        assert self.messages is not None
         known_bindings = set(DEFAULT_BINDINGS) | set(self.bindings or {}) | {"task"}
 
         for turn in self.messages:
@@ -166,19 +156,12 @@ class TrainingRecipe:
             if missing:
                 raise ValueError(f"MessageTurn references unknown binding(s): {sorted(missing)}")
 
-        has_target = any(turn.target for turn in self.messages)
-        has_low_level = any(turn.stream == "low_level" for turn in self.messages)
-        if not (has_target or has_low_level):
-            raise ValueError(
-                "Message recipes must contain at least one supervised turn — "
-                "either ``target: true`` (text CE) or ``stream: low_level`` "
-                "(flow/action loss)."
-            )
+        if not any(turn.target for turn in self.messages):
+            raise ValueError("Message recipes must contain at least one target turn.")
 
     def _validate_blend_recipe(self) -> None:
         """Ensure each blend component is a non-empty, weighted message recipe."""
-        if self.blend is None:
-            raise ValueError("Cannot validate a blend recipe without blend components.")
+        assert self.blend is not None
         if not self.blend:
             raise ValueError("Blend recipes must contain at least one component.")
 

src/lerobot/configs/recipes/subtask.yaml

@@ -1,16 +0,0 @@
-# Predicts subtasks from tasks and trains subtask-conditioned action flow without memory or plans.
-# Requires `subtask` annotations; samples with missing `if_present` bindings do not render.
-
-blend:
-
-  high_level_subtask:
-    weight: 0.30
-    messages:
-      - {role: user, content: "${task}", stream: high_level}
-      - {role: assistant, content: "${subtask}", stream: high_level, target: true, if_present: subtask}
-
-  low_level_execution:
-    weight: 0.70
-    messages:
-      # The low-level stream trains action flow on the generated or annotated subtask.
-      - {role: user, content: "${subtask}", stream: low_level, if_present: subtask}

src/lerobot/configs/recipes/subtask_joint.yaml

@@ -1,13 +0,0 @@
-# Paper-style joint sequence (pi0.5 §IV-B): one sample supervises the subtask
-# text with CE and, because the assistant turn is part of the prefix, conditions
-# the FAST and flow action losses on the same annotated subtask in one forward.
-# The supervised span is attended causally; the action losses see task + subtask.
-#
-# Pair with `--policy.joint_subtask_conditioning=true` at inference so the flow
-# prefix reproduces this layout (task turn with state + causal generated subtask).
-# Samples without a `subtask` annotation fall back to a plain task-prompt
-# low-level sample via `if_present`.
-
-messages:
-  - {role: user, content: "${task}", stream: low_level}
-  - {role: assistant, content: "${subtask}", stream: low_level, target: true, if_present: subtask}

src/lerobot/configs/recipes/subtask_mem.yaml

@@ -1,30 +0,0 @@
-# Trains subtask prediction, subtask-conditioned action flow, and memory updates without plans.
-# Requires `subtask` and `memory`; missing `if_present` bindings skip the affected sub-recipe.
-
-blend:
-
-  high_level_subtask:
-    weight: 0.25
-    messages:
-      - {role: user, content: "${task}", stream: high_level}
-      - {role: assistant, content: "${subtask}", stream: high_level, target: true, if_present: subtask}
-
-  low_level_execution:
-    weight: 0.60
-    messages:
-      # The low-level stream trains action flow on the generated or annotated subtask.
-      - {role: user, content: "${subtask}", stream: low_level, if_present: subtask}
-
-  memory_update:
-    # `active_at` densifies sparse boundaries while preserving the prior-memory/subtask mapping.
-    # Inference controls update timing through `subtask_change` events.
-    weight: 0.15
-    bindings:
-      prior_memory: "nth_prev(style=memory, offset=1)"
-      current_memory: "active_at(t, style=memory)"
-      completed_subtask: "nth_prev(style=subtask, offset=1)"
-    messages:
-      - {role: user, content: "${task}", stream: high_level}
-      - {role: assistant, content: "Previous memory: ${prior_memory}", stream: high_level, if_present: prior_memory}
-      - {role: user, content: "Completed subtask: ${completed_subtask}", stream: high_level, if_present: completed_subtask}
-      - {role: assistant, content: "${current_memory}", stream: high_level, target: true, if_present: current_memory}

src/lerobot/configs/recipes/subtask_mem_vqa_speech.yaml

@@ -1,72 +0,0 @@
-# Adds memory, spoken interjection responses, and camera-grounded VQA to subtask/action training.
-# Missing optional annotations skip only their sub-recipe; `say` tool calls tokenize as `<say>...</say>`.
-
-blend:
-
-  high_level_subtask:
-    weight: 0.25
-    messages:
-      - {role: user, content: "${task}", stream: high_level}
-      - {role: assistant, content: "${subtask}", stream: high_level, target: true, if_present: subtask}
-
-  low_level_execution:
-    weight: 0.40
-    messages:
-      # The low-level stream trains action flow on the generated or annotated subtask.
-      - {role: user, content: "${subtask}", stream: low_level, if_present: subtask}
-
-  memory_update:
-    # `active_at` densifies sparse boundaries while preserving the prior-memory/subtask mapping.
-    # Inference controls update timing through `subtask_change` events.
-    weight: 0.10
-    bindings:
-      prior_memory: "nth_prev(style=memory, offset=1)"
-      current_memory: "active_at(t, style=memory)"
-      completed_subtask: "nth_prev(style=subtask, offset=1)"
-    messages:
-      - {role: user, content: "${task}", stream: high_level}
-      - {role: assistant, content: "Previous memory: ${prior_memory}", stream: high_level, if_present: prior_memory}
-      - {role: user, content: "Completed subtask: ${completed_subtask}", stream: high_level, if_present: completed_subtask}
-      - {role: assistant, content: "${current_memory}", stream: high_level, target: true, if_present: current_memory}
-
-  user_interjection_response:
-    weight: 0.10
-    bindings:
-      interjection: "emitted_at(t, style=interjection)"
-      speech: "emitted_at(t, role=assistant, tool_name=say)"
-    messages:
-      - {role: user, content: "${task}", stream: high_level}
-      - {role: user, content: "${interjection}", stream: high_level, if_present: interjection}
-      # The assistant target is a `say` tool call flattened to a `<say>...</say>` marker.
-      - {role: assistant, stream: high_level, target: true, if_present: speech, tool_calls_from: speech}
-
-  # Each camera uses a separate VQA sub-recipe for view-specific binding.
-  ask_vqa_top:
-    weight: 0.075
-    route: vqa
-    bindings:
-      vqa_query: "emitted_at(t, style=vqa, role=user, camera=observation.images.front)"
-      vqa: "emitted_at(t, style=vqa, role=assistant, camera=observation.images.front)"
-    messages:
-      - role: user
-        stream: high_level
-        if_present: vqa_query
-        content:
-          - {type: image, feature: observation.images.front}
-          - {type: text, text: "${vqa_query}"}
-      - {role: assistant, content: "${vqa}", stream: high_level, target: true, if_present: vqa}
-
-  ask_vqa_wrist:
-    weight: 0.075
-    route: vqa
-    bindings:
-      vqa_query: "emitted_at(t, style=vqa, role=user, camera=observation.images.wrist)"
-      vqa: "emitted_at(t, style=vqa, role=assistant, camera=observation.images.wrist)"
-    messages:
-      - role: user
-        stream: high_level
-        if_present: vqa_query
-        content:
-          - {type: image, feature: observation.images.wrist}
-          - {type: text, text: "${vqa_query}"}
-      - {role: assistant, content: "${vqa}", stream: high_level, target: true, if_present: vqa}

src/lerobot/configs/train.py

@@ -18,7 +18,6 @@ import multiprocessing
 import os
 import tempfile
 from dataclasses import dataclass, field
-from enum import Enum
 from pathlib import Path
 from typing import Any
 
@@ -27,8 +26,6 @@ from huggingface_hub import hf_hub_download
 from huggingface_hub.errors import HfHubHTTPError
 
 from lerobot import envs
-from lerobot.configs.accelerator import AcceleratorConfig, ActivationCheckpointingMode
-from lerobot.configs.parallelism import ParallelismConfig
 from lerobot.optim import LRSchedulerConfig, OptimizerConfig
 from lerobot.utils.constants import PRETRAINED_MODEL_DIR
 from lerobot.utils.hub import HubMixin, find_latest_hub_checkpoint
@@ -42,34 +39,6 @@ from .rewards import RewardModelConfig
 TRAIN_CONFIG_NAME = "train_config.json"
 
 
-class CheckpointFormat(str, Enum):
-    """Model-artifact format inside training checkpoints.
-
-    Selects only the *model* artifact; the training_state layout is format-independent (the
-    optimizer channel is always DCP under sharded runs, safetensors+json otherwise).
-
-    - SAFETENSORS (default): a full `model.safetensors` — maximum compatibility, one gather per
-      save under sharding.
-    - DCP: sharded `pytorch_model_fsdp_0/*.distcp` only — fastest save/resume; convert with
-      `lerobot-convert-dcp` before distributing.
-    - SAFETENSORS_AND_DCP: both artifacts, written independently.
-    """
-
-    SAFETENSORS = "safetensors"
-    DCP = "dcp"
-    SAFETENSORS_AND_DCP = "safetensors_dcp"
-
-    @property
-    def wants_safetensors(self) -> bool:
-        """True when a full `model.safetensors` artifact should be written."""
-        return self in (CheckpointFormat.SAFETENSORS, CheckpointFormat.SAFETENSORS_AND_DCP)
-
-    @property
-    def wants_dcp(self) -> bool:
-        """True when sharded DCP model shards (`pytorch_model_fsdp_0/`) should be written."""
-        return self in (CheckpointFormat.DCP, CheckpointFormat.SAFETENSORS_AND_DCP)
-
-
 def _migrate_legacy_rabc_fields(config: dict[str, Any]) -> dict[str, Any] | None:
     """Return migrated payload for legacy RA-BC fields, or None when no migration is needed."""
     legacy_fields = (
@@ -152,16 +121,9 @@ class TrainPipelineConfig(HubMixin):
     # Checkpoint is saved every `save_freq` training iterations and after the last training step.
     # A non-positive value disables periodic saving, keeping only the final checkpoint.
     save_freq: int = 20_000
-    # Model-artifact format inside checkpoints; non-default values require a sharded run.
-    checkpoint_format: CheckpointFormat = CheckpointFormat.SAFETENSORS
     use_policy_training_preset: bool = True
     optimizer: OptimizerConfig | None = None
     scheduler: LRSchedulerConfig | None = None
-    # Process topology: dp_replicate / dp_shard (HSDP) and context-parallel degree placeholders.
-    parallelism: ParallelismConfig = field(default_factory=ParallelismConfig)
-    # Execution runtime handed to the Accelerator: mixed precision, gradient accumulation,
-    # FSDP/DDP tuning knobs, compile & activation-checkpointing placeholders.
-    accelerator: AcceleratorConfig = field(default_factory=AcceleratorConfig)
     eval: EvalConfig = field(default_factory=EvalConfig)
     wandb: WandBConfig = field(default_factory=WandBConfig)
     peft: PeftConfig | None = None
@@ -329,60 +291,6 @@ class TrainPipelineConfig(HubMixin):
         if self.save_checkpoint_to_hub and not (self.policy is not None and self.policy.repo_id):
             raise ValueError("save_checkpoint_to_hub requires --policy.repo_id.")
 
-        self._validate_distributed()
-
-    def _validate_distributed(self) -> None:
-        """Fail-fasts for the distributed-training scope.
-
-        Raises:
-            ValueError: If the config requests anything outside the verified scope: context
-                parallelism or CFG parallelism (reserved placeholders), the compile or
-                activation-checkpointing placeholders, a DCP checkpoint format on a
-                non-sharded run, or — under sharded training — fp16 mixed precision, PEFT,
-                reward-model training, in-training environment evaluation, or multi-optimizer
-                configs.
-        """
-        if self.parallelism.cp_size > 1:
-            raise ValueError(
-                "Context parallelism is not implemented yet: --parallelism.context_parallel.* "
-                "degrees must be 1 (reserved for the CP engine round)."
-            )
-        if self.parallelism.cfg_parallel != 1:
-            raise ValueError(
-                "CFG parallelism is inference-only and must be 1 for training "
-                "(cfg_parallel is reserved for the serving round)."
-            )
-        if self.accelerator.compile.enabled:
-            raise ValueError("--accelerator.compile is a placeholder and not wired yet.")
-        if self.accelerator.activation_checkpointing.mode is not ActivationCheckpointingMode.NONE:
-            raise ValueError("--accelerator.activation_checkpointing is a placeholder and not wired yet.")
-        if self.checkpoint_format is not CheckpointFormat.SAFETENSORS and not self.parallelism.is_sharded:
-            raise ValueError(
-                f"checkpoint_format={self.checkpoint_format.value} requires a sharded run "
-                "(--parallelism.dp_shard != 1); non-sharded checkpoints are always safetensors."
-            )
-        if self.parallelism.is_sharded:
-            if self.accelerator.mixed_precision == "fp16":
-                raise ValueError(
-                    "fp16 is not supported under sharded training (GradScaler over DTensor "
-                    "gradients is unverified); use bf16 or full precision."
-                )
-            if self.peft is not None:
-                raise ValueError("PEFT is not supported under sharded training yet.")
-            if self.is_reward_model_training:
-                raise ValueError(
-                    "Reward-model training is not supported under sharded training yet "
-                    "(reward models declare no FSDP wrap units and have no sharded save path)."
-                )
-            if self.env is not None and self.env_eval_freq > 0:
-                raise ValueError(
-                    "In-training environment evaluation is not supported under sharded training "
-                    "(a rank-0-only rollout of a sharded model deadlocks on collectives); set "
-                    "--env_eval_freq=0 and evaluate with lerobot-eval on saved checkpoints."
-                )
-            if self.optimizer is not None and self.optimizer.builds_multiple_optimizers:
-                raise ValueError("Multi-optimizer configs are not supported under sharded training.")
-
     @classmethod
     def __get_path_fields__(cls) -> list[str]:
         """Keys for draccus pretrained-path loading."""

src/lerobot/data_processing/sarm_annotations/subtask_annotation.py

@@ -76,7 +76,7 @@ import torch
 from pydantic import BaseModel, Field
 from transformers import AutoProcessor, Qwen3VLMoeForConditionalGeneration
 
-from lerobot.datasets import LeRobotDataset, resolve_episode_indices
+from lerobot.datasets import LeRobotDataset
 
 
 # Pydantic Models for SARM Subtask Annotation
@@ -1049,10 +1049,7 @@ def main():
     torch_dtype = {"bfloat16": torch.bfloat16, "float16": torch.float16, "float32": torch.float32}[args.dtype]
 
     # Determine episodes
-    resolved_episodes = resolve_episode_indices(args.episodes, dataset.meta.total_episodes)
-    episode_indices = (
-        resolved_episodes if resolved_episodes is not None else list(range(dataset.meta.total_episodes))
-    )
+    episode_indices = args.episodes or list(range(dataset.meta.total_episodes))
 
     existing_annotations = load_annotations_from_dataset(dataset.root, prefix="sparse")
     if args.skip_existing:

src/lerobot/datasets/__init__.py

@@ -52,7 +52,7 @@ from .pipeline_features import aggregate_pipeline_dataset_features, create_initi
 from .pyav_utils import check_video_encoder_parameters_pyav, detect_available_encoders_pyav
 from .sampler import EpisodeAwareSampler, compute_sampler_state
 from .streaming_dataset import StreamingLeRobotDataset
-from .utils import DEFAULT_EPISODES_PATH, create_lerobot_dataset_card, resolve_episode_indices
+from .utils import DEFAULT_EPISODES_PATH, create_lerobot_dataset_card
 from .video_utils import VideoEncodingManager
 
 # NOTE: Low-level I/O functions (cast_stats_to_numpy, get_parquet_file_size_in_mb, etc.)
@@ -97,7 +97,6 @@ __all__ = [
     "reencode_dataset",
     "remove_feature",
     "resolve_delta_timestamps",
-    "resolve_episode_indices",
     "safe_stop_image_writer",
     "split_dataset",
     "write_stats",

src/lerobot/datasets/aggregate.py

@@ -58,38 +58,12 @@ type ChunkFile = tuple[int, int]
 
 
 class IndexState(TypedDict):
-    """The current write cursor for a non-video (parquet) output stream during aggregation.
-
-    **Attributes**:
-        - **chunk** (`int`) -- The chunk index currently being written to.
-        - **file** (`int`) -- The file index, within `chunk`, currently being written to.
-        - **src_to_dst** (`dict[ChunkFile, ChunkFile]`, *optional*) -- Maps each source dataset's
-          `(chunk, file)` to the destination `(chunk, file)` its rows were merged into.
-    """
-
     chunk: int
     file: int
     src_to_dst: NotRequired[dict[ChunkFile, ChunkFile]]
 
 
 class VideoIndex(TypedDict):
-    """The current write cursor for a video output stream during aggregation.
-
-    **Attributes**:
-        - **chunk** (`int`) -- The chunk index currently being written to.
-        - **file** (`int`) -- The file index, within `chunk`, currently being written to.
-        - **latest_duration** (`float`) -- The duration, in seconds, appended to the current destination
-          file so far.
-        - **episode_duration** (`float`) -- The duration, in seconds, of the episode currently being
-          concatenated.
-        - **src_to_offset** (`dict[ChunkFile, float]`, *optional*) -- Maps each source `(chunk, file)` to
-          the time offset, in seconds, at which it was appended into its destination file.
-        - **src_to_dst** (`dict[ChunkFile, ChunkFile]`, *optional*) -- Maps each source `(chunk, file)` to
-          the destination `(chunk, file)` its video was concatenated into.
-        - **dst_file_durations** (`dict[ChunkFile, float]`, *optional*) -- The final total duration, in
-          seconds, of each completed destination file.
-    """
-
     chunk: int
     file: int
     latest_duration: float
@@ -106,7 +80,7 @@ def merge_video_feature_info_for_aggregate(all_metadata: list[LeRobotDatasetMeta
     """Create a merged video feature info dictionary for aggregation. The video encoder info is merged field-by-field: each key is kept only when every source agrees; otherwise that key is set to ``null`` (or ``{}`` for ``video.extra_options``) and a warning is logged.
 
     Args:
-        all_metadata (`list`): List of `LeRobotDatasetMetadata` objects to merge.
+        all_metadata: List of LeRobotDatasetMetadata objects to merge.
 
     Returns:
         dict: A dictionary of merged video feature info.
@@ -152,7 +126,7 @@ def validate_all_metadata(all_metadata: list[LeRobotDatasetMetadata]) -> tuple[i
     Video encoder info is not considered for validation but is merged during aggregation in ``merge_video_feature_info_for_aggregate``.
 
     Args:
-        all_metadata (`list`): List of `LeRobotDatasetMetadata` objects to validate.
+        all_metadata: List of LeRobotDatasetMetadata objects to validate.
 
     Returns:
         tuple: A tuple containing (fps, robot_type, features) from the first metadata.
@@ -161,6 +135,7 @@ def validate_all_metadata(all_metadata: list[LeRobotDatasetMetadata]) -> tuple[i
         ValueError: If any metadata has different fps, robot_type, or features
                    than the first metadata in the list.
     """
+
     fps = all_metadata[0].fps
     robot_type = all_metadata[0].robot_type
     features = all_metadata[0].features
@@ -189,13 +164,14 @@ def update_data_df(
     previously aggregated data in the destination dataset.
 
     Args:
-        df (`DataFrame`): DataFrame containing the data to be updated.
-        src_meta (`LeRobotDatasetMetadata`): Source dataset metadata.
-        dst_meta (`LeRobotDatasetMetadata`): Destination dataset metadata.
+        df: DataFrame containing the data to be updated.
+        src_meta: Source dataset metadata.
+        dst_meta: Destination dataset metadata.
 
     Returns:
         pd.DataFrame: Updated DataFrame with adjusted indices.
     """
+
     df["episode_index"] = df["episode_index"] + dst_meta.info.total_episodes
     df["index"] = df["index"] + dst_meta.info.total_frames
 
@@ -221,15 +197,16 @@ def update_meta_data(
     to correctly map source file indices to their destination locations.
 
     Args:
-        df (`DataFrame`): DataFrame containing the metadata to be updated.
-        dst_meta (`LeRobotDatasetMetadata`): Destination dataset metadata.
-        meta_idx (`IndexState`): Dictionary containing current metadata chunk and file indices.
-        data_idx (`IndexState`): Dictionary containing current data chunk and file indices.
-        videos_idx (`VideoIndexState`): Dictionary containing current video indices and timestamps.
+        df: DataFrame containing the metadata to be updated.
+        dst_meta: Destination dataset metadata.
+        meta_idx: Dictionary containing current metadata chunk and file indices.
+        data_idx: Dictionary containing current data chunk and file indices.
+        videos_idx: Dictionary containing current video indices and timestamps.
 
     Returns:
         pd.DataFrame: Updated DataFrame with adjusted indices and timestamps.
     """
+
     df["meta/episodes/chunk_index"] = df["meta/episodes/chunk_index"] + meta_idx["chunk"]
     df["meta/episodes/file_index"] = df["meta/episodes/file_index"] + meta_idx["file"]
 
@@ -390,21 +367,15 @@ def aggregate_datasets(
     4. Finalizing the aggregated dataset with proper statistics
 
     Args:
-        repo_ids (`list`): List of repository IDs for the datasets to aggregate.
-        aggr_repo_id (`str`): Repository ID for the aggregated output dataset.
-        roots (`list[pathlib.Path] | None`, *optional*): List of root paths for the source
-            datasets.
-        aggr_root (`pathlib.Path | None`, *optional*): Root path for the aggregated dataset.
-        data_files_size_in_mb (`int | None`, *optional*): Maximum size for data files in MB. Falls
-            back to `DEFAULT_DATA_FILE_SIZE_IN_MB` when not set.
-        video_files_size_in_mb (`int | None`, *optional*): Maximum size for video files in MB. Falls
-            back to `DEFAULT_VIDEO_FILE_SIZE_IN_MB` when not set.
-        chunk_size (`int | None`, *optional*): Maximum number of files per chunk. Falls back to
-            `DEFAULT_CHUNK_SIZE` when not set.
-        concatenate_videos (`bool`, *optional*, defaults to `True`): When `False`, keep one mp4 per
-            source file instead of packing into shards.
-        concatenate_data (`bool`, *optional*, defaults to `True`): When `False`, keep one parquet per
-            source file instead of packing into shards.
+        repo_ids: List of repository IDs for the datasets to aggregate.
+        aggr_repo_id: Repository ID for the aggregated output dataset.
+        roots: Optional list of root paths for the source datasets.
+        aggr_root: Optional root path for the aggregated dataset.
+        data_files_size_in_mb: Maximum size for data files in MB (defaults to DEFAULT_DATA_FILE_SIZE_IN_MB)
+        video_files_size_in_mb: Maximum size for video files in MB (defaults to DEFAULT_VIDEO_FILE_SIZE_IN_MB)
+        chunk_size: Maximum number of files per chunk (defaults to DEFAULT_CHUNK_SIZE)
+        concatenate_videos: When False, keep one mp4 per source file instead of packing into shards.
+        concatenate_data: When False, keep one parquet per source file instead of packing into shards.
     """
     logger.info("Start aggregate_datasets")
 
@@ -487,14 +458,12 @@ def aggregate_videos(
     Creates new video files when size limits are exceeded.
 
     Args:
-        src_meta (`LeRobotDatasetMetadata`): Source dataset metadata.
-        dst_meta (`LeRobotDatasetMetadata`): Destination dataset metadata.
-        videos_idx (`VideoIndexState`): Dictionary tracking video chunk and file indices.
-        video_files_size_in_mb (`float`): Maximum size for video files in MB.
-        chunk_size (`int`): Maximum number of files per chunk.
-        concatenate_videos (`bool`, *optional*, defaults to `True`): When `False`, keep one mp4 per
-            source file instead of packing into shards.
-
+        src_meta: Source dataset metadata.
+        dst_meta: Destination dataset metadata.
+        videos_idx: Dictionary tracking video chunk and file indices.
+        video_files_size_in_mb: Maximum size for video files in MB (defaults to DEFAULT_VIDEO_FILE_SIZE_IN_MB)
+        chunk_size: Maximum number of files per chunk (defaults to DEFAULT_CHUNK_SIZE)
+        concatenate_videos: When False, keep one mp4 per source file instead of packing into shards.
     Returns:
         dict: Updated videos_idx with current chunk and file indices.
     """
@@ -612,13 +581,12 @@ def aggregate_data(
     have multiple data files (e.g., from a previous merge operation).
 
     Args:
-        src_meta (`LeRobotDatasetMetadata`): Source dataset metadata.
-        dst_meta (`LeRobotDatasetMetadata`): Destination dataset metadata.
-        data_idx (`IndexState`): Dictionary tracking data chunk and file indices.
-        data_files_size_in_mb (`float`): Maximum size for data files in MB.
-        chunk_size (`int`): Maximum number of files per chunk.
-        concatenate_data (`bool`, *optional*, defaults to `True`): When `False`, keep one parquet per
-            source file instead of packing into shards.
+        src_meta: Source dataset metadata.
+        dst_meta: Destination dataset metadata.
+        data_idx: Dictionary tracking data chunk and file indices.
+        data_files_size_in_mb: Maximum size for data files in MB.
+        chunk_size: Maximum number of files per chunk.
+        concatenate_data: When False, keep one parquet per source file instead of packing into shards.
 
     Returns:
         dict: Updated data_idx with current chunk and file indices.
@@ -692,11 +660,11 @@ def aggregate_metadata(
     and writes them to the destination with proper file rotation.
 
     Args:
-        src_meta (`LeRobotDatasetMetadata`): Source dataset metadata.
-        dst_meta (`LeRobotDatasetMetadata`): Destination dataset metadata.
-        meta_idx (`IndexState`): Dictionary tracking metadata chunk and file indices.
-        data_idx (`IndexState`): Dictionary tracking data chunk and file indices.
-        videos_idx (`VideoIndexState`): Dictionary tracking video indices and timestamps.
+        src_meta: Source dataset metadata.
+        dst_meta: Destination dataset metadata.
+        meta_idx: Dictionary tracking metadata chunk and file indices.
+        data_idx: Dictionary tracking data chunk and file indices.
+        videos_idx: Dictionary tracking video indices and timestamps.
 
     Returns:
         dict: Updated meta_idx with current chunk and file indices.
@@ -759,22 +727,18 @@ def append_or_create_parquet_file(
     from becoming too large. Handles both regular parquet files and those containing images.
 
     Args:
-        df (`DataFrame`): DataFrame to write to the parquet file.
-        src_path (`Path`): Path to the source file, used for size estimation.
-        idx (`IndexState`): Dictionary containing current `chunk` and `file` indices.
-        max_mb (`float`): Maximum allowed file size in MB before rotation.
-        chunk_size (`int`): Maximum number of files per chunk before incrementing the chunk index.
-        default_path (`str`): Format string for generating file paths.
-        contains_images (`bool`, *optional*, defaults to `False`): Whether the data contains images
-            requiring special handling.
-        aggr_root (`pathlib.Path | None`, *optional*): Root path for the aggregated dataset.
-        hf_features (`datasets.features.features.Features | None`, *optional*): HuggingFace Features
-            schema used for proper image typing.
-        concatenate (`bool`, *optional*, defaults to `True`): When `False`, always rotate to a new
-            file instead of appending to the current one.
-        one_row_group_per_episode (`bool`, *optional*, defaults to `False`): Whether to emit one
-            parquet row group per episode. Set to `True` for data parquet files; left `False` for the
-            episodes-metadata parquet, which already has one row per episode.
+        df: DataFrame to write to the parquet file.
+        src_path: Path to the source file (used for size estimation).
+        idx: Dictionary containing current 'chunk' and 'file' indices.
+        max_mb: Maximum allowed file size in MB before rotation.
+        chunk_size: Maximum number of files per chunk before incrementing chunk index.
+        default_path: Format string for generating file paths.
+        contains_images: Whether the data contains images requiring special handling.
+        aggr_root: Root path for the aggregated dataset.
+        hf_features: Optional HuggingFace Features schema for proper image typing.
+        concatenate: When False, always rotate to a new file instead of appending to the current one.
+        one_row_group_per_episode: True for DATA parquet (emit one row group per episode); False for
+            the episodes-metadata parquet (already one row per episode).
 
     Returns:
         tuple: (updated_idx, (dst_chunk, dst_file)) where updated_idx is the index dict
@@ -838,8 +802,8 @@ def finalize_aggregation(
     aggregated statistics from all source datasets.
 
     Args:
-        aggr_meta (`LeRobotDatasetMetadata`): Aggregated dataset metadata.
-        all_metadata (`list`): List of all source dataset metadata objects.
+        aggr_meta: Aggregated dataset metadata.
+        all_metadata: List of all source dataset metadata objects.
     """
     logger.info("write tasks")
     write_tasks(aggr_meta.tasks, aggr_meta.root)

src/lerobot/datasets/compute_stats.py

@@ -28,22 +28,16 @@ DEFAULT_QUANTILES = [0.01, 0.10, 0.50, 0.90, 0.99]
 
 
 class RunningQuantileStats:
-    """Maintains running statistics for batches of vectors.
+    """
+    Maintains running statistics for batches of vectors, including mean,
+    standard deviation, min, max, and approximate quantiles.
 
-    Includes mean, standard deviation, min, max, and approximate quantiles. Statistics are computed per
-    feature dimension and updated incrementally
+    Statistics are computed per feature dimension and updated incrementally
     as new batches are observed. Quantiles are estimated using histograms,
     which adapt dynamically if the observed data range expands.
     """
 
     def __init__(self, quantile_list: list[float] | None = None, num_quantile_bins: int = 5000):
-        """Initialize empty running statistics.
-
-        Args:
-            quantile_list: Quantiles to track (e.g. `0.01` for the 1st percentile). Defaults to
-                `DEFAULT_QUANTILES` (1st, 10th, 50th, 90th, 99th percentiles).
-            num_quantile_bins: Number of histogram bins used to estimate quantiles.
-        """
         self._count = 0
         self._mean = None
         self._mean_of_squares = None
@@ -210,9 +204,8 @@ def estimate_num_samples(
     dataset_len: int, min_num_samples: int = 100, max_num_samples: int = 10_000, power: float = 0.75
 ) -> int:
     """Heuristic to estimate the number of samples based on dataset size.
-
-    The power controls the sample growth relative to dataset size. Lower the power for less number of
-    samples.
+    The power controls the sample growth relative to dataset size.
+    Lower the power for less number of samples.
 
     For default arguments, we have:
     - from 1 to ~500, num_samples=100
@@ -228,24 +221,11 @@ def estimate_num_samples(
 
 
 def sample_indices(data_len: int) -> list[int]:
-    """Return evenly-spaced indices into a sequence of length `data_len`, sized by `estimate_num_samples`."""
     num_samples = estimate_num_samples(data_len)
     return np.round(np.linspace(0, data_len - 1, num_samples)).astype(int).tolist()
 
 
 def auto_downsample_height_width(img: np.ndarray, target_size: int = 150, max_size_threshold: int = 300):
-    """Downsample a `(C, H, W)` image by integer striding if either dimension exceeds `max_size_threshold`.
-
-    Args:
-        img (`np.ndarray`): Input image in `(C, H, W)` layout to potentially downsample.
-        target_size (`int`, *optional*, defaults to 150): Approximate size, in pixels, that the
-            largest side should be reduced to.
-        max_size_threshold (`int`, *optional*, defaults to 300): Size, in pixels, above which the
-            largest side of `img` triggers downsampling.
-
-    Returns:
-        `img` unchanged, or strided down so its largest side is roughly `target_size`.
-    """
     _, height, width = img.shape
 
     if max(width, height) < max_size_threshold:
@@ -257,16 +237,6 @@ def auto_downsample_height_width(img: np.ndarray, target_size: int = 150, max_si
 
 
 def sample_images(image_paths: list[str]) -> np.ndarray:
-    """Load and downsample a sampled subset of `image_paths` into a single `uint8` array.
-
-    Args:
-        image_paths (`list[str]`): Paths of all images for the episode/feature, from which a
-            subset is sampled (see `sample_indices`).
-
-    Returns:
-        A `(N, C, H, W)` `uint8` array of the sampled, downsampled images (see
-        `auto_downsample_height_width`), where `N` is chosen by `sample_indices`.
-    """
     sampled_indices = sample_indices(len(image_paths))
 
     images = None
@@ -439,7 +409,6 @@ def _compute_basic_stats(
     Args:
         array: Reshaped array ready for statistics computation
         sample_count: Number of samples represented in the data
-        quantile_list: Quantiles to fill with the mean value. Defaults to `DEFAULT_QUANTILES`.
 
     Returns:
         Dictionary with basic statistics and quantiles set to mean values
@@ -478,14 +447,13 @@ def get_feature_stats(
     - Global: axis=None computes statistics over entire array
 
     Args:
-        array (`np.ndarray`): Input data array with a shape appropriate for the specified `axis`.
-        axis (`int | tuple[int, ...] | None`): Axis or axes along which to compute statistics:
-            `(0, 2, 3)` for image data (batch, channels, height, width), `0` or `(0,)` for
-            vector/tabular data (samples, features), `(1,)` to compute across features, or
-            `None` for global statistics over the entire array.
-        keepdims (`bool`): If `True`, reduced axes are kept as dimensions of size 1.
-        quantile_list (`list[float] | None`, *optional*): Quantiles to compute (e.g. `0.01` for
-            the 1st percentile). Defaults to `DEFAULT_QUANTILES` when not provided.
+        array: Input data array with shape appropriate for the specified axis
+        axis: Axis or axes along which to compute statistics
+            - (0, 2, 3): For image data (batch, channels, height, width)
+            - 0 or (0,): For vector/tabular data (samples, features)
+            - (1,): For computing across features
+            - None: For global statistics over entire array
+        keepdims: If True, reduced axes are kept as dimensions with size 1
 
     Returns:
         Dictionary containing:
@@ -528,13 +496,10 @@ def compute_episode_stats(
     - Strings: Skipped (no statistics computed)
 
     Args:
-        episode_data (`dict[str, list[str] | np.ndarray]`): Mapping from feature name to its data
-            for the episode: a list of file paths for `image`/`video` features, or a numpy array
-            for numerical features.
-        features (`dict`): Dataset feature metadata, keyed by feature name, describing each
-            feature's `dtype` and shape.
-        quantile_list (`list[float] | None`, *optional*): Quantiles to compute (e.g. `0.01` for
-            the 1st percentile). Defaults to `DEFAULT_QUANTILES` when not provided.
+        episode_data: Dictionary mapping feature names to data
+            - For images/videos: list of file paths
+            - For numerical data: numpy arrays
+        features: Dictionary describing each feature's dtype and shape
 
     Returns:
         Dictionary mapping feature names to their statistics dictionaries.
@@ -665,6 +630,7 @@ def aggregate_stats(stats_list: list[dict[str, dict]]) -> dict[str, dict[str, np
     - new_mean = (mean of all data, weighted by counts)
     - new_std = (std of all data)
     """
+
     _assert_type_and_shape(stats_list)
 
     data_keys = {key for stats in stats_list for key in stats}
@@ -724,16 +690,16 @@ def compute_relative_action_stats(
     statistics suitable for normalization.
 
     Args:
-        hf_dataset (`datasets.Dataset`): The underlying HuggingFace dataset, must expose
-            `"action"`, `"observation.state"`, and `"episode_index"` columns.
-        features (`dict`): Dataset feature metadata; must contain `"action"` with a `"shape"`
-            entry and optionally `"names"`.
-        chunk_size (`int`): Number of consecutive frames per action chunk.
-        exclude_joints (`list[str] | None`, *optional*): Joint names whose dimensions should
-            remain absolute instead of being converted to relative actions.
-        num_workers (`int`, *optional*, defaults to 0): Number of parallel threads used for
-            computation. Values `<= 1` run single-threaded; NumPy releases the GIL so threads
-            give real parallelism here.
+        hf_dataset: The underlying HuggingFace dataset with "action",
+            "observation.state", and "episode_index" columns.
+        features: Dataset feature metadata (must contain "action" with "shape"
+            and optionally "names").
+        chunk_size: Number of consecutive frames per action chunk.
+        exclude_joints: Joint names whose dimensions should remain absolute
+            (not converted to relative actions).
+        num_workers: Number of parallel threads for computation. Values ≤1
+            mean single-threaded. Numpy releases the GIL so threads give
+            real parallelism here.
 
     Returns:
         Statistics dict with keys "mean", "std", "min", "max", "q01", …, "q99".

src/lerobot/datasets/dataset_metadata.py

@@ -529,9 +529,9 @@ class LeRobotDatasetMetadata:
         return self.info.video_files_size_in_mb
 
     def get_task_index(self, task: str) -> int | None:
-        """Given a task in natural language, returns its task_index if the task already exists in the dataset.
-
-        Otherwise return None.
+        """
+        Given a task in natural language, returns its task_index if the task already exists in the dataset,
+        otherwise return None.
         """
         if task in self.tasks.index:
             return int(self.tasks.loc[task].task_index)
@@ -774,7 +774,6 @@ class LeRobotDatasetMetadata:
         }
 
     def __repr__(self):
-        """A short summary: repo ID, total episode/frame counts, and feature keys."""
         feature_keys = list(self.features)
         return (
             f"{self.__class__.__name__}({{\n"

src/lerobot/datasets/dataset_reader.py

@@ -39,7 +39,6 @@ from .io_utils import (
     hf_transform_to_torch,
     load_nested_dataset,
 )
-from .utils import resolve_episode_indices
 from .video_utils import decode_video_frames
 
 
@@ -84,7 +83,7 @@ class DatasetReader:
         """
         self._meta = meta
         self.root = root
-        self.episodes = resolve_episode_indices(episodes, meta.total_episodes)
+        self.episodes = episodes
         self._tolerance_s = tolerance_s
         self._video_backend = video_backend
         if image_transforms is not None and not callable(image_transforms):
@@ -164,34 +163,10 @@ class DatasetReader:
     def _load_hf_dataset(self) -> datasets.Dataset:
         """hf_dataset contains all the observations, states, actions, rewards, etc."""
         features = get_hf_features_from_features(self._meta.features)
-        self._validate_language_columns_declared(features)
         hf_dataset = load_nested_dataset(self.root / "data", features=features, episodes=self.episodes)
         hf_dataset.set_transform(hf_transform_to_torch)
         return hf_dataset
 
-    def _validate_language_columns_declared(self, features: datasets.Features) -> None:
-        """Require language columns stored in Parquet to be declared in metadata."""
-        # Leave empty datasets to fail through the normal loading path.
-        try:
-            sample = next((self.root / "data").glob("*/*.parquet"))
-        except StopIteration:
-            return
-
-        from pyarrow import parquet as _pq  # noqa: PLC0415
-
-        # LeRobot shards are schema-uniform, so one schema represents the dataset.
-        schema_names = set(_pq.read_schema(sample).names)
-        from .language import LANGUAGE_COLUMNS  # noqa: PLC0415
-
-        missing = sorted(set(LANGUAGE_COLUMNS) & schema_names - set(features))
-        if missing:
-            raise ValueError(
-                f"Dataset Parquet files contain language feature(s) missing from metadata: {missing}. "
-                "Metadata must describe the stored data; add the entries returned by "
-                "lerobot.datasets.language.language_feature_info() to meta/info.json['features'] "
-                "or rerun the annotation pipeline's metadata synchronization."
-            )
-
     def _check_cached_episodes_sufficient(self) -> bool:
         """Check if the cached dataset contains all requested episodes and their video files."""
         if self.hf_dataset is None or len(self.hf_dataset) == 0:
@@ -293,9 +268,7 @@ class DatasetReader:
         return result
 
     def _query_videos(self, query_timestamps: dict[str, list[float]], ep_idx: int) -> dict[str, torch.Tensor]:
-        """Decode the requested per-camera frame timestamps from `ep_idx`'s videos.
-
-        Note: When using data workers (e.g. DataLoader with num_workers>0), do not call this function
+        """Note: When using data workers (e.g. DataLoader with num_workers>0), do not call this function
         in the main process (e.g. by using a second Dataloader with num_workers=0). It will result in a
         Segmentation Fault.
         """

src/lerobot/datasets/dataset_tools.py

@@ -118,11 +118,10 @@ def delete_episodes(
     consistent with its own metadata.
 
     Args:
-        dataset (`LeRobotDataset`): The source LeRobotDataset.
-        episode_indices (`list`): List of episode indices to delete.
-        output_dir (`str | pathlib.Path | None`, *optional*): Root directory where the edited dataset
-            will be stored. If not specified, defaults to `$HF_LEROBOT_HOME/repo_id`.
-        repo_id (`str | None`, *optional*): Identifier for the edited dataset.
+        dataset: The source LeRobotDataset.
+        episode_indices: List of episode indices to delete.
+        output_dir: Root directory where the edited dataset will be stored. If not specified, defaults to $HF_LEROBOT_HOME/repo_id. Equivalent to new_root in EditDatasetConfig.
+        repo_id: Edited dataset identifier. Equivalent to new_repo_id in EditDatasetConfig.
     """
     if not episode_indices:
         raise ValueError("No episodes to delete")
@@ -186,11 +185,10 @@ def split_dataset(
     output split stays consistent with its own metadata.
 
     Args:
-        dataset (`LeRobotDataset`): The source LeRobotDataset to split.
-        splits (`dict`): Either a dict mapping split names to episode indices, or a dict mapping
-            split names to fractions (must sum to <= 1.0).
-        output_dir (`str | pathlib.Path | None`, *optional*): Root directory where the split
-            datasets will be stored. If not specified, defaults to `$HF_LEROBOT_HOME/repo_id`.
+        dataset: The source LeRobotDataset to split.
+        splits: Either a dict mapping split names to episode indices, or a dict mapping
+                split names to fractions (must sum to <= 1.0).
+        output_dir: Root directory where the split datasets will be stored. If not specified, defaults to $HF_LEROBOT_HOME/repo_id.
 
     Examples:
       Split by specific episodes
@@ -282,14 +280,11 @@ def merge_datasets(
     This is a wrapper around the aggregate_datasets functionality with a cleaner API.
 
     Args:
-        datasets (`list`): List of LeRobotDatasets to merge.
-        output_repo_id (`str`): Identifier for the merged dataset.
-        output_dir (`str | pathlib.Path | None`, *optional*): Root directory where the merged dataset
-            will be stored. If not specified, defaults to `$HF_LEROBOT_HOME/output_repo_id`.
-        concatenate_videos (`bool`, *optional*, defaults to `True`): When `False`, keep one mp4 per
-            source file instead of packing them into shards.
-        concatenate_data (`bool`, *optional*, defaults to `True`): When `False`, keep one parquet file
-            per source file instead of packing them into shards.
+        datasets: List of LeRobotDatasets to merge.
+        output_repo_id: Merged dataset identifier.
+        output_dir: Root directory where the merged dataset will be stored. If not specified, defaults to $HF_LEROBOT_HOME/output_repo_id.
+        concatenate_videos: When False, keep one mp4 per source file instead of packing into shards.
+        concatenate_data: When False, keep one parquet per source file instead of packing into shards.
     """
     if not datasets:
         raise ValueError("No datasets to merge")
@@ -332,14 +327,11 @@ def modify_features(
     regardless of how many features are being added or removed.
 
     Args:
-        dataset (`LeRobotDataset`): The source LeRobotDataset.
-        add_features (`dict[str, tuple[numpy.ndarray | torch.Tensor | collections.abc.Callable, dict]] | None`, *optional*):
-            Dict mapping feature names to `(feature_values, feature_info)` tuples.
-        remove_features (`str | list[str] | None`, *optional*): Feature name(s) to remove. Can be a
-            single string or a list.
-        output_dir (`str | pathlib.Path | None`, *optional*): Root directory where the edited dataset
-            will be stored. If not specified, defaults to `$HF_LEROBOT_HOME/repo_id`.
-        repo_id (`str | None`, *optional*): Identifier for the edited dataset.
+        dataset: The source LeRobotDataset.
+        add_features: Optional dict mapping feature names to (feature_values, feature_info) tuples.
+        remove_features: Optional feature name(s) to remove. Can be a single string or list.
+        output_dir: Root directory where the edited dataset will be stored. If not specified, defaults to $HF_LEROBOT_HOME/repo_id. Equivalent to new_root in EditDatasetConfig.
+        repo_id: Edited dataset identifier. Equivalent to new_repo_id in EditDatasetConfig.
 
     Returns:
         New dataset with features modified.
@@ -438,11 +430,10 @@ def add_features(
     copies the dataset once regardless of how many features are being added.
 
     Args:
-        dataset (`LeRobotDataset`): The source LeRobotDataset.
-        features (`dict`): Dictionary mapping feature names to `(feature_values, feature_info)` tuples.
-        output_dir (`str | pathlib.Path | None`, *optional*): Root directory where the edited dataset
-            will be stored. If not specified, defaults to `$HF_LEROBOT_HOME/repo_id`.
-        repo_id (`str | None`, *optional*): Identifier for the edited dataset.
+        dataset: The source LeRobotDataset.
+        features: Dictionary mapping feature names to (feature_values, feature_info) tuples.
+        output_dir: Root directory where the edited dataset will be stored. If not specified, defaults to $HF_LEROBOT_HOME/repo_id. Equivalent to new_root in EditDatasetConfig.
+        repo_id: Edited dataset identifier. Equivalent to new_repo_id in EditDatasetConfig.
 
     Returns:
         New dataset with all features added.
@@ -476,12 +467,10 @@ def remove_feature(
     """Remove features from a LeRobotDataset.
 
     Args:
-        dataset (`LeRobotDataset`): The source LeRobotDataset.
-        feature_names (`str | list[str]`): Name(s) of features to remove. Can be a single string or
-            a list.
-        output_dir (`str | pathlib.Path | None`, *optional*): Root directory where the edited dataset
-            will be stored. If not specified, defaults to `$HF_LEROBOT_HOME/repo_id`.
-        repo_id (`str | None`, *optional*): Identifier for the edited dataset.
+        dataset: The source LeRobotDataset.
+        feature_names: Name(s) of features to remove. Can be a single string or list.
+        output_dir: Root directory where the edited dataset will be stored. If not specified, defaults to $HF_LEROBOT_HOME/repo_id. Equivalent to new_root in EditDatasetConfig.
+        repo_id: Edited dataset identifier. Equivalent to new_repo_id in EditDatasetConfig.
 
     Returns:
         New dataset with features removed.
@@ -960,7 +949,7 @@ def _copy_and_reindex_episodes_metadata(
 
 
 def _write_parquet(df: pd.DataFrame, path: Path, meta: LeRobotDatasetMetadata) -> None:
-    """Write DataFrame to parquet.
+    """Write DataFrame to parquet
 
     This ensures images are properly embedded and the file can be loaded correctly by HF datasets.
     """
@@ -1468,14 +1457,13 @@ def modify_tasks(
     - meta/info.json (total_tasks)
 
     Args:
-        dataset (`LeRobotDataset`): The source LeRobotDataset to modify.
-        new_task (`str | None`, *optional*): Default task applied to any episode not covered by
-            `episode_tasks` or a matching `task_replacements` entry.
-        episode_tasks (`dict[int, str] | None`, *optional*): Dict mapping episode indices to task
-            strings. Takes precedence over both `task_replacements` and `new_task`.
-        task_replacements (`dict[str, str] | None`, *optional*): Dict mapping existing task strings to
-            new ones. Applied to episodes whose current task matches a key. Every key must be an
-            existing task.
+        dataset: The source LeRobotDataset to modify.
+        new_task: Default task applied to any episode not covered by `episode_tasks` or a
+            matching `task_replacements` entry.
+        episode_tasks: Optional dict mapping episode indices to task strings. Takes precedence
+            over both `task_replacements` and `new_task`.
+        task_replacements: Optional dict mapping existing task strings to new ones. Applied to
+            episodes whose current task matches a key. Every key must be an existing task.
 
     At least one of `new_task`, `episode_tasks`, or `task_replacements` must be provided.
 
@@ -1606,19 +1594,19 @@ def recompute_stats(
     """Recompute stats.json from scratch by iterating all episodes.
 
     Args:
-        dataset (`LeRobotDataset`): The LeRobotDataset to recompute stats for.
-        skip_image_video (`bool`, *optional*, defaults to `True`): If `True`, only recompute stats for
-            numeric features (action, state, etc.) and keep existing image/video stats unchanged.
-        relative_action (`bool`, *optional*, defaults to `False`): If `True`, compute action stats in
-            relative space by iterating all valid action chunks and subtracting the current state.
-            This matches the normalization distribution the model sees during training with
-            `use_relative_actions=True`.
-        relative_exclude_joints (`list[str] | None`, *optional*): Joint names to exclude from relative
-            conversion when `relative_action=True`. These dims keep absolute stats.
-        chunk_size (`int`, *optional*, defaults to 50): Action chunk size used for relative stats
-            computation. Should match `policy.chunk_size`. Only used when `relative_action=True`.
-        num_workers (`int`, *optional*, defaults to 0): Number of parallel threads for relative action
-            stats computation. Values <=1 mean single-threaded. Only used when `relative_action=True`.
+        dataset: The LeRobotDataset to recompute stats for.
+        skip_image_video: If True (default), only recompute stats for numeric features
+            (action, state, etc.) and keep existing image/video stats unchanged.
+        relative_action: If True, compute action stats in relative space by
+            iterating all valid action chunks and subtracting the current state.
+            This matches the normalization distribution the model sees during
+            training with ``use_relative_actions=True``.
+        relative_exclude_joints: Joint names to exclude from relative conversion when
+            relative_action=True. These dims keep absolute stats.
+        chunk_size: Action chunk size used for relative stats computation. Should match
+            ``policy.chunk_size``. Only used when ``relative_action=True``.
+        num_workers: Number of parallel threads for relative action stats computation.
+            Values ≤1 mean single-threaded. Only used when ``relative_action=True``.
 
     Returns:
         The same dataset with updated stats.
@@ -1721,22 +1709,24 @@ def convert_image_to_video_dataset(
     LeRobot dataset structure with videos stored in chunked MP4 files.
 
     Args:
-        dataset (`LeRobotDataset`): The source LeRobot dataset with images.
-        output_dir (`pathlib.Path | None`, *optional*): Root directory where the converted dataset will
-            be stored. When `None`, defaults to `$HF_LEROBOT_HOME/repo_id`.
-        repo_id (`str | None`, *optional*): Identifier for the converted dataset.
-        rgb_encoder (`lerobot.configs.video.RGBEncoderConfig | None`, *optional*): Video encoder settings
-            applied to RGB cameras. When `None`, `rgb_encoder_defaults` is used.
-        depth_encoder (`lerobot.configs.video.DepthEncoderConfig | None`, *optional*): Video encoder
-            settings applied to depth-map cameras, including the quantization parameters persisted to
-            the dataset metadata. When `None`, `depth_encoder_defaults` is used.
-        episode_indices (`list[int] | None`, *optional*): Episode indices to convert. When `None`, all
-            episodes are converted.
-        num_workers (`int`, *optional*, defaults to 4): Number of threads for parallel processing.
-        max_episodes_per_batch (`int | None`, *optional*): Maximum episodes per video batch, to bound
-            memory use. `None` means no limit.
-        max_frames_per_batch (`int | None`, *optional*): Maximum frames per video batch, to bound memory
-            use. `None` means no limit.
+        dataset: The source LeRobot dataset with images.
+        output_dir: Root directory where the converted dataset will be stored. When
+            ``None``, defaults to ``$HF_LEROBOT_HOME/repo_id``. Equivalent to
+            ``new_root`` in ``EditDatasetConfig``.
+        repo_id: Converted dataset identifier. Equivalent to ``new_repo_id`` in
+            ``EditDatasetConfig``.
+        rgb_encoder: Video encoder settings applied to RGB cameras. When ``None``,
+            :func:`~lerobot.configs.video.rgb_encoder_defaults` is used.
+        depth_encoder: Video encoder settings applied to depth-map cameras, including
+            the quantization parameters persisted to the dataset metadata. When
+            ``None``, :func:`~lerobot.configs.video.depth_encoder_defaults` is used.
+        episode_indices: Episode indices to convert. When ``None``, all episodes are
+            converted.
+        num_workers: Number of threads for parallel processing.
+        max_episodes_per_batch: Maximum episodes per video batch, to bound memory use.
+            ``None`` means no limit.
+        max_frames_per_batch: Maximum frames per video batch, to bound memory use.
+            ``None`` means no limit.
 
     Returns:
         A new :class:`LeRobotDataset` with images encoded as videos.
@@ -1976,17 +1966,18 @@ def reencode_dataset(
     Videos are re-encoded in-place and the video information in ``info.json`` is refreshed.
 
     Args:
-        dataset (`LeRobotDataset`): An existing :class:`LeRobotDataset` whose videos will be re-encoded.
-        rgb_encoder (`lerobot.configs.video.RGBEncoderConfig | None`, *optional*): Target encoder
-            configuration applied to every RGB video file. If `None`, re-encoding is skipped for RGB
-            videos.
-        depth_encoder (`lerobot.configs.video.DepthEncoderConfig | None`, *optional*): Target encoder
-            configuration applied to every depth video file. If `None`, re-encoding is skipped for depth
-            videos. Quantization parameters will not override the ones in the current dataset.
-        encoder_threads (`int | None`, *optional*): Per-encoder thread count forwarded to
-            `reencode_video`. `None` lets the codec decide.
-        num_workers (`int | None`, *optional*): Number of parallel processes. `None` or `0` means
-            sequential (no multiprocessing); `1+` spawns a `ProcessPoolExecutor`.
+        dataset: An existing :class:`LeRobotDataset` whose videos will be
+            re-encoded.
+        rgb_encoder: Target encoder configuration applied to every RGB video
+            file. If ``None``, re-encoding is skipped for RGB videos.
+        depth_encoder: Target encoder configuration applied to every depth video
+            file. If ``None``, re-encoding is skipped for depth videos.
+            Quantization parameters will not override the ones in the current dataset.
+        encoder_threads: Per-encoder thread count forwarded to
+            :func:`reencode_video`. ``None`` lets the codec decide.
+        num_workers: Number of parallel processes. ``None`` or ``0`` means
+            sequential (no multiprocessing); ``1+`` spawns a
+            :class:`~concurrent.futures.ProcessPoolExecutor`.
 
     Returns:
         The same :class:`LeRobotDataset` instance with its metadata updated

src/lerobot/datasets/dataset_writer.py

@@ -200,7 +200,8 @@ class DatasetWriter:
             self.image_writer.save_image(image=image, fpath=fpath, compress_level=compress_level)
 
     def add_frame(self, frame: dict) -> None:
-        """Add a single frame to the current episode buffer.
+        """
+        Add a single frame to the current episode buffer.
 
         Apart from images written to a temporary directory, nothing is written to disk
         until ``save_episode()`` is called.

src/lerobot/datasets/depth_utils.py

@@ -13,7 +13,9 @@
 # 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.
-"""Depth encoding/decoding helpers for :class:`DepthEncoderConfig`."""
+"""
+Depth encoding/decoding helpers for :class:`DepthEncoderConfig`.
+"""
 
 import math
 from typing import Literal
@@ -90,24 +92,13 @@ def quantize_depth(
     ``depth_min``, ``depth_max``, and ``shift`` are always in **metres**.
 
     Args:
-        depth (`numpy.ndarray[tuple[int, ...], numpy.dtype[numpy.uint16]] | numpy.ndarray[tuple[int, ...], numpy.dtype[numpy.float32]] | torch.Tensor`): Depth
-            map to quantize. A `torch.Tensor` is moved to CPU before conversion.
-        depth_min (`float`, *optional*, defaults to 0.01): Depth, in metres, mapped to quantum
-            `0`.
-        depth_max (`float`, *optional*, defaults to 10.0): Depth, in metres, mapped to quantum
-            `DEPTH_QMAX`.
-        shift (`float`, *optional*, defaults to 3.5): Depth shift, in metres, used in log mode.
-            Must satisfy `depth_min + shift > 0`.
-        use_log (`bool`, *optional*, defaults to `True`): If `True`, quantize in log space, which
-            allocates more quanta to near-range depth.
-        pix_fmt (`str`, *optional*, defaults to `"gray12le"`): Pixel format used to build the
-            `av.VideoFrame` when `video_backend="pyav"`.
-        video_backend (`str | None`, *optional*, defaults to `"pyav"`): Video backend used for
-            encoding. When `"pyav"`, returns an `av.VideoFrame`; otherwise returns the raw
-            `uint16` array.
-        input_unit (`Literal`, *optional*, defaults to `"auto"`): Input unit policy: `"auto"`
-            infers the unit from `depth`'s dtype, while `"mm"` or `"m"` force millimetres or
-            metres respectively.
+        depth: Depth map; ``torch.Tensor`` is moved to CPU for conversion.
+        depth_min: Depth (metres) at quantum ``0``.
+        depth_max: Depth (metres) at quantum :data:`DEPTH_QMAX`.
+        shift: Depth shift (metres); used in log mode. Must satisfy ``depth_min + shift > 0``.
+        use_log: If ``True`` (default), quantize in log space.
+        video_backend: Video backend to use for encoding. Defaults to "pyav".
+        input_unit: Input unit policy (``"auto"``, ``"mm"``, ``"m"``).
 
     Returns:
         ``numpy.ndarray``, ``dtype=uint16``, same shape as ``depth``, values in
@@ -183,28 +174,15 @@ def dequantize_depth(
     Output layout is determined by ``output_channel_last``.
 
     Args:
-        quantized (`numpy.ndarray[tuple[int, ...], numpy.dtype[numpy.uint16]] | av.video.frame.VideoFrame | torch.Tensor`): 12-bit
-            codes in `[0, DEPTH_QMAX]`, as a numpy array, `av.VideoFrame`, or `torch.Tensor`
-            (any integer or float dtype).
-        depth_min (`float`, *optional*, defaults to 0.01): Depth, in metres, mapped to quantum
-            `0`. Must match the value passed to `quantize_depth`.
-        depth_max (`float`, *optional*, defaults to 10.0): Depth, in metres, mapped to quantum
-            `DEPTH_QMAX`. Must match the value passed to `quantize_depth`.
-        shift (`float`, *optional*, defaults to 3.5): Depth shift, in metres, used in log mode.
-            Must match the value passed to `quantize_depth`.
-        use_log (`bool`, *optional*, defaults to `True`): If `True`, invert the log-space mapping
-            used by `quantize_depth`. Must match the encoding call.
-        pix_fmt (`str`, *optional*, defaults to `"gray12le"`): Pixel format used to extract the
-            plane data when `quantized` is an `av.VideoFrame`.
-        output_unit (`Literal`, *optional*, defaults to `"mm"`): `"mm"` returns `uint16`
-            millimetres, clipped to `[0, 65535]`, when returning a numpy array, or `float32`
-            millimetres when `output_tensor=True`. `"m"` returns `float32` metres in
-            `[depth_min, depth_max]`.
-        output_tensor (`bool`, *optional*, defaults to `True`): If `True`, return a
-            `torch.Tensor` instead of a numpy array.
-        output_channel_last (`bool`, *optional*, defaults to `False`): If `True`, add the
-            restored singleton channel dimension as the last axis instead of the third-to-last
-            axis.
+        quantized: 12-bit codes in ``[0, DEPTH_QMAX]``. ``np.ndarray``,
+            ``av.VideoFrame``, or ``torch.Tensor`` (any integer or float dtype).
+        depth_min, depth_max, shift, use_log: Same as :func:`quantize_depth` (metres).
+        pix_fmt: Pixel format used to extract the plane from an ``av.VideoFrame``.
+        output_unit: ``"mm"`` returns ``uint16`` millimetres (rint, clip
+            ``[0, 65535]``) when returning a numpy array, or ``float32`` mm when
+            ``output_tensor=True``. ``"m"`` returns ``float32`` metres in
+            ``[depth_min, depth_max]``.
+        output_tensor: If True, return a ``torch.Tensor`` instead of a numpy array.
 
     Returns:
         Depth map in the requested unit and dtype.

src/lerobot/datasets/factory.py

@@ -29,7 +29,6 @@ from .dataset_metadata import LeRobotDatasetMetadata
 from .lerobot_dataset import LeRobotDataset
 from .multi_dataset import MultiLeRobotDataset
 from .streaming_dataset import StreamingLeRobotDataset
-from .utils import resolve_episode_indices
 
 
 def resolve_delta_timestamps(
@@ -85,24 +84,14 @@ def make_dataset(cfg: TrainPipelineConfig) -> LeRobotDataset | MultiLeRobotDatas
 
     if isinstance(cfg.dataset.repo_id, str):
         ds_meta = LeRobotDatasetMetadata(
-            cfg.dataset.repo_id,
-            root=cfg.dataset.root,
-            revision=cfg.dataset.revision,
-            repo_type=cfg.dataset.repo_type,
+            cfg.dataset.repo_id, root=cfg.dataset.root, revision=cfg.dataset.revision
         )
         delta_timestamps = resolve_delta_timestamps(cfg.trainable_config, ds_meta)
-        episodes = resolve_episode_indices(
-            cfg.dataset.episodes, ds_meta.total_episodes, cfg.dataset.exclude_episodes
-        )
         if not cfg.dataset.streaming:
-            if cfg.dataset.repo_type == "bucket":
-                raise ValueError(
-                    "repo_type='bucket' is streaming-only: set dataset.streaming=true to train from an HF Storage Bucket."
-                )
             dataset = LeRobotDataset(
                 cfg.dataset.repo_id,
                 root=cfg.dataset.root,
-                episodes=episodes,
+                episodes=cfg.dataset.episodes,
                 delta_timestamps=delta_timestamps,
                 image_transforms=image_transforms,
                 revision=cfg.dataset.revision,
@@ -115,14 +104,13 @@ def make_dataset(cfg: TrainPipelineConfig) -> LeRobotDataset | MultiLeRobotDatas
             dataset = StreamingLeRobotDataset(
                 cfg.dataset.repo_id,
                 root=cfg.dataset.root,
-                episodes=episodes,
+                episodes=cfg.dataset.episodes,
                 delta_timestamps=delta_timestamps,
                 image_transforms=image_transforms,
                 revision=cfg.dataset.revision,
                 max_num_shards=cfg.num_workers,
                 tolerance_s=cfg.tolerance_s,
                 return_uint8=True,
-                repo_type=cfg.dataset.repo_type,
             )
     else:
         raise NotImplementedError("The MultiLeRobotDataset isn't supported for now.")

src/lerobot/datasets/feature_utils.py

@@ -101,10 +101,10 @@ def create_empty_dataset_info(
         fps (int): The frames per second of the data.
         features (dict): The LeRobot features dictionary for the dataset.
         use_videos (bool): Whether the dataset will store videos.
-        robot_type (str | None, *optional*): The type of robot used, if any.
-        chunks_size (int | None, *optional*): Max files per chunk directory. Defaults to ``DEFAULT_CHUNK_SIZE``.
-        data_files_size_in_mb (int | None, *optional*): Max parquet file size in MB. Defaults to ``DEFAULT_DATA_FILE_SIZE_IN_MB``.
-        video_files_size_in_mb (int | None, *optional*): Max video file size in MB. Defaults to ``DEFAULT_VIDEO_FILE_SIZE_IN_MB``.
+        robot_type (str | None): The type of robot used, if any.
+        chunks_size (int | None): Max files per chunk directory. Defaults to ``DEFAULT_CHUNK_SIZE``.
+        data_files_size_in_mb (int | None): Max parquet file size in MB. Defaults to ``DEFAULT_DATA_FILE_SIZE_IN_MB``.
+        video_files_size_in_mb (int | None): Max video file size in MB. Defaults to ``DEFAULT_VIDEO_FILE_SIZE_IN_MB``.
 
     Returns:
         DatasetInfo: A typed dataset information object with initial metadata.
@@ -170,7 +170,7 @@ def check_delta_timestamps(
             deltas in seconds.
         fps (int): The frames per second of the dataset.
         tolerance_s (float): The allowed tolerance in seconds.
-        raise_value_error (bool, *optional*, defaults to `True`): If True, raises an error on failure.
+        raise_value_error (bool): If True, raises an error on failure.
 
     Returns:
         bool: True if all deltas are valid, False otherwise.
@@ -219,18 +219,6 @@ def get_delta_indices(delta_timestamps: dict[str, list[float]], fps: int) -> dic
 
 
 def validate_frame(frame: dict, features: dict) -> None:
-    """Check that `frame` has a `"task"` key and matches `features` (minus auto-populated defaults).
-
-    Args:
-        frame (`dict`): The frame to validate, mapping feature names to their values, as passed by the
-            caller to `add_frame`.
-        features (`dict`): The dataset's feature specification, mapping feature names to their dtype and
-            shape metadata.
-
-    Raises:
-        ValueError: If `frame` is missing `"task"`, or has missing/extra features, or a feature's dtype
-            or shape doesn't match its definition in `features`.
-    """
     # DEFAULT_FEATURES (timestamp, frame_index, episode_index, index, task_index) are
     # auto-populated by the recording pipeline (add_frame / save_episode) and must not
     # be supplied by the caller. Excluding them here means any frame dict that contains
@@ -287,7 +275,7 @@ def validate_feature_dtype_and_shape(
     Args:
         name (str): The name of the feature.
         feature (dict): The feature specification from the LeRobot features dictionary.
-        value (`numpy.ndarray | PIL.Image.Image | str`): The value of the feature to validate.
+        value: The value of the feature to validate.
 
     Returns:
         str: An error message if validation fails, otherwise an empty string.
@@ -349,7 +337,7 @@ def validate_feature_image_or_video(
     Args:
         name (str): The name of the feature.
         expected_shape (list[str]): The expected shape, e.g. (C, H, W) or (H, W, C).
-        value (`numpy.ndarray | PIL.Image.Image`): The image or video frame data to validate.
+        value: The image data to validate.
 
     Returns:
         str: An error message if validation fails, otherwise an empty string.
@@ -395,8 +383,7 @@ def validate_feature_language(name: str, value) -> str:
 
     Args:
         name (str): The name of the feature.
-        value (`Any`): The value supplied for the language feature. Only checked for being `None`; any
-            other value is dropped with a warning.
+        value: The value to validate.
 
     Returns:
         str: Always an empty string — language values are non-fatal.

src/lerobot/datasets/image_writer.py

@@ -27,10 +27,7 @@ logger = logging.getLogger(__name__)
 
 
 def safe_stop_image_writer(func):
-    """Decorator: on an exception from `func`, stop the `dataset` kwarg's image writer before re-raising."""
-
     def wrapper(*args, **kwargs):
-        """Call `func`; on any exception, stop `kwargs["dataset"].writer.image_writer` before re-raising."""
         try:
             return func(*args, **kwargs)
         except BaseException:
@@ -129,7 +126,8 @@ def save_kwargs_for_path(fpath: Path, compress_level: int) -> dict:
 
 
 def write_image(image: np.ndarray | PIL.Image.Image, fpath: Path, compress_level: int = 1):
-    """Saves a NumPy array or PIL Image to a file.
+    """
+    Saves a NumPy array or PIL Image to a file.
 
     This function handles both NumPy arrays and PIL Image objects, converting
     the former to a PIL Image before saving. It includes error handling for
@@ -140,7 +138,7 @@ def write_image(image: np.ndarray | PIL.Image.Image, fpath: Path, compress_level
     Args:
         image (np.ndarray | PIL.Image.Image): The image data to save.
         fpath (Path): The destination file path for the image.
-        compress_level (int, optional, *optional*, defaults to 1): The compression level for the saved
+        compress_level (int, optional): The compression level for the saved
             image, as used by PIL.Image.save(). Defaults to 1.
             Refer to: https://github.com/huggingface/lerobot/pull/2135
             for more details on the default value rationale.
@@ -165,7 +163,6 @@ def write_image(image: np.ndarray | PIL.Image.Image, fpath: Path, compress_level
 
 
 def worker_thread_loop(queue: queue.Queue):
-    """Pop `(image_array, fpath, compress_level)` items from `queue` and write each until a `None` sentinel."""
     while True:
         item = queue.get()
         if item is None:
@@ -177,7 +174,6 @@ def worker_thread_loop(queue: queue.Queue):
 
 
 def worker_process(queue: queue.Queue, num_threads: int):
-    """Run `num_threads` `worker_thread_loop` threads against `queue` and block until they all exit."""
     threads = []
     for _ in range(num_threads):
         t = threading.Thread(target=worker_thread_loop, args=(queue,))
@@ -189,9 +185,9 @@ def worker_process(queue: queue.Queue, num_threads: int):
 
 
 class AsyncImageWriter:
-    """This class abstracts away the initialisation of processes or/and threads.
-
-    It saves images on disk asynchronously, which is critical to control a robot and record data
+    """
+    This class abstract away the initialisation of processes or/and threads to
+    save images on disk asynchronously, which is critical to control a robot and record data
     at a high frame rate.
 
     When `num_processes=0`, it creates a threads pool of size `num_threads`.
@@ -204,15 +200,6 @@ class AsyncImageWriter:
     """
 
     def __init__(self, num_processes: int = 0, num_threads: int = 1):
-        """Start the thread or process pool.
-
-        Args:
-            num_processes: Number of worker subprocesses. `0` uses threads only (in this process).
-            num_threads: Number of writer threads per process (or in this process, if `num_processes=0`).
-
-        Raises:
-            ValueError: If both `num_threads` and `num_processes` are non-positive.
-        """
         self.num_processes = num_processes
         self.num_threads = num_threads
         self.queue = None
@@ -243,18 +230,15 @@ class AsyncImageWriter:
     def save_image(
         self, image: torch.Tensor | np.ndarray | PIL.Image.Image, fpath: Path, compress_level: int = 1
     ):
-        """Enqueue `image` to be written to `fpath` asynchronously; returns immediately."""
         if isinstance(image, torch.Tensor):
             # Convert tensor to numpy array to minimize main process time
             image = image.cpu().numpy()
         self.queue.put((image, fpath, compress_level))
 
     def wait_until_done(self):
-        """Block until every enqueued image has been written to disk."""
         self.queue.join()
 
     def stop(self):
-        """Signal all worker threads/processes to exit and wait for them to join. No-op if already stopped."""
         if self._stopped:
             return
 

src/lerobot/datasets/io_utils.py

@@ -46,7 +46,6 @@ from .utils import (
 
 
 def get_parquet_file_size_in_mb(parquet_path: str | Path) -> float:
-    """Return the uncompressed size, in megabytes, of a parquet file's column data (from its metadata)."""
     metadata = pq.read_metadata(parquet_path)
     total_uncompressed_size = 0
     for row_group in range(metadata.num_row_groups):
@@ -58,24 +57,20 @@ def get_parquet_file_size_in_mb(parquet_path: str | Path) -> float:
 
 
 def get_hf_dataset_size_in_mb(hf_ds: Dataset) -> int:
-    """Return the in-memory (Arrow buffer) size of a Hugging Face `Dataset`, in megabytes."""
     return hf_ds.data.nbytes // (1024**2)
 
 
 def load_nested_dataset(
     pq_dir: Path, features: datasets.Features | None = None, episodes: list[int] | None = None
 ) -> Dataset:
-    """Find parquet files in provided directory {pq_dir}/chunk-xxx/file-xxx.parquet.
-
-    Convert parquet files to pyarrow memory mapped in a cache folder for efficient RAM usage, then
-    concatenate all pyarrow references to return HF Dataset format.
+    """Find parquet files in provided directory {pq_dir}/chunk-xxx/file-xxx.parquet
+    Convert parquet files to pyarrow memory mapped in a cache folder for efficient RAM usage
+    Concatenate all pyarrow references to return HF Dataset format
 
     Args:
-        pq_dir (`Path`): Directory containing parquet files.
-        features (`datasets.features.features.Features | None`, *optional*): Features schema used to ensure
-            consistent loading of complex types like images.
-        episodes (`list[int] | None`, *optional*): List of episode indices to filter. Uses PyArrow
-            predicate pushdown for efficiency.
+        pq_dir: Directory containing parquet files
+        features: Optional features schema to ensure consistent loading of complex types like images
+        episodes: Optional list of episode indices to filter. Uses PyArrow predicate pushdown for efficiency.
     """
     paths = sorted(pq_dir.glob("*/*.parquet"))
     if len(paths) == 0:
@@ -88,7 +83,6 @@ def load_nested_dataset(
 
 
 def get_parquet_num_frames(parquet_path: str | Path) -> int:
-    """Return the number of rows in a parquet file, read from its metadata (no data is loaded)."""
     metadata = pq.read_metadata(parquet_path)
     return metadata.num_rows
 
@@ -124,7 +118,6 @@ def embed_images(dataset: datasets.Dataset) -> datasets.Dataset:
 
 
 def write_info(info: DatasetInfo, local_dir: Path) -> None:
-    """Write dataset info metadata to its standard file path (the inverse of `load_info`)."""
     write_json(info.to_dict(), local_dir / INFO_PATH)
 
 
@@ -183,14 +176,12 @@ def load_stats(local_dir: Path) -> dict[str, dict[str, np.ndarray]] | None:
 
 
 def write_tasks(tasks: pandas.DataFrame, local_dir: Path) -> None:
-    """Write the task-prompt table to its standard parquet file path (the inverse of `load_tasks`)."""
     path = local_dir / DEFAULT_TASKS_PATH
     path.parent.mkdir(parents=True, exist_ok=True)
     tasks.to_parquet(path)
 
 
 def load_tasks(local_dir: Path) -> pandas.DataFrame:
-    """Load the task-prompt table from its standard file path, indexed by task string."""
     tasks = pd.read_parquet(local_dir / DEFAULT_TASKS_PATH)
     tasks.index.name = "task"
     return tasks
@@ -198,13 +189,12 @@ def load_tasks(local_dir: Path) -> pandas.DataFrame:
 
 def write_episodes(episodes: Dataset, local_dir: Path) -> None:
     """Write episode metadata to a parquet file in the LeRobot v3.0 format.
-
     This function writes episode-level metadata to a single parquet file.
     Used primarily during dataset conversion (v2.1 → v3.0) and in test fixtures.
 
     Args:
-        episodes (`Dataset`): Hugging Face `Dataset` containing the episode metadata.
-        local_dir (`Path`): Root directory where the dataset is stored.
+        episodes: HuggingFace Dataset containing episode metadata
+        local_dir: Root directory where the dataset will be stored
     """
     episode_size_mb = get_hf_dataset_size_in_mb(episodes)
     if episode_size_mb > DEFAULT_DATA_FILE_SIZE_IN_MB:
@@ -220,7 +210,6 @@ def write_episodes(episodes: Dataset, local_dir: Path) -> None:
 
 
 def load_episodes(local_dir: Path) -> datasets.Dataset:
-    """Load episode metadata, excluding per-episode `stats/*` columns (for faster access to the rest)."""
     episodes = load_nested_dataset(local_dir / EPISODES_DIR)
     # Select episode features/columns containing references to episode data and videos
     # (e.g. tasks, dataset_from_index, dataset_to_index, data/chunk_index, data/file_index, etc.)
@@ -236,9 +225,9 @@ def load_image_as_numpy(
 
     Args:
         fpath (str | Path): Path to the image file.
-        dtype (np.dtype, *optional*, defaults to `float32`): The desired data type of the output array. If floating,
+        dtype (np.dtype): The desired data type of the output array. If floating,
             pixels are scaled to [0, 1]. Only used for RGB images.
-        channel_first (bool, *optional*, defaults to `True`): If True, converts the image to (C, H, W) format.
+        channel_first (bool): If True, converts the image to (C, H, W) format.
             Otherwise, it remains in (H, W, C) format.
 
     Returns:

src/lerobot/datasets/language_render.py

@@ -162,32 +162,14 @@ def render_sample(
     task: str | None = None,
     dataset_ctx: Any | None = None,
 ) -> RenderedMessages | None:
-    """Render recipe-defined messages and supervision for one dataset sample.
+    """Render the chat-style messages for a single dataset sample.
 
-    Resolves bindings against ``persistent`` and ``events`` at frame timestamp
-    ``t``. Blend recipes first route matching sparse VQA annotations, then use
-    deterministic weighted selection for the remaining samples. Returns
-    ``None`` when the selected recipe provides no text or low-level action
-    supervision for this sample.
+    Resolves the recipe's bindings against ``persistent`` and ``events`` rows
+    at frame timestamp ``t``, then expands the recipe's message templates.
+    Returns ``None`` if the resolved sample contains no target message.
     """
     persistent_rows = _normalize_rows(persistent or [])
     event_rows = _normalize_rows(events or [])
-
-    # Route sparse VQA frames to a matching view-specific component before weighted selection.
-    # This avoids dropping annotated frames or selecting VQA without annotations.
-    if recipe.blend is not None:
-        vqa_rendered = _render_vqa_if_present(
-            recipe,
-            persistent=persistent_rows,
-            events=event_rows,
-            t=t,
-            sample_idx=sample_idx,
-            task=task,
-            dataset_ctx=dataset_ctx,
-        )
-        if vqa_rendered is not None:
-            return vqa_rendered
-
     selected_recipe = _select_recipe(recipe, sample_idx)
     bindings = _resolve_bindings(
         selected_recipe,
@@ -201,58 +183,6 @@ def render_sample(
     return _render_message_recipe(selected_recipe, bindings)
 
 
-def _render_vqa_if_present(
-    recipe: TrainingRecipe,
-    *,
-    persistent: Sequence[LanguageRow],
-    events: Sequence[LanguageRow],
-    t: float,
-    sample_idx: int,
-    task: str | None,
-    dataset_ctx: Any | None,
-) -> RenderedMessages | None:
-    """Render a matching VQA component, or return ``None`` for normal selection.
-
-    Multiple matching views are selected deterministically by relative weight.
-    """
-    if recipe.blend is None:
-        return None
-    renderable: list[tuple[float, RenderedMessages]] = []
-    for component in recipe.blend.values():
-        if component.route != "vqa":
-            continue
-        bindings = _resolve_bindings(
-            component,
-            persistent=persistent,
-            events=events,
-            t=t,
-            sample_idx=sample_idx,
-            task=task,
-            dataset_ctx=dataset_ctx,
-        )
-        rendered = _render_message_recipe(component, bindings)
-        if rendered is not None:
-            if component.weight is None:
-                raise ValueError("Routed VQA blend components must define a weight.")
-            renderable.append((component.weight, rendered))
-
-    if not renderable:
-        return None
-    if len(renderable) == 1:
-        return renderable[0][1]
-
-    # Choose among matching cameras by their validated positive relative weights.
-    total = sum(weight for weight, _ in renderable)
-    digest = hashlib.blake2b(f"vqa:{sample_idx}".encode(), digest_size=8).digest()
-    draw = int.from_bytes(digest, "big") / 2**64 * total
-    cumulative = 0.0
-    for weight, rendered in renderable:
-        cumulative += weight
-        if draw < cumulative:
-            return rendered
-    return renderable[-1][1]
-
-
 def _select_recipe(recipe: TrainingRecipe, sample_idx: int) -> TrainingRecipe:
     """Pick a deterministic blend component for ``sample_idx`` (or return ``recipe``)."""
     if recipe.blend is None:
@@ -271,8 +201,7 @@ def _select_recipe(recipe: TrainingRecipe, sample_idx: int) -> TrainingRecipe:
         cumulative += component.weight or 0.0
         if draw < cumulative:
             return component
-    if last_component is None:
-        raise ValueError("Blend recipes must contain at least one component.")
+    assert last_component is not None
     return last_component
 
 
@@ -392,8 +321,7 @@ def _render_message_recipe(
     bindings: dict[str, LanguageRow | str | None],
 ) -> RenderedMessages | None:
     """Expand ``recipe.messages`` into rendered chat messages using ``bindings``."""
-    if recipe.messages is None:
-        raise ValueError("Cannot render a blend recipe as a message recipe.")
+    assert recipe.messages is not None
     messages: list[dict[str, Any]] = []
     streams: list[str | None] = []
     target_indices: list[int] = []
@@ -418,9 +346,7 @@ def _render_message_recipe(
         if turn.target:
             target_indices.append(message_idx)
 
-    # Keep samples with either text targets or low-level action supervision.
-    has_low_level = any(stream == "low_level" for stream in streams)
-    if not target_indices and not has_low_level:
+    if not target_indices:
         return None
 
     rendered = {
@@ -477,12 +403,14 @@ def _validate_rendered(rendered: RenderedMessages) -> None:
 
     if len(streams) != len(messages):
         raise ValueError("message_streams must be aligned with messages.")
-    # Require text or low-level action supervision.
-    if not target_indices and not any(s == "low_level" for s in streams):
-        raise ValueError("Rendered samples must contain a target message or a low_level-stream message.")
+    if not target_indices:
+        raise ValueError("Rendered samples must contain at least one target message.")
     for idx in target_indices:
         if idx < 0 or idx >= len(messages):
             raise ValueError(f"Target message index {idx} is out of bounds.")
+    # ``stream`` is enforced non-None at MessageTurn construction time
+    # (see ``MessageTurn.__post_init__``), so a missing stream here would
+    # mean the dataclass invariant was bypassed; no need to re-check.
 
 
 def _nth_relative(

src/lerobot/datasets/lerobot_dataset.py

@@ -44,14 +44,6 @@ logger = logging.getLogger(__name__)
 
 
 class LeRobotDataset(torch.utils.data.Dataset):
-    """A PyTorch `Dataset` over episodic robot data: per-frame state/action tensors, optional videos.
-
-    Backed by parquet files (`data/`) for tabular observation/action/reward data, optional video files
-    (`videos/`) for image observations, and a `meta/` directory holding `info.json` (shapes, keys, fps),
-    `stats.json` (normalization statistics), and per-episode metadata. See `__init__`'s docstring for the
-    on-disk layout, and `create()` for building a new (empty) dataset from scratch.
-    """
-
     def __init__(
         self,
         repo_id: str,
@@ -76,7 +68,8 @@ class LeRobotDataset(torch.utils.data.Dataset):
         *,
         token: str | bool | None = None,
     ):
-        """2 modes are available for instantiating this class, depending on 2 different use cases.
+        """
+        2 modes are available for instantiating this class, depending on 2 different use cases:
 
         1. Your dataset already exists:
             - On your local disk in the 'root' folder. This is typically the case when you recorded your
@@ -175,9 +168,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
                 conversion. This works for both image-backed and video-backed observations and can later be
                 updated with `set_image_transforms()` or cleared with `clear_image_transforms()`.
                 Defaults to None.
-            delta_timestamps (dict[list[float]] | None, optional): Per-feature timestamp offsets (in
-                seconds, relative to a frame's own timestamp) of additional frames to return alongside it.
-                Defaults to None.
+            delta_timestamps (dict[list[float]] | None, optional): _description_. Defaults to None.
             tolerance_s (float, optional): Tolerance in seconds used to ensure data timestamps are actually in
                 sync with the fps value. It is used at the init of the dataset to make sure that each
                 timestamps is separated to the next by 1/fps +/- tolerance_s. This also applies to frames
@@ -194,11 +185,6 @@ class LeRobotDataset(torch.utils.data.Dataset):
                 True.
             video_backend (str | None, optional): Video backend to use for decoding videos. Defaults to torchcodec when available int the platform; otherwise, defaults to 'pyav'.
                 You can also use the 'pyav' decoder used by Torchvision, which used to be the default option, or 'video_reader' which is another decoder of Torchvision.
-            return_uint8 (bool, optional): For RGB videos, whether to return raw uint8 frames instead of
-                the default float32 frames normalized to [0, 1]. Defaults to False.
-            depth_output_unit (str, optional): Physical unit depth maps are dequantized to at load time:
-                "mm" (millimeters) or "m" (metres). Has no effect on datasets without depth cameras.
-                Defaults to "mm".
             batch_encoding_size (int, optional): Number of episodes to accumulate before batch encoding videos.
                 Set to 1 for immediate encoding (default), or higher for batched encoding. Defaults to 1.
             rgb_encoder (RGBEncoderConfig | None, optional): Video encoder settings for cameras
@@ -409,7 +395,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
 
     @property
     def hf_dataset(self) -> datasets.Dataset:
-        """The underlying Hugging Face Dataset object."""
+        """The underlying Hugging Face Dataset object"""
         self.reader = self._ensure_reader()
         if self.reader.hf_dataset is None:
             self.reader.load_and_activate()
@@ -554,7 +540,6 @@ class LeRobotDataset(torch.utils.data.Dataset):
         return self.hf_dataset[idx]
 
     def __repr__(self):
-        """A short summary: repo ID, selected episode/sample counts, and feature keys."""
         feature_keys = list(self.features)
         return (
             f"{self.__class__.__name__}({{\n"
@@ -753,10 +738,6 @@ class LeRobotDataset(torch.utils.data.Dataset):
                 during capture instead of writing images first.
             encoder_queue_maxsize: Max buffered frames per camera when using
                 streaming encoding.
-            video_files_size_in_mb: Max video file size in MB. Defaults to
-                ``DEFAULT_VIDEO_FILE_SIZE_IN_MB``.
-            data_files_size_in_mb: Max parquet file size in MB. Defaults to
-                ``DEFAULT_DATA_FILE_SIZE_IN_MB``.
 
         Returns:
             A new :class:`LeRobotDataset` in write mode.

src/lerobot/datasets/multi_dataset.py

@@ -51,20 +51,6 @@ class MultiLeRobotDataset(torch.utils.data.Dataset):
         *,
         token: str | bool | None = None,
     ):
-        """Construct a `LeRobotDataset` for each `repo_id` and concatenate them.
-
-        Args:
-            repo_ids: The Hub repo IDs (or local dataset names, if `root` is set) to load.
-            root: Root directory containing the underlying datasets. Defaults to `$HF_LEROBOT_HOME`.
-            episodes: Optional mapping from `repo_id` to the episode indices to load from it.
-            image_transforms: Transform applied to visual observations in each underlying dataset.
-            delta_timestamps: Passed through to each underlying `LeRobotDataset`.
-            tolerances_s: Optional mapping from `repo_id` to its timestamp tolerance, in seconds. Defaults
-                to `1e-4` for every dataset.
-            download_videos: Whether to download video files for each underlying dataset.
-            video_backend: The video decoding backend to use.
-            token: Hugging Face Hub authentication token.
-        """
         super().__init__()
         self.repo_ids = repo_ids
         self.root = Path(root) if root else HF_LEROBOT_HOME
@@ -154,7 +140,6 @@ class MultiLeRobotDataset(torch.utils.data.Dataset):
 
     @property
     def features(self) -> datasets.Features:
-        """The union of all underlying datasets' features (minus `disabled_features`)."""
         features = {}
         for dataset in self._datasets:
             features.update(
@@ -201,26 +186,17 @@ class MultiLeRobotDataset(torch.utils.data.Dataset):
 
     @property
     def tolerance_s(self) -> float:
-        """Tolerance in seconds used to discard loaded frames when their timestamps aren't close enough.
-
-        Only used when `delta_timestamps` is provided or when loading video frames from mp4 files.
+        """Tolerance in seconds used to discard loaded frames when their timestamps
+        are not close enough from the requested frames. It is only used when `delta_timestamps`
+        is provided or when loading video frames from mp4 files.
         """
         # 1e-4 to account for possible numerical error
         return 1 / self.fps - 1e-4
 
     def __len__(self):
-        """The total number of frames across all underlying datasets."""
         return self.num_frames
 
     def __getitem__(self, idx: int) -> dict[str, torch.Tensor]:
-        """Return the frame at `idx`, resolved to the underlying dataset it falls in.
-
-        Adds a `"dataset_index"` key identifying which underlying dataset the frame came from, and drops
-        any `disabled_features` keys.
-
-        Raises:
-            IndexError: If `idx` is out of bounds.
-        """
         if idx >= len(self):
             raise IndexError(f"Index {idx} out of bounds.")
         # Determine which dataset to get an item from based on the index.
@@ -243,7 +219,6 @@ class MultiLeRobotDataset(torch.utils.data.Dataset):
         return item
 
     def __repr__(self):
-        """A summary: repo IDs, sample/episode counts, media type, fps, camera keys, and transforms."""
         return (
             f"{self.__class__.__name__}(\n"
             f"  Repository IDs: '{self.repo_ids}',\n"

src/lerobot/datasets/pipeline_features.py

@@ -26,13 +26,12 @@ from lerobot.utils.feature_utils import hw_to_dataset_features
 def create_initial_features(
     action: RobotAction | None = None, observation: RobotObservation | None = None
 ) -> dict[PipelineFeatureType, dict[str, Any]]:
-    """Creates the initial features dict for the dataset from action and observation specs.
+    """
+    Creates the initial features dict for the dataset from action and observation specs.
 
     Args:
-        action (`dict[str, typing.Any] | None`, *optional*): A dictionary of action feature names to their
-            types/shapes.
-        observation (`dict[str, typing.Any] | None`, *optional*): A dictionary of observation feature names
-            to their types/shapes.
+        action: A dictionary of action feature names to their types/shapes.
+        observation: A dictionary of observation feature names to their types/shapes.
 
     Returns:
         The initial features dictionary structured by PipelineFeatureType.
@@ -47,14 +46,12 @@ def create_initial_features(
 
 # Helper to filter state/action keys based on compiled regex patterns.
 def should_keep(key: str, patterns: tuple[re.Pattern] | None) -> bool:
-    """Return `True` if `patterns` is `None` or any pattern in it matches `key`."""
     if patterns is None:
         return True
     return any(pat.search(key) for pat in patterns)
 
 
 def strip_prefix(key: str, prefixes_to_strip: tuple[str]) -> str:
-    """Remove the first prefix in `prefixes_to_strip` that `key` starts with, if any."""
     for prefix in prefixes_to_strip:
         if key.startswith(prefix):
             return key[len(prefix) :]
@@ -76,22 +73,20 @@ def aggregate_pipeline_dataset_features(
     exclude_images: bool = False,
     patterns: Sequence[str] | None = None,
 ) -> dict[str, dict]:
-    """Aggregates and filters pipeline features to create a dataset-ready features dictionary.
+    """
+    Aggregates and filters pipeline features to create a dataset-ready features dictionary.
 
     This function transforms initial features using the pipeline, categorizes them as action or observations
     (image or state), filters them based on `exclude_images` and `patterns`, and finally
     formats them for use with a Hugging Face LeRobot Dataset.
 
     Args:
-        pipeline (`DataProcessorPipeline`): The processor pipeline to apply to `initial_features`.
-        initial_features (`dict`): A dictionary of raw feature specs for actions and observations, keyed by
-            `PipelineFeatureType`.
-        use_videos (`bool`, *optional*, defaults to `True`): Controls the storage dtype for image features.
-            If `True`, images are stored as `"video"`; if `False`, they are stored as `"image"`.
-        exclude_images (`bool`, *optional*, defaults to `False`): If `True`, image features are dropped
-            entirely from the output.
-        patterns (`collections.abc.Sequence[str] | None`, *optional*): A sequence of regex patterns used to
-            filter action and state features.
+        pipeline: The DataProcessorPipeline to apply.
+        initial_features: A dictionary of raw feature specs for actions and observations.
+        use_videos: Controls the storage dtype for image features. If True, images are stored as "video"; if False, they are stored as "image".
+        exclude_images: If True, image features are dropped entirely from the output.
+        patterns: A sequence of regex patterns to filter action and state features.
+                  Image features are not affected by this filter.
 
     Returns:
         A dictionary of features formatted for a Hugging Face LeRobot Dataset.

src/lerobot/datasets/pyav_utils.py

@@ -41,11 +41,10 @@ def write_u16_plane(plane: av.video.plane.VideoPlane, src: np.ndarray, fill_valu
     leave the padding untouched.
 
     Args:
-        plane (`VideoPlane`): Destination 16-bit plane to copy into.
-        src (`ndarray`): Source image, shape `(height, width)`, dtype `uint16`.
-        fill_value (`int | None`, *optional*): If given, every pixel of the plane
-            (including the row padding) is set to this value first, so the padding
-            holds clean data instead of garbage.
+        plane: Destination 16-bit plane.
+        src: Source image, shape ``(height, width)``, dtype ``uint16``.
+        fill_value: If given, every pixel (padding included) is set to this first, so the
+            padding holds clean data instead of garbage.
     """
     height, width = src.shape
     stride_u16 = plane.line_size // np.dtype(np.uint16).itemsize

src/lerobot/datasets/sampler.py

@@ -55,8 +55,7 @@ class EpisodeAwareSampler:
         seed: int = 0,
         absolute_to_relative_idx: dict[int, int] | None = None,
     ):
-        """Build the sampler from per-episode `[from, to)` frame-index boundaries.
-
+        """
         Args:
             dataset_from_indices: Start index of each episode in the dataset.
             dataset_to_indices: End index of each episode in the dataset.
@@ -65,13 +64,6 @@ class EpisodeAwareSampler:
             drop_n_last_frames: Frames to drop from the end of each episode.
             shuffle: Whether to shuffle the indices.
             seed: Seed the permutation is derived from (together with the epoch).
-            absolute_to_relative_idx: Optional mapping from absolute dataset frame index to the relative
-                index actually yielded (e.g. when the sampler is used over a filtered subset).
-
-        Raises:
-            ValueError: If `drop_n_first_frames`/`drop_n_last_frames` is negative, if
-                `dataset_from_indices`/`dataset_to_indices` have different lengths, or if no episode has
-                any frames remaining after dropping.
         """
         if drop_n_first_frames < 0:
             raise ValueError(f"drop_n_first_frames must be >= 0, got {drop_n_first_frames}")
@@ -124,15 +116,12 @@ class EpisodeAwareSampler:
         return [self._frame_index(k) for k in range(self._num_frames)]
 
     def set_epoch(self, epoch: int) -> None:
-        """Set the epoch the next `__iter__` call will use, without consuming an auto-advance."""
         self._epoch = epoch
 
     def state_dict(self) -> dict:
-        """Return `{"epoch": ..., "start_index": ...}`, enough to resume mid-epoch sample-exactly."""
         return {"epoch": self._epoch, "start_index": self._start_index}
 
     def load_state_dict(self, state: dict) -> None:
-        """Restore the epoch and within-epoch offset from a `state_dict()`-produced dict."""
         self._epoch = state["epoch"]
         self._start_index = state["start_index"]
 
@@ -151,10 +140,6 @@ class EpisodeAwareSampler:
         return absolute_idx
 
     def __iter__(self) -> Iterator[int]:
-        """Yield frame indices for the current epoch (from `set_epoch`/`load_state_dict`), then advance it.
-
-        Shuffled if `self.shuffle`, using a permutation seeded from `(seed, epoch)`.
-        """
         # Advance epoch state eagerly, not on first consumption of the generator.
         epoch, start = self._epoch, self._start_index
         self._epoch += 1
@@ -171,7 +156,6 @@ class EpisodeAwareSampler:
                 yield self._frame_index(k)
 
     def __len__(self) -> int:
-        """The total number of frames across the sampled episodes (full length, even mid-resume)."""
         return self._num_frames
 
 

src/lerobot/datasets/streaming_dataset.py

@@ -44,13 +44,17 @@ from .video_utils import (
 
 
 class LookBackError(Exception):
-    """Exception raised when trying to look back in the history of a Backtrackable object."""
+    """
+    Exception raised when trying to look back in the history of a Backtrackable object.
+    """
 
     pass
 
 
 class LookAheadError(Exception):
-    """Exception raised when trying to look ahead in the future of a Backtrackable object."""
+    """
+    Exception raised when trying to look ahead in the future of a Backtrackable object.
+    """
 
     pass
 
@@ -60,10 +64,11 @@ class _ShardExhaustedError(Exception):
 
 
 class Backtrackable[T]:
-    """Wrap any iterator/iterable so you can step back up to `history` items and look ahead.
+    """
+    Wrap any iterator/iterable so you can step back up to `history` items
+    and look ahead up to `lookahead` items.
 
-    Looking ahead is bounded by `lookahead` items. This is useful for streaming datasets where you need
-    to access previous and future items
+    This is useful for streaming datasets where you need to access previous and future items
     but can't load the entire dataset into memory.
 
     Example:
@@ -93,16 +98,6 @@ class Backtrackable[T]:
     __slots__ = ("_source", "_back_buf", "_ahead_buf", "_cursor", "_history", "_lookahead")
 
     def __init__(self, iterable: Iterable[T], *, history: int = 1, lookahead: int = 0):
-        """Wrap `iterable`, buffering up to `history` past items and `lookahead` future items.
-
-        Args:
-            iterable: The iterable to wrap.
-            history: How many past items `prev()`/`peek_back()` can reach. Must be `>= 1`.
-            lookahead: How many future items `peek_ahead()` can reach. Must be `> 0`.
-
-        Raises:
-            ValueError: If `history < 1` or `lookahead <= 0`.
-        """
         if history < 1:
             raise ValueError("history must be >= 1")
         if lookahead <= 0:
@@ -116,11 +111,9 @@ class Backtrackable[T]:
         self._lookahead = lookahead
 
     def __iter__(self) -> "Backtrackable[T]":
-        """Return `self`; `Backtrackable` is its own iterator."""
         return self
 
     def __next__(self) -> T:
-        """Return the next item, consuming from the back buffer first if `prev()` stepped back."""
         # If we've stepped back, consume from back buffer first
         if self._cursor < 0:  # -1 means "last item", etc.
             self._cursor += 1
@@ -135,9 +128,9 @@ class Backtrackable[T]:
         return item
 
     def prev(self) -> T:
-        """Step one item back in history and return it.
-
-        Raises `LookBackError` if already at the oldest buffered item.
+        """
+        Step one item back in history and return it.
+        Raises IndexError if already at the oldest buffered item.
         """
         if len(self._back_buf) + self._cursor <= 1:
             raise LookBackError("At start of history")
@@ -146,15 +139,17 @@ class Backtrackable[T]:
         return self._back_buf[self._cursor]
 
     def peek_back(self, n: int = 1) -> T:
-        """Look `n` items back (n=1 == previous item) without moving the cursor."""
+        """
+        Look `n` items back (n=1 == previous item) without moving the cursor.
+        """
         if n < 0 or n + 1 > len(self._back_buf) + self._cursor:
             raise LookBackError("peek_back distance out of range")
 
         return self._back_buf[self._cursor - (n + 1)]
 
     def peek_ahead(self, n: int = 1) -> T:
-        """Look `n` items ahead (n=1 == next item) without moving the cursor.
-
+        """
+        Look `n` items ahead (n=1 == next item) without moving the cursor.
         Fills the ahead buffer if necessary.
         """
         if n < 1:
@@ -174,9 +169,9 @@ class Backtrackable[T]:
         return self._ahead_buf[n - 1]
 
     def history(self) -> list[T]:
-        """Return a copy of the buffered history (most recent last).
-
-        The list length is at most the `history` argument passed at construction.
+        """
+        Return a copy of the buffered history (most recent last).
+        The list length ≤ `history` argument passed at construction.
         """
         if self._cursor == 0:
             return list(self._back_buf)
@@ -185,12 +180,14 @@ class Backtrackable[T]:
         return list(self._back_buf)[: self._cursor or None]
 
     def can_peek_back(self, steps: int = 1) -> bool:
-        """Check if we can go back `steps` items without raising a `LookBackError`."""
+        """
+        Check if we can go back `steps` items without raising an IndexError.
+        """
         return steps < len(self._back_buf) + self._cursor
 
     def can_peek_ahead(self, steps: int = 1) -> bool:
-        """Check if we can peek ahead `steps` items.
-
+        """
+        Check if we can peek ahead `steps` items.
         This may involve trying to fill the ahead buffer.
         """
         if self._lookahead > 0 and steps > self._lookahead:
@@ -278,8 +275,6 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
             episodes (list[int] | None, optional): If specified, this will only load episodes specified by
                 their episode_index in this list.
             image_transforms (Callable | None, optional): Transform to apply to image data.
-            delta_timestamps (dict[list[float]] | None, optional): Per-feature timestamp offsets (in
-                seconds, relative to a frame's own timestamp) of additional frames to return alongside it.
             tolerance_s (float, optional): Tolerance in seconds for timestamp matching.
             revision (str, optional): Git revision id (branch name, tag, or commit hash).
             force_cache_sync (bool, optional): Flag to sync and refresh local files first.
@@ -289,8 +284,6 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
             seed (int, optional): Reproducibility random seed.
             rng (np.random.Generator | None, optional): Random number generator.
             shuffle (bool, optional): Whether to shuffle the dataset across exhaustions. Defaults to True.
-            return_uint8 (bool, optional): For RGB videos, whether to return raw uint8 frames instead of
-                the default float32 frames normalized to [0, 1].
             depth_output_unit (str, optional): Physical unit depth maps are dequantized to ("m" or "mm").
                 Defaults to "mm".
             repo_type: "dataset" (default) or "bucket" to stream from an HF Storage Bucket
@@ -390,17 +383,14 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
 
     @property
     def num_frames(self):
-        """The total number of frames in the dataset."""
         return self.meta.total_frames
 
     @property
     def num_episodes(self):
-        """The total number of episodes in the dataset."""
         return self.meta.total_episodes
 
     @property
     def fps(self):
-        """The dataset's recording frame rate."""
         return self.meta.fps
 
     @property
@@ -425,11 +415,6 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
     # could be used with a ThreadPoolExecutor to run `make_frame` (especially video decoding)
     # in parallel, feeding a queue from which this iterator will yield processed items.
     def __iter__(self) -> Iterator[dict[str, torch.Tensor]]:
-        """Yield frames via reservoir-buffered random sampling across shards, streaming indefinitely.
-
-        Samples a random shard, then a random frame from a fixed-size buffer refilled from that shard, so
-        no full shuffle or shard is ever fully materialized in memory.
-        """
         if self.video_decoder_cache is None:
             self.video_decoder_cache = VideoDecoderCache()
 
@@ -507,7 +492,7 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
             return dict.fromkeys(self.meta.video_keys, [start_ts])
 
     def _make_padding_camera_frame(self, camera_key: str):
-        """Variable-shape padding frame for the given camera key, shaped (H, W, C)."""
+        """Variable-shape padding frame for given camera keys, given in (H, W, C)"""
         return torch.zeros(self.meta.info.features[camera_key]["shape"]).permute(-1, 0, 1)
 
     def _get_video_frame_padding_mask(
@@ -538,7 +523,7 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         return padding_mask
 
     def make_frame(self, dataset_iterator: Backtrackable) -> Generator:
-        """Makes a frame starting from a dataset iterator."""
+        """Makes a frame starting from a dataset iterator"""
         try:
             item = next(dataset_iterator)
         except StopIteration as e:
@@ -631,13 +616,12 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         return query_timestamps
 
     def _query_videos(self, query_timestamps: dict[str, list[float]], ep_idx: int) -> dict:
-        """Decode the requested per-camera frame timestamps from `ep_idx`'s videos.
-
-        Note: When using data workers (e.g. DataLoader with num_workers>0), do not call this function
+        """Note: When using data workers (e.g. DataLoader with num_workers>0), do not call this function
         in the main process (e.g. by using a second Dataloader with num_workers=0). It will result in a
         Segmentation Fault. This probably happens because a memory reference to the video loader is created in
         the main process and a subprocess fails to access it.
         """
+
         item = {}
         for video_key, query_ts in query_timestamps.items():
             root = self.meta.url_root if self.streaming and not self.streaming_from_local else self.root
@@ -680,9 +664,8 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         """Get frames with delta offsets using the backtrackable iterator.
 
         Args:
-            dataset_iterator (Backtrackable): The backtrackable iterator to peek/step through for delta
-                frames.
             current_item (dict): Current item from the iterator.
+            ep_idx (int): Episode index.
 
         Returns:
             tuple: (query_result, padding) - frames at delta offsets and padding info.
@@ -787,7 +770,8 @@ class StreamingLeRobotDataset(torch.utils.data.IterableDataset):
         return query_result, padding
 
     def _validate_delta_timestamp_keys(self, delta_timestamps: dict[list[float]]) -> None:
-        """Validate that all keys in delta_timestamps correspond to actual features in the dataset.
+        """
+        Validate that all keys in delta_timestamps correspond to actual features in the dataset.
 
         Raises:
             ValueError: If any delta timestamp key doesn't correspond to a dataset feature.

src/lerobot/datasets/utils.py

@@ -18,7 +18,6 @@ import dataclasses
 import importlib.resources
 import json
 import logging
-from collections.abc import Sequence
 from dataclasses import dataclass, field
 from pathlib import Path
 
@@ -63,21 +62,11 @@ hub_api.create_tag("{repo_id}", tag="_version_", repo_type="dataset")
 """
 
 
-class CompatibilityError(Exception):
-    """Base class for errors raised when a dataset's `codebase_version` doesn't match this install."""
-
-    ...
+class CompatibilityError(Exception): ...
 
 
 class BackwardCompatibilityError(CompatibilityError):
-    """Raised when a dataset was saved with an older, unsupported `codebase_version`."""
-
     def __init__(self, repo_id: str, version: packaging.version.Version):
-        """Build the error message pointing the user at the v2.1-to-v3.0 conversion script.
-
-        Raises:
-            NotImplementedError: If `version` isn't the one supported legacy version (2.1).
-        """
         if version.major == 2 and version.minor == 1:
             message = V30_MESSAGE.format(repo_id=repo_id, version=version)
         else:
@@ -88,10 +77,7 @@ class BackwardCompatibilityError(CompatibilityError):
 
 
 class ForwardCompatibilityError(CompatibilityError):
-    """Raised when a dataset was saved with a newer `codebase_version` than this install supports."""
-
     def __init__(self, repo_id: str, version: packaging.version.Version):
-        """Build the error message pointing the user at upgrading their `lerobot` install."""
         message = FUTURE_MESSAGE.format(repo_id=repo_id, version=version)
         super().__init__(message)
 
@@ -112,47 +98,6 @@ VIDEO_DIR = "videos"
 
 CHUNK_FILE_PATTERN = "chunk-{chunk_index:03d}/file-{file_index:03d}"
 IMAGE_FILE_PATTERN = "frame-{frame_index:06d}.png"
-
-
-def resolve_episode_indices(
-    episodes: Sequence[int] | None,
-    total_episodes: int,
-    exclude_episodes: Sequence[int] | None = None,
-) -> list[int] | None:
-    """Resolve an optional episode allowlist and exclusion list against dataset bounds.
-
-    ``None`` is preserved when no filtering is requested so callers can retain
-    their native "all episodes" fast path. Invalid indices are ignored with a
-    warning, and the input order is preserved.
-    """
-    if total_episodes < 0:
-        raise ValueError(f"total_episodes must be non-negative, got {total_episodes}")
-
-    if episodes is None and not exclude_episodes:
-        return None
-
-    candidates = list(range(total_episodes)) if episodes is None else list(episodes)
-    invalid = [episode for episode in candidates if not 0 <= episode < total_episodes]
-    if invalid:
-        logger.warning(
-            "Ignoring episode indices outside the dataset range [0, %d): %s",
-            total_episodes,
-            invalid,
-        )
-    candidates = [episode for episode in candidates if 0 <= episode < total_episodes]
-
-    excluded = set(exclude_episodes or [])
-    invalid_excluded = sorted(episode for episode in excluded if not 0 <= episode < total_episodes)
-    if invalid_excluded:
-        logger.warning(
-            "Ignoring excluded episode indices outside the dataset range [0, %d): %s",
-            total_episodes,
-            invalid_excluded,
-        )
-    excluded = {episode for episode in excluded if 0 <= episode < total_episodes}
-    return [episode for episode in candidates if episode not in excluded]
-
-
 DEPTH_FILE_PATTERN = "frame-{frame_index:06d}.tiff"
 DEFAULT_TASKS_PATH = "meta/tasks.parquet"
 DEFAULT_EPISODES_PATH = EPISODES_DIR + "/" + CHUNK_FILE_PATTERN + ".parquet"
@@ -202,12 +147,6 @@ class DatasetInfo:
     tools: list[dict] | None = None
 
     def __post_init__(self) -> None:
-        """Coerce feature shapes from list to tuple, and validate `fps`/`chunks_size`/file-size fields.
-
-        Raises:
-            ValueError: If `fps`, `chunks_size`, `data_files_size_in_mb`, or `video_files_size_in_mb` isn't
-                positive.
-        """
         # Coerce feature shapes from list to tuple — JSON deserialisation
         # returns lists, but the rest of the codebase expects tuples.
         for ft in self.features.values():
@@ -258,11 +197,6 @@ class DatasetInfo:
     # Once all callers have been migrated to attribute access, remove these.
     # ---------------------------------------------------------------------------
     def __getitem__(self, key: str):
-        """Deprecated dict-style read; use attribute access instead.
-
-        Raises:
-            KeyError: If `key` isn't a field on this class.
-        """
         import warnings
 
         warnings.warn(
@@ -277,7 +211,6 @@ class DatasetInfo:
             raise KeyError(key) from err
 
     def __setitem__(self, key: str, value) -> None:
-        """Deprecated dict-style write; use attribute assignment instead."""
         import warnings
 
         warnings.warn(
@@ -315,7 +248,6 @@ def has_legacy_hub_download_metadata(root: Path) -> bool:
 
 
 def update_chunk_file_indices(chunk_idx: int, file_idx: int, chunks_size: int) -> tuple[int, int]:
-    """Advance to the next `(chunk_idx, file_idx)`, rolling over to a new chunk once `chunks_size` is hit."""
     if file_idx == chunks_size - 1:
         file_idx = 0
         chunk_idx += 1
@@ -381,7 +313,7 @@ def check_version_compatibility(
         repo_id (str): The repository ID for logging purposes.
         version_to_check (str | packaging.version.Version): The version of the dataset.
         current_version (str | packaging.version.Version): The current version of the codebase.
-        enforce_breaking_major (bool, *optional*, defaults to `True`): If True, raise an error on major version mismatch.
+        enforce_breaking_major (bool): If True, raise an error on major version mismatch.
 
     Raises:
         BackwardCompatibilityError: If the dataset version is from a newer, incompatible
@@ -408,9 +340,9 @@ def get_repo_versions(repo_id: str, *, token: str | bool | None = None) -> list[
 
     Args:
         repo_id (str): The repository ID on the Hugging Face Hub.
-        token (`str | bool | None`, *optional*): Authentication token used for Hub requests. Pass a string
-            token, `True` to require the locally stored token, `False` to disable authentication, or `None`
-            to use the Hugging Face Hub default.
+        token: Authentication token used for Hub requests. Pass a string token,
+            ``True`` to require the locally stored token, ``False`` to disable
+            authentication, or ``None`` to use the Hugging Face Hub default.
 
     Returns:
         list[packaging.version.Version]: A list of valid versions found.
@@ -440,7 +372,7 @@ def get_safe_version(
     Args:
         repo_id (str): The repository ID on the Hugging Face Hub.
         version (str | packaging.version.Version): The target version.
-        token (`str | bool | None`, *optional*): Authentication token forwarded to the Hub version lookup.
+        token: Authentication token forwarded to the Hub version lookup.
 
     Returns:
         str: The safe version string (e.g., "v1.2.3") to use as a revision.
@@ -487,7 +419,7 @@ def create_branch(repo_id: str, *, branch: str, repo_type: str | None = None) ->
     Args:
         repo_id (str): The ID of the repository.
         branch (str): The name of the branch to create.
-        repo_type (str | None, *optional*): The type of the repository (e.g., "dataset").
+        repo_type (str | None): The type of the repository (e.g., "dataset").
     """
     api = HfApi()
 
@@ -512,12 +444,10 @@ def create_lerobot_dataset_card(
     https://huggingface.co/docs/hub/repositories-licenses.
 
     Args:
-        tags (list | None, *optional*): A list of tags to add to the dataset card.
-        dataset_info (DatasetInfo | None, *optional*): The dataset's info object, which will
+        tags (list | None): A list of tags to add to the dataset card.
+        dataset_info (DatasetInfo | None): The dataset's info object, which will
             be displayed on the card.
-        kwargs (`Any`, *optional*): Values used to replace placeholders in the card template, e.g. `license`, which
-            must be a valid license identifier from
-            https://huggingface.co/docs/hub/repositories-licenses.
+        **kwargs: Additional keyword arguments to populate the card template.
 
     Returns:
         DatasetCard: The generated dataset card object.
@@ -552,12 +482,10 @@ def create_lerobot_dataset_card(
 
 
 def is_float_in_list(target, float_list, threshold=1e-6):
-    """Return `True` if `float_list` contains a value within `threshold` of `target`."""
     return any(abs(target - x) <= threshold for x in float_list)
 
 
 def find_float_index(target, float_list, threshold=1e-6):
-    """Return the index of the first value in `float_list` within `threshold` of `target`, or -1."""
     for i, x in enumerate(float_list):
         if abs(target - x) <= threshold:
             return i
@@ -565,7 +493,9 @@ def find_float_index(target, float_list, threshold=1e-6):
 
 
 def safe_shard(dataset: datasets.IterableDataset, index: int, num_shards: int) -> datasets.Dataset:
-    """Safe shards the dataset."""
+    """
+    Safe shards the dataset.
+    """
     shard_idx = min(dataset.num_shards, index + 1) - 1
 
     return dataset.shard(num_shards, index=shard_idx)

src/lerobot/datasets/video_utils.py

@@ -61,18 +61,19 @@ def decode_video_frames(
     return_uint8: bool = False,
     is_depth: bool = False,
 ) -> torch.Tensor:
-    """Decodes video frames using the specified backend.
+    """
+    Decodes video frames using the specified backend.
 
     Args:
         video_path (Path): Path to the video file.
         timestamps (list[float]): List of timestamps to extract frames.
         tolerance_s (float): Allowed deviation in seconds for frame retrieval.
-        backend (str, optional, *optional*): Backend to use for decoding. Defaults to "torchcodec" when available
+        backend (str, optional): Backend to use for decoding. Defaults to "torchcodec" when available
             in the platform; otherwise, defaults to "pyav". The legacy value "video_reader" is
             accepted for one release as an alias for "pyav" and will be removed in a future version.
-        return_uint8 (bool, *optional*, defaults to `False`): For RGB videos, if True return raw uint8 frames without float32 normalization.
+        return_uint8 (bool): For RGB videos, if True return raw uint8 frames without float32 normalization.
             This reduces memory for DataLoader IPC; normalization can be done on GPU afterward.
-        is_depth (bool, *optional*, defaults to `False`): Set to True if the video is a depth map (1 channel, uint12).
+        is_depth (bool): Set to True if the video is a depth map (1 channel, uint12).
 
     Returns:
         torch.Tensor: Decoded frames (RGB: float32 in [0,1] by default, or uint8 if return_uint8=True, Depth: uint12).
@@ -123,16 +124,14 @@ def decode_video_frames_pyav(
     video can be adjusted at encoding time to trade off decoding speed against file size.
 
     Args:
-        video_path (`pathlib.Path | str`): Path to the video file.
-        timestamps (`list`): List of timestamps, in seconds, to extract frames for.
-        tolerance_s (`float`): Allowed deviation in seconds between a queried timestamp
-            and the closest decoded frame.
-        log_loaded_timestamps (`bool`, *optional*, defaults to `False`): Whether to log
-            every decoded frame's timestamp at INFO level.
-        return_uint8 (`bool`, *optional*, defaults to `False`): For RGB videos, whether to
-            return raw uint8 frames instead of the default float32 frames normalized to [0, 1].
-        is_depth (`bool`, *optional*, defaults to `False`): Whether the video is a depth map
-            (1 channel, uint12).
+        video_path: Path to the video file.
+        timestamps: List of timestamps (in seconds) to extract frames for.
+        tolerance_s: Allowed deviation in seconds between a queried timestamp and the closest
+            decoded frame.
+        log_loaded_timestamps: When True, log every decoded frame's timestamp at INFO level.
+        return_uint8: For RGB videos, if True return raw uint8 frames (C, H, W).
+            Otherwise, return float32 in [0, 1] range.
+        is_depth: Set to True if the video is a depth map (1 channel, uint12).
 
     Returns:
         torch.Tensor of shape (len(timestamps), C, H, W).
@@ -266,31 +265,15 @@ class VideoDecoderCache:
     ever opened until the process exits).
 
     Args:
-        max_size (`int | None | object`, *optional*, defaults to `<unset>`): Maximum
-            number of decoders to retain. `None` disables eviction and restores legacy unbounded
-            behaviour. The sentinel default defers to the value of `LEROBOT_VIDEO_DECODER_CACHE_SIZE`
-            if set, otherwise `DEFAULT_DECODER_CACHE_SIZE`.
+        max_size: Maximum number of decoders to retain. ``None`` disables
+            eviction and restores legacy unbounded behaviour. Defaults to the
+            value of ``LEROBOT_VIDEO_DECODER_CACHE_SIZE`` if set, otherwise
+            :data:`DEFAULT_DECODER_CACHE_SIZE`.
     """
 
-    class _UnsetSentinel:
-        """Singleton marker distinguishing "not passed" from an explicit `None` `max_size`.
-
-        Has a fixed `__repr__` (unlike a bare `object()`) so it renders identically across
-        processes, which keeps the class docstring's `defaults to` clause stable.
-        """
-
-        def __repr__(self) -> str:
-            """Return `"<unset>"`, a stable placeholder for docstrings/logging."""
-            return "<unset>"
-
-    _SENTINEL: ClassVar[object] = _UnsetSentinel()
+    _SENTINEL: ClassVar[object] = object()
 
     def __init__(self, max_size: int | None | object = _SENTINEL):
-        """Create the cache. See the class docstring for `max_size`.
-
-        Raises:
-            ValueError: If `max_size` is neither `None` nor a positive integer.
-        """
         if max_size is VideoDecoderCache._SENTINEL:
             max_size = _default_max_cache_size()
         if max_size is not None and max_size <= 0:
@@ -300,7 +283,6 @@ class VideoDecoderCache:
         self._lock = Lock()
 
     def __contains__(self, video_path: object) -> bool:
-        """Return `True` if `video_path` (as `str`) has a cached decoder."""
         with self._lock:
             return str(video_path) in self._cache
 
@@ -356,7 +338,7 @@ class VideoDecoderCache:
 
 
 class FrameTimestampError(ValueError):
-    """Helper error to indicate the retrieved timestamps exceed the queried ones."""
+    """Helper error to indicate the retrieved timestamps exceed the queried ones"""
 
     pass
 
@@ -375,16 +357,11 @@ def decode_video_frames_torchcodec(
     """Loads frames associated with the requested timestamps of a video using torchcodec.
 
     Args:
-        video_path (`pathlib.Path | str`): Path to the video file.
-        timestamps (`list`): List of timestamps, in seconds, to extract frames for.
-        tolerance_s (`float`): Allowed deviation in seconds between a queried timestamp
-            and the closest decoded frame.
-        log_loaded_timestamps (`bool`, *optional*, defaults to `False`): Whether to log
-            every decoded frame's timestamp at INFO level.
-        decoder_cache (`lerobot.datasets.video_utils.VideoDecoderCache | None`, *optional*): Decoder
-            cache to fetch the `VideoDecoder` from. Uses the module-level default cache if `None`.
-        return_uint8 (`bool`, *optional*, defaults to `False`): For RGB videos, whether to
-            return raw uint8 frames instead of the default float32 frames normalized to [0, 1].
+        video_path: Path to the video file.
+        timestamps: List of timestamps to extract frames.
+        tolerance_s: Allowed deviation in seconds for frame retrieval.
+        log_loaded_timestamps: Whether to log loaded timestamps.
+        decoder_cache: Optional decoder cache instance. Uses default if None.
 
     Note: Setting device="cuda" outside the main process, e.g. in data loader workers, will lead to CUDA initialization errors.
 
@@ -474,19 +451,19 @@ def encode_video_frames(
     RGB frames are encoded directly.
 
     Args:
-        imgs_dir (`pathlib.Path | str`): Directory containing the frames to encode, named
-            `frame-000000` onwards (`.png` for RGB, `.tiff` for depth).
-        video_path (`pathlib.Path | str`): Output path for the encoded `.mp4` file.
-        fps (`int`): Frame rate of the output video.
-        video_encoder (`lerobot.configs.video.VideoEncoderConfig | None`, *optional*): Encoder settings
-            (codec, pixel format, quality, ...). When `None`, `rgb_encoder_defaults` is used. Pass a
-            `DepthEncoderConfig` to encode depth frames.
-        encoder_threads (`int | None`, *optional*): Per-encoder thread count forwarded to the codec.
-            `None` lets the codec decide.
-        log_level (`int | None`, *optional*, defaults to 24): libav log level to set while encoding,
-            or `None` to leave the current logging configuration unchanged.
-        overwrite (`bool`, *optional*, defaults to `False`): When `False` and `video_path` already
-            exists, skip encoding and log a warning. When `True`, re-encode and replace the existing file.
+        imgs_dir: Directory containing the frames to encode, named ``frame-000000``
+            onwards (``.png`` for RGB, ``.tiff`` for depth).
+        video_path: Output path for the encoded ``.mp4`` file.
+        fps: Frame rate of the output video.
+        video_encoder: Encoder settings (codec, pixel format, quality, ...). When
+            ``None``, :func:`rgb_encoder_defaults` is used. Pass a
+            :class:`~lerobot.configs.video.DepthEncoderConfig` to encode depth frames.
+        encoder_threads: Per-encoder thread count forwarded to the codec. ``None``
+            lets the codec decide.
+        log_level: libav log level to set while encoding, or ``None`` to leave the
+            current logging configuration unchanged.
+        overwrite: When ``False`` and ``video_path`` already exists, skip encoding and
+            log a warning. When ``True``, re-encode and replace the existing file.
     """
     if video_encoder is None:
         video_encoder = rgb_encoder_defaults()
@@ -575,21 +552,16 @@ def reencode_video(
     """Re-encode a video file, optionally trimming it to ``[start_time_s, end_time_s)``.
 
     Args:
-        input_video_path (`pathlib.Path | str`): Existing video file to read.
-        output_video_path (`pathlib.Path | str`): Path for the re-encoded file.
-        video_encoder (`lerobot.configs.video.VideoEncoderConfig | None`, *optional*): Encoder
-            configuration. Defaults to `rgb_encoder_defaults`.
-        encoder_threads (`int | None`, *optional*): Optional thread count forwarded to
-            `VideoEncoderConfig.get_codec_options`.
-        log_level (`int | None`, *optional*, defaults to 24): libav log level while encoding,
-            or `None` to leave logging unchanged.
-        overwrite (`bool`, *optional*, defaults to `False`): When `False` and `output_video_path`
-            already exists, skip and log a warning.
-        start_time_s (`float | None`, *optional*): When set, trim the output to start at this
-            timestamp, in seconds.
-        end_time_s (`float | None`, *optional*): When set, trim the output to end at this
-            timestamp, in seconds, exclusive.
+        input_video_path: Existing video file to read.
+        output_video_path: Path for the re-encoded file.
+        video_encoder: Encoder configuration. Defaults to :func:`rgb_encoder_defaults`.
+        encoder_threads: Optional thread count forwarded to :meth:`VideoEncoderConfig.get_codec_options`.
+        log_level: libav log level while encoding, or ``None`` to leave logging unchanged. Defaults to WARNING.
+        overwrite: When ``False`` and ``output_video_path`` already exists, skip and log a warning.
+        start_time_s: When set, trim the output to start at this timestamp (seconds).
+        end_time_s: When set, trim the output to end at this timestamp (seconds, exclusive).
     """
+
     video_encoder = video_encoder or rgb_encoder_defaults()
 
     if (start_time_s is not None and start_time_s < 0) or (end_time_s is not None and end_time_s < 0):
@@ -679,26 +651,25 @@ def concatenate_video_files(
     overwrite: bool = True,
     compatibility_check: bool = False,
 ):
-    """Concatenate multiple video files into a single video file using pyav.
+    """
+    Concatenate multiple video files into a single video file using pyav.
 
     This function takes a list of video input file paths and concatenates them into a single
     output video file. It uses ffmpeg's concat demuxer with stream copy mode for fast
     concatenation without re-encoding.
 
     Args:
-        input_video_paths (`list`): Ordered list of input video file paths to concatenate.
-        output_video_path (`Path`): Path to the output video file.
-        overwrite (`bool`, *optional*, defaults to `True`): Whether to overwrite the output
-            video file if it already exists.
-        compatibility_check (`bool`, *optional*, defaults to `False`): Whether to check that
-            the input videos share the same height, width, fps, codec, and pixel format
-            before concatenating.
+        input_video_paths: Ordered list of input video file paths to concatenate.
+        output_video_path: Path to the output video file.
+        overwrite: Whether to overwrite the output video file if it already exists. Default is True.
+        compatibility_check: Whether to check if the input videos are compatible. Default is False.
 
     Note:
         - Creates a temporary directory for intermediate files that is cleaned up after use.
         - Uses ffmpeg's concat demuxer which requires all input videos to have the same
           codec, resolution, and frame rate for proper concatenation.
     """
+
     output_video_path = Path(output_video_path)
 
     if output_video_path.exists() and not overwrite:
@@ -796,17 +767,6 @@ class _CameraEncoderThread(threading.Thread):
         stop_event: threading.Event,
         encoder_threads: int | None = None,
     ):
-        """Set up the thread; frames are only consumed once `start()` is called.
-
-        Args:
-            video_path: Output MP4 path.
-            fps: Output frame rate.
-            video_encoder: Codec/quality settings; `DepthEncoderConfig` selects depth-map encoding.
-            frame_queue: Queue this thread reads `(frame, ...)` items from.
-            result_queue: Queue the final stats are pushed to once encoding finishes.
-            stop_event: Set by the caller to signal this thread to stop early.
-            encoder_threads: Number of threads passed to the codec, if it supports one.
-        """
         super().__init__(daemon=True)
         self.video_path = video_path
         self.fps = fps
@@ -818,10 +778,6 @@ class _CameraEncoderThread(threading.Thread):
         self.encoder_threads = encoder_threads
 
     def run(self) -> None:
-        """Encode frames from `frame_queue` to `video_path` until a stop sentinel or `stop_event`.
-
-        Pushes the accumulated `RunningQuantileStats` to `result_queue` once encoding finishes.
-        """
         from .compute_stats import RunningQuantileStats, auto_downsample_height_width
 
         container = None
@@ -942,8 +898,7 @@ class StreamingVideoEncoder:
         queue_maxsize: int = 30,
         encoder_threads: int | None = None,
     ):
-        """Create the manager; per-camera encoder threads are started lazily on first frame.
-
+        """
         Args:
             fps: Frames per second for the output videos.
             rgb_encoder: Video encoder settings applied to all RGB cameras.
@@ -1150,9 +1105,11 @@ class StreamingVideoEncoder:
 @dataclass
 class VideoFrame:
     # TODO(rcadene, lhoestq): move to Hugging Face `datasets` repo
-    """Provides a type for a dataset containing video frames.
+    """
+    Provides a type for a dataset containing video frames.
 
     Example:
+
     ```python
     data_dict = [{"image": {"path": "videos/episode_0.mp4", "timestamp": 0.3}}]
     features = {"image": VideoFrame()}
@@ -1164,7 +1121,6 @@ class VideoFrame:
     _type: str = field(default="VideoFrame", init=False, repr=False)
 
     def __call__(self):
-        """Return the pyarrow struct type backing this feature, as required by `datasets.Features`."""
         return self.pa_type
 
 
@@ -1179,11 +1135,6 @@ with warnings.catch_warnings():
 
 
 def get_audio_info(video_path: Path | str) -> dict:
-    """Read audio-stream metadata (channels, codec, bit rate, sample rate, etc.) from a video file.
-
-    Returns:
-        A dict of `"audio.*"` keys, or `{"has_audio": False}` if `video_path` has no audio stream.
-    """
     # Set logging level
     logging.getLogger("libav").setLevel(av.logging.WARNING)
 
@@ -1222,13 +1173,13 @@ def get_video_info(
     """Build the ``video.*`` / ``audio.*`` info dict persisted in ``info.json``.
 
     Args:
-        video_path (`pathlib.Path | str`): Path to the encoded video file to probe.
-        video_encoder (`lerobot.configs.video.VideoEncoderConfig | None`, *optional*): If provided,
-            record the exact encoder settings used to encode this video. Stream-derived values take
-            precedence — encoder fields are only written for keys not already populated from the
-            video file itself. When a `DepthEncoderConfig` is passed, the depth quantization
-            parameters (`depth_min` / `depth_max` / `shift` / `use_log`) are recorded so frames can
-            be dequantized on read.
+        video_path: Path to the encoded video file to probe.
+        video_encoder: If provided, record the exact encoder settings used to encode this
+            video. Stream-derived values take precedence — encoder fields are only written for keys
+            not already populated from the video file itself. When a
+            :class:`~lerobot.configs.video.DepthEncoderConfig` is passed, the depth
+            quantization parameters (``depth_min`` / ``depth_max`` / ``shift`` /
+            ``use_log``) are recorded so frames can be dequantized on read.
 
     Returns:
         The ``video.*`` / ``audio.*`` info dict, including ``is_depth_map`` which is
@@ -1276,10 +1227,11 @@ def get_video_info(
 
 
 def get_video_duration_in_s(video_path: Path | str) -> float:
-    """Get the duration of a video file in seconds using PyAV.
+    """
+    Get the duration of a video file in seconds using PyAV.
 
     Args:
-        video_path (`pathlib.Path | str`): Path to the video file.
+        video_path: Path to the video file.
 
     Returns:
         Duration of the video in seconds.
@@ -1297,7 +1249,8 @@ def get_video_duration_in_s(video_path: Path | str) -> float:
 
 
 class VideoEncodingManager:
-    """Context manager that ensures proper video encoding and data cleanup even if exceptions occur.
+    """
+    Context manager that ensures proper video encoding and data cleanup even if exceptions occur.
 
     This manager handles:
     - Batch encoding for any remaining episodes when recording interrupted
@@ -1305,23 +1258,16 @@ class VideoEncodingManager:
     - Removing empty image directories
 
     Args:
-        dataset (`LeRobotDataset`): The LeRobotDataset instance.
+        dataset: The LeRobotDataset instance
     """
 
     def __init__(self, dataset):
-        """Store the `LeRobotDataset` this manager will finalize/clean up on exit."""
         self.dataset = dataset
 
     def __enter__(self):
-        """Return `self`; no setup is needed on entry."""
         return self
 
     def __exit__(self, exc_type, exc_val, exc_tb):
-        """Finalize the dataset, cancelling pending videos and cleaning up interrupted-episode files.
-
-        Runs unconditionally (even if `exc_type` is set), so partial/interrupted recordings still leave
-        a consistent dataset on disk.
-        """
         writer = self.dataset.writer
         if writer is not None:
             if exc_type is not None and writer._streaming_encoder is not None:

src/lerobot/distributed/__init__.py

@@ -1,43 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Distributed-training runtime for LeRobot.
-
-This package owns everything that turns the declarative topology in
-:class:`lerobot.configs.parallelism.ParallelismConfig` into a running engine:
-mesh math (:class:`~lerobot.distributed.parallel_dims.ParallelDims`), the
-`Accelerator` factory (:func:`~lerobot.distributed.factory.make_accelerator`),
-sharding-aware checkpoint helpers, and small rank utilities.
-
-Setup-order contract (normative):
-CP dispatch install -> activation checkpointing -> torch.compile ->
-``fully_shard``/DDP (via ``accelerator.prepare``) -> optimizer rebind.
-Only the last two steps are active today; CP/AC/compile are configured
-placeholders wired in later rounds.
-"""
-
-from .factory import guard_against_env_interference, make_accelerator, set_fsdp_wrap_modules
-from .parallel_dims import ParallelDims
-from .utils import finalize_sharded_policy, is_main_process, strip_accelerate_cp_hooks
-
-__all__ = [
-    "ParallelDims",
-    "finalize_sharded_policy",
-    "guard_against_env_interference",
-    "is_main_process",
-    "make_accelerator",
-    "set_fsdp_wrap_modules",
-    "strip_accelerate_cp_hooks",
-]

src/lerobot/distributed/checkpoint.py

@@ -1,195 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Sharding-aware checkpoint primitives.
-
-Two artifact channels with distinct owners:
-
-- the **distributable** ``model.safetensors``: produced by ``PreTrainedPolicy.save_pretrained``
-  through :func:`full_model_state_dict` — a collective full gather when the model is sharded;
-- the **resume** channel (sharded runs): torch DCP directories written/read through accelerate's
-  ``save/load_fsdp_model`` and ``save/load_fsdp_optimizer`` (``pytorch_model_fsdp_0/`` and
-  ``optimizer_0/``, names imported from accelerate constants), which reshard on load across
-  topology changes.
-
-Every function that touches sharded state is a collective and must run on ALL ranks.
-"""
-
-from pathlib import Path
-from typing import TYPE_CHECKING
-
-import torch
-from torch import nn
-
-if TYPE_CHECKING:
-    from accelerate import Accelerator
-
-
-def is_sharded_module(module: nn.Module) -> bool:
-    """True when `fully_shard` owns this module's parameters (FSDP2's in-place class swap).
-
-    Args:
-        module (nn.Module): The module to inspect (a torch.compile wrapper is looked through
-            via `_orig_mod`).
-
-    Returns:
-        bool: True when the module (or its compiled `_orig_mod`) is an `FSDPModule`.
-    """
-    from torch.distributed.fsdp import FSDPModule
-
-    if isinstance(module, FSDPModule):
-        return True
-    # torch.compile wraps the sharded module; mirror accelerate's `_orig_mod` check.
-    orig_mod = getattr(module, "_orig_mod", None)
-    return orig_mod is not None and isinstance(orig_mod, FSDPModule)
-
-
-def full_model_state_dict(module: nn.Module) -> dict[str, torch.Tensor]:
-    """The module's full (unsharded) state dict, however its parameters are laid out.
-
-    Sharded modules gather through torch's DCP state-dict API: a COLLECTIVE that must run on
-    every rank; with ``cpu_offload=True`` the full dict materializes on the main rank only and
-    every other rank receives a literal ``{}`` (runtime-verified — a
-    rank-0-gated call deadlocks). Plain modules return ``module.state_dict()`` on every rank.
-
-    Args:
-        module (nn.Module): The (possibly sharded) module to read the state dict from.
-
-    Returns:
-        dict[str, torch.Tensor]: The full state dict — on the main rank only (``{}``
-            elsewhere) when the module is sharded, on every rank otherwise.
-    """
-    if not is_sharded_module(module):
-        return module.state_dict()
-
-    from torch.distributed.checkpoint.state_dict import StateDictOptions, get_model_state_dict
-
-    return get_model_state_dict(module, options=StateDictOptions(full_state_dict=True, cpu_offload=True))
-
-
-def _fsdp_plugin(accelerator: "Accelerator") -> object:
-    """The accelerator's FSDP plugin, required by every DCP save/load helper below.
-
-    Args:
-        accelerator (Accelerator): The accelerator that prepared the sharded model.
-
-    Returns:
-        object: The FSDP plugin held by `accelerator.state`.
-
-    Raises:
-        RuntimeError: If the accelerator was not configured with an FSDP plugin.
-    """
-    plugin = getattr(accelerator.state, "fsdp_plugin", None)
-    if plugin is None:
-        raise RuntimeError("Sharded checkpointing requires an FSDP-prepared Accelerator.")
-    return plugin
-
-
-def save_sharded_model(accelerator: "Accelerator", model: nn.Module, output_dir: Path) -> None:
-    """Write the DCP model shards (`pytorch_model_fsdp_0/`). Collective: call on all ranks.
-
-    Args:
-        accelerator (Accelerator): The accelerator that prepared the sharded model.
-        model (nn.Module): The prepared (sharded) model to save.
-        output_dir (Path): The directory the shard subdirectory is created in.
-    """
-    from accelerate.utils import save_fsdp_model
-
-    # accelerate 1.14's DCP helpers do string containment checks on the path:
-    # always hand them str, never Path.
-    save_fsdp_model(_fsdp_plugin(accelerator), accelerator, model, str(output_dir))
-
-
-def load_sharded_model(accelerator: "Accelerator", model: nn.Module, input_dir: Path) -> None:
-    """Load DCP model shards into the prepared (sharded) model. Collective: call on all ranks.
-
-    Args:
-        accelerator (Accelerator): The accelerator that prepared the sharded model.
-        model (nn.Module): The prepared (sharded) model to load into.
-        input_dir (Path): The directory containing the `pytorch_model_fsdp_0/` shard
-            subdirectory.
-    """
-    from accelerate.utils import load_fsdp_model
-    from accelerate.utils.constants import FSDP_MODEL_NAME
-
-    # Pass the exact shard directory: accelerate's load resolves it with a substring check
-    # ("pytorch_model_fsdp" in the path -> use as-is), which misfires on run paths that happen
-    # to contain the marker; the exact dir makes the check deterministic.
-    load_fsdp_model(_fsdp_plugin(accelerator), accelerator, model, str(input_dir / f"{FSDP_MODEL_NAME}_0"))
-
-
-def save_sharded_optimizer(
-    accelerator: "Accelerator", optimizer: torch.optim.Optimizer, model: nn.Module, output_dir: Path
-) -> None:
-    """Write the DCP optimizer shards (`optimizer_0/`). Collective: call on all ranks.
-
-    Args:
-        accelerator (Accelerator): The accelerator that prepared the model and optimizer.
-        optimizer (torch.optim.Optimizer): The prepared optimizer to save the state from.
-        model (nn.Module): The prepared (sharded) model the optimizer state is keyed by.
-        output_dir (Path): The directory the shard subdirectory is created in.
-    """
-    from accelerate.utils import save_fsdp_optimizer
-
-    save_fsdp_optimizer(_fsdp_plugin(accelerator), accelerator, optimizer, model, str(output_dir))
-
-
-def load_sharded_optimizer(
-    accelerator: "Accelerator", optimizer: torch.optim.Optimizer, model: nn.Module, input_dir: Path
-) -> None:
-    """Load DCP optimizer shards into the prepared optimizer. Collective: call on all ranks.
-
-    Must run AFTER ``accelerator.prepare()``: FSDP2's prepare rebinds the optimizer's param
-    groups to sharded DTensors but never migrates ``optimizer.state`` — the resharding load is
-    the only correct way to restore it.
-
-    Args:
-        accelerator (Accelerator): The accelerator that prepared the model and optimizer.
-        optimizer (torch.optim.Optimizer): The prepared optimizer to restore the state into.
-        model (nn.Module): The prepared (sharded) model the optimizer state is keyed by.
-        input_dir (Path): The directory containing the `optimizer_0/` shard subdirectory.
-    """
-    from accelerate.utils import load_fsdp_optimizer
-    from accelerate.utils.constants import OPTIMIZER_NAME
-
-    # Exact shard directory for the same reason as load_sharded_model: accelerate's substring
-    # check ("optimizer" in the path) would misread e.g. --job_name=optimizer_sweep run paths.
-    load_fsdp_optimizer(
-        _fsdp_plugin(accelerator), accelerator, optimizer, model, str(input_dir / f"{OPTIMIZER_NAME}_0")
-    )
-
-
-def dcp_to_safetensors(dcp_dir: Path, output_dir: Path, *, delete_dcp: bool = False) -> Path:
-    """Merge a DCP shard directory into a single `model.safetensors` (offline, single process).
-
-    Thin wrapper over `accelerate.utils.merge_fsdp_weights`, which loads the shards without a
-    process group, writes safetensors directly, and — when asked — removes the merged shard
-    directory itself, only on the main process and only once the merge has succeeded.
-
-    Args:
-        dcp_dir (Path): The DCP shard directory to merge (e.g. `.../pytorch_model_fsdp_0`).
-        output_dir (Path): The directory the merged `model.safetensors` is written into.
-        delete_dcp (bool): Whether to remove the shard directory once it has been merged.
-            Defaults to False.
-
-    Returns:
-        Path: The written `model.safetensors` file's path.
-    """
-    from accelerate.utils import merge_fsdp_weights
-
-    merge_fsdp_weights(
-        str(dcp_dir), str(output_dir), safe_serialization=True, remove_checkpoint_dir=delete_dcp
-    )
-    return output_dir / "model.safetensors"

src/lerobot/distributed/factory.py

@@ -1,147 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""The `Accelerator` factory — the only place accelerate gets configured.
-
-`torchrun` is the launcher; every accelerate parameter comes from `TrainPipelineConfig`
-(`cfg.parallelism` + `cfg.accelerator`) so a run is reproducible from its `train_config.json`
-alone. `accelerate launch` without a `--config_file` remains equivalent (it only sets rendezvous
-env vars in that mode); the yaml flow is superseded.
-"""
-
-import os
-from typing import TYPE_CHECKING
-
-from lerobot.configs.parallelism import world_size_from_env
-from lerobot.configs.train import TrainPipelineConfig
-
-if TYPE_CHECKING:
-    from accelerate import Accelerator
-
-    from lerobot.policies.pretrained import PreTrainedPolicy
-
-# Env vars through which `accelerate launch --config_file` (or a stray shell) would configure
-# accelerate behind the config system's back. Plugin `__post_init__`s read these silently as
-# field fallbacks (ACCELERATE_DYNAMO_* enables torch.compile through the default
-# TorchDynamoPlugin; ACCELERATE_GRADIENT_ACCUMULATION_STEPS overrides the explicitly passed
-# value inside Accelerator.__init__), which would make train_config.json lie about what ran.
-_ACCELERATE_ENV_PREFIXES = ("FSDP_", "PARALLELISM_CONFIG_", "ACCELERATE_DYNAMO_")
-_ACCELERATE_ENV_VARS = (
-    "ACCELERATE_USE_FSDP",
-    "ACCELERATE_USE_PARALLELISM_CONFIG",
-    "ACCELERATE_GRADIENT_ACCUMULATION_STEPS",
-)
-_ENV_OVERRIDE = "LEROBOT_ALLOW_ACCELERATE_ENV"
-
-
-def guard_against_env_interference() -> None:
-    """Hard-error when accelerate-configuring env vars are set.
-
-    A silently env-overridden "reproducible" config is worse than a stop: users migrating from
-    the old `accelerate launch --config_file fsdp.yaml` flow get a precise error instead of a
-    config that lies. Set LEROBOT_ALLOW_ACCELERATE_ENV=1 to acknowledge and proceed.
-
-    Raises:
-        RuntimeError: If any accelerate-configuring environment variable is set and the
-            LEROBOT_ALLOW_ACCELERATE_ENV override is not.
-    """
-    if os.environ.get(_ENV_OVERRIDE):
-        return
-    offending = sorted(
-        name
-        for name in os.environ
-        if name in _ACCELERATE_ENV_VARS or name.startswith(_ACCELERATE_ENV_PREFIXES)
-    )
-    if offending:
-        raise RuntimeError(
-            f"Accelerate-configuring environment variables are set: {', '.join(offending)}. "
-            "LeRobot manages accelerate exclusively through TrainPipelineConfig "
-            "(--parallelism.* / --accelerator.*); launch with plain torchrun and remove these "
-            "variables (the `accelerate launch --config_file` flow is superseded), or set "
-            f"{_ENV_OVERRIDE}=1 to acknowledge that they may override your config."
-        )
-
-
-def make_accelerator(cfg: TrainPipelineConfig) -> "Accelerator":
-    """Resolve the topology against the launched world and build the `Accelerator`.
-
-    Must run once per process, before any other component needs the device or the process
-    group (`Accelerator.__init__` initializes both and builds the device mesh).
-
-    Args:
-        cfg (TrainPipelineConfig): The full training config; `cfg.parallelism` is resolved in
-            place against the launched world size and `cfg.accelerator` builds the result.
-
-    Returns:
-        Accelerator: The configured accelerator, with device and process group initialized.
-
-    Raises:
-        ValueError: If `cfg.checkpoint_format` requires DCP but the topology resolved to a
-            non-sharded run.
-    """
-    guard_against_env_interference()
-    cfg.parallelism.resolve(world_size_from_env())
-    # The parse-time format check ran against the declared degrees, where the dp_shard=-1
-    # sentinel counts as sharded; it may resolve to an unsharded run (e.g. -1 at world size 1).
-    # Re-check against the concrete degrees so the recorded format never lies about the
-    # artifacts a checkpoint will actually contain.
-    if cfg.checkpoint_format.wants_dcp and not cfg.parallelism.is_sharded:
-        raise ValueError(
-            f"checkpoint_format={cfg.checkpoint_format.value} requires a sharded run, but the "
-            f"topology resolved to a non-sharded one (dp_replicate={cfg.parallelism.dp_replicate}, "
-            f"dp_shard={cfg.parallelism.dp_shard}); non-sharded checkpoints are always safetensors."
-        )
-    return cfg.accelerator.build(
-        cfg.parallelism,
-        cpu=cfg.trainable_config.device == "cpu",
-    )
-
-
-def set_fsdp_wrap_modules(accelerator: "Accelerator", policy: "PreTrainedPolicy") -> None:
-    """Resolve the FSDP wrap-unit class names onto the plugin before `accelerator.prepare()`.
-
-    Resolution order: user override (`--accelerator.fsdp.wrap_modules`, already on the plugin)
-    -> the policy's `_fsdp_wrap_modules` declaration -> hard error. Root-only wrapping — the
-    silent default when no wrap source exists — is never accepted: it quietly forfeits all
-    sharding memory savings.
-
-    No-op for the size-based policy (`--accelerator.fsdp.min_num_params`), which needs no class
-    names, and for non-sharded runs (no fsdp plugin).
-
-    Args:
-        accelerator (Accelerator): The accelerator whose FSDP plugin receives the wrap-unit
-            class names.
-        policy (PreTrainedPolicy): The trainable whose class may declare `_fsdp_wrap_modules`.
-
-    Raises:
-        ValueError: If sharded class-based wrapping is configured but neither a user override
-            nor a policy declaration supplies wrap-unit class names.
-    """
-    plugin = getattr(accelerator.state, "fsdp_plugin", None)
-    if plugin is None or plugin.min_num_params:
-        return
-    if plugin.transformer_cls_names_to_wrap:  # user override, set at build time
-        return
-    # getattr, not attribute access: non-policy trainables (no `_fsdp_wrap_modules` attribute)
-    # must reach the actionable error below, not an AttributeError.
-    declared = getattr(type(policy), "_fsdp_wrap_modules", None)
-    if not declared:
-        raise ValueError(
-            f"Policy '{type(policy).__name__}' declares no FSDP wrap units. Sharded training "
-            "requires wrap-unit class names: set --accelerator.fsdp.wrap_modules='[\"MyBlock\"]' "
-            "(or --accelerator.fsdp.min_num_params for a size-based policy), or declare "
-            "`_fsdp_wrap_modules` on the policy class."
-        )
-    plugin.transformer_cls_names_to_wrap = list(declared)

src/lerobot/distributed/parallel_dims.py

@@ -1,112 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Runtime mesh math derived from the declarative :class:`ParallelismConfig`.
-
-`ParallelDims` is the training script's single source of truth for topology-derived numbers
-(data-parallel world size and rank, sample accounting inputs) and — once the CP engine lands —
-the owner of LeRobot's private ``(dp_replicate, dp_shard, ring, ulysses)`` mesh. It is a runtime
-object and is never serialized (the config it derives from is what lands in
-``train_config.json``).
-"""
-
-from dataclasses import dataclass
-
-import torch.distributed as dist
-
-from lerobot.configs.parallelism import ParallelismConfig
-
-
-@dataclass(frozen=True)
-class ParallelDims:
-    """Concrete parallelism degrees bound to a world size (canonical row-major rank layout)."""
-
-    dp_replicate: int
-    dp_shard: int
-    ring: int
-    ulysses: int
-    world_size: int
-    device_type: str
-
-    @classmethod
-    def from_config(cls, cfg: ParallelismConfig, world_size: int, device_type: str) -> "ParallelDims":
-        """Bind a *resolved* config to the actual runtime world size (cross-checked here).
-
-        Args:
-            cfg (ParallelismConfig): The declarative topology, already resolved via
-                `ParallelismConfig.resolve(world_size)`.
-            world_size (int): The launched world size the declared degrees must multiply to.
-            device_type (str): The accelerator device type backing the mesh (e.g. "cuda").
-
-        Returns:
-            ParallelDims: The concrete parallelism degrees bound to this world.
-
-        Raises:
-            ValueError: If the config is unresolved (`dp_shard == -1`) or its degrees do not
-                multiply to `world_size`.
-        """
-        total = cfg.dp_replicate * cfg.dp_shard * cfg.cp_size
-        if cfg.dp_shard == -1 or total != world_size:
-            raise ValueError(
-                f"ParallelismConfig is not resolved against this world: dp_replicate="
-                f"{cfg.dp_replicate} * dp_shard={cfg.dp_shard} * cp={cfg.cp_size} != "
-                f"world_size={world_size}. Call ParallelismConfig.resolve(world_size) first "
-                "(make_accelerator does this)."
-            )
-        return cls(
-            dp_replicate=cfg.dp_replicate,
-            dp_shard=cfg.dp_shard,
-            ring=cfg.context_parallel.ring_degree,
-            ulysses=cfg.context_parallel.ulysses_degree,
-            world_size=world_size,
-            device_type=device_type,
-        )
-
-    @property
-    def cp_size(self) -> int:
-        """Total context-parallel degree (`ring * ulysses`)."""
-        return self.ring * self.ulysses
-
-    @property
-    def is_sharded(self) -> bool:
-        """Whether parameters are sharded (`dp_shard > 1` or any context parallelism)."""
-        return self.dp_shard > 1 or self.cp_size > 1
-
-    @property
-    def dp_world_size(self) -> int:
-        """Number of distinct data-parallel workers — the divisor for all sample accounting."""
-        return self.dp_replicate * self.dp_shard
-
-    @property
-    def dp_rank(self) -> int:
-        """This process's data-parallel coordinate (CP peers share one dp_rank).
-
-        With the canonical row-major layout and (ring, ulysses) innermost, CP peers are
-        contiguous global ranks, so the dp coordinate is the integer quotient by cp_size —
-        the same arithmetic accelerate's mesh-aware dataloader applies.
-        """
-        global_rank = dist.get_rank() if dist.is_initialized() else 0
-        return global_rank // self.cp_size
-
-    def cp_mesh(self) -> None:
-        """Private (ring, ulysses) mesh for the CP engine — reserved for the CP round.
-
-        Raises:
-            NotImplementedError: Always — context parallelism is not implemented yet.
-        """
-        raise NotImplementedError(
-            "Context parallelism is not implemented yet; ParallelDims.cp_mesh is reserved for "
-            "the CP engine round (a private mesh aligned with accelerate's cp block)."
-        )

src/lerobot/distributed/utils.py

@@ -1,94 +0,0 @@
-#!/usr/bin/env python
-
-# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Rank utilities and post-`prepare()` sharding finalization."""
-
-import logging
-from typing import TYPE_CHECKING
-
-import torch.distributed as dist
-from torch import nn
-
-if TYPE_CHECKING:
-    from lerobot.distributed.parallel_dims import ParallelDims
-
-
-def is_main_process() -> bool:
-    """True on the process that owns rank-0-only side effects (file writes, uploads, logging).
-
-    Torch-native on purpose: persistence code must not depend on an `Accelerator` handle —
-    `_save_pretrained` and the hub publishers run in contexts that have none. Outside
-    distributed runs every process is the main process.
-
-    Returns:
-        bool: True when this process is rank 0 or no process group is initialized.
-    """
-    return not dist.is_initialized() or dist.get_rank() == 0
-
-
-def strip_accelerate_cp_hooks(model: nn.Module) -> int:
-    """Remove accelerate's context-parallel forward-pre-hooks from every module.
-
-    When `cp_size > 1` is declared, `accelerator.prepare()` unconditionally attaches hooks that
-    silently replace any `attention_mask` kwarg of `*self_attn` modules with `is_causal=True`
-    (`accelerate.big_modeling._attach_context_parallel_hooks`) — mask corruption for policies
-    with non-causal attention. LeRobot implements CP itself and never enters accelerate's CP
-    context, so these hooks are pure hazard. Deterministically identified by their defining
-    module; a version canary pins that identity.
-
-    Args:
-        model (nn.Module): The prepared model to strip the hooks from (all submodules are
-            visited).
-
-    Returns:
-        int: The number of hooks removed.
-    """
-    removed = 0
-    for module in model.modules():
-        for hook_id, hook in list(module._forward_pre_hooks.items()):
-            if getattr(hook, "__module__", None) == "accelerate.big_modeling":
-                del module._forward_pre_hooks[hook_id]
-                module._forward_pre_hooks_with_kwargs.pop(hook_id, None)
-                removed += 1
-    return removed
-
-
-def finalize_sharded_policy(policy: nn.Module, parallel_dims: "ParallelDims") -> None:
-    """Sharding correctness protocol, applied once, immediately after `accelerator.prepare()`.
-
-    1. Strip accelerate's CP mask hooks (only attached when cp > 1 was declared).
-    2. Register the policy's non-`forward` entry points (`_fsdp_forward_methods`) so FSDP2
-       unshards parameters around `select_action` & co. — without this, any inference-style
-       call on a sharded policy crashes on mixed Tensor/DTensor.
-
-    No-op for DDP/single-process runs.
-
-    Args:
-        policy (nn.Module): The policy as returned by `accelerator.prepare()`.
-        parallel_dims (ParallelDims): The run's resolved topology; decides whether the protocol
-            applies.
-    """
-    if not parallel_dims.is_sharded:
-        return
-    if parallel_dims.cp_size > 1:
-        removed = strip_accelerate_cp_hooks(policy)
-        logging.info("Stripped %d accelerate context-parallel attention-mask hooks.", removed)
-
-    from torch.distributed.fsdp import FSDPModule, register_fsdp_forward_method
-
-    if isinstance(policy, FSDPModule):
-        for method_name in getattr(type(policy), "_fsdp_forward_methods", ()):
-            if callable(getattr(policy, method_name, None)):
-                register_fsdp_forward_method(policy, method_name)

src/lerobot/jobs/hf.py

@@ -432,7 +432,7 @@ def submit_to_hf(cfg: TrainPipelineConfig) -> None:
 
     # Finish as soon as the model is pushed, rather than waiting out the platform's
     # post-run finalization before the job stage flips to COMPLETED. This matches the
-    # exact log line emitted by lerobot.common.train_utils.publish_trained_model — the two must stay
+    # exact log line emitted by PreTrainedPolicy.push_model_to_hub — the two must stay
     # in sync. If it ever stops matching we just fall back to stage-based completion
     # (~30s slower), so the contract is an optimization, not a correctness requirement.
     success_marker = f"Model pushed to https://huggingface.co/{repo_id}"

src/lerobot/motors/motors_bus.py

@@ -314,16 +314,11 @@ class SerialMotorsBus(MotorsBusBase):
     To find the port, you can run our utility script:
     ```bash
     lerobot-find-port.py
-    ```
-
-    which prints:
-
-    ```
-    Finding all available ports for the MotorsBus.
-    ["/dev/tty.usbmodem575E0032081", "/dev/tty.usbmodem575E0031751"]
-    Remove the usb cable from your MotorsBus and press Enter when done.
-    The port of this MotorsBus is /dev/tty.usbmodem575E0031751.
-    Reconnect the usb cable.
+    >>> Finding all available ports for the MotorsBus.
+    >>> ["/dev/tty.usbmodem575E0032081", "/dev/tty.usbmodem575E0031751"]
+    >>> Remove the usb cable from your MotorsBus and press Enter when done.
+    >>> The port of this MotorsBus is /dev/tty.usbmodem575E0031751.
+    >>> Reconnect the usb cable.
     ```
 
     Example of usage for 1 Feetech sts3215 motor connected to the bus:
@@ -600,7 +595,7 @@ class SerialMotorsBus(MotorsBusBase):
         ID, and finally programs the bus' default baud-rate.
 
         Args:
-            motor (str): Key of the motor in `motors`.
+            motor (str): Key of the motor in :pyattr:`motors`.
             initial_baudrate (int | None, optional): Current baud-rate (skips scanning when provided).
                 Defaults to None.
             initial_id (int | None, optional): Current ID (skips scanning when provided). Defaults to None.
@@ -671,7 +666,7 @@ class SerialMotorsBus(MotorsBusBase):
         """Enable torque on selected motors.
 
         Args:
-            motors (int | str | list[str] | None, optional): Same semantics as [`~motors.motors_bus.MotorsBus.disable_torque`].
+            motors (int | str | list[str] | None, optional): Same semantics as :pymeth:`disable_torque`.
                 Defaults to `None`.
             num_retry (int, optional): Number of additional retry attempts on communication failure.
                 Defaults to 0.
@@ -684,12 +679,10 @@ class SerialMotorsBus(MotorsBusBase):
 
         This helper is useful to temporarily disable torque when configuring motors.
 
-        Example:
-            ```python
-            >>> with bus.torque_disabled():  # doctest: +SKIP
+        Examples:
+            >>> with bus.torque_disabled():
             ...     # Safe operations here
             ...     pass
-            ```
         """
         self.disable_torque(motors)
         try:
@@ -702,7 +695,7 @@ class SerialMotorsBus(MotorsBusBase):
 
         Args:
             timeout_ms (int | None, optional): Timeout in *milliseconds*. If `None` (default) the method falls
-                back to `default_timeout`.
+                back to :pyattr:`default_timeout`.
         """
         timeout_ms = timeout_ms if timeout_ms is not None else self.default_timeout
         self.port_handler.setPacketTimeoutMillis(timeout_ms)
@@ -753,8 +746,8 @@ class SerialMotorsBus(MotorsBusBase):
 
         Args:
             calibration_dict (dict[str, MotorCalibration]): Calibration obtained from
-                [`~motors.motors_bus.MotorsBus.read_calibration`] or crafted by the user.
-            cache (bool, optional): Save the calibration to `calibration`. Defaults to True.
+                :pymeth:`read_calibration` or crafted by the user.
+            cache (bool, optional): Save the calibration to :pyattr:`calibration`. Defaults to True.
         """
         pass
 
@@ -762,7 +755,7 @@ class SerialMotorsBus(MotorsBusBase):
         """Restore factory calibration for the selected motors.
 
         Homing offset is set to ``0`` and min/max position limits are set to the full usable range.
-        The in-memory `calibration` is cleared.
+        The in-memory :pyattr:`calibration` is cleared.
 
         Args:
             motors (NameOrID | Sequence[NameOrID] | None, optional): Selection of motors. `None` (default)
@@ -1076,9 +1069,9 @@ class SerialMotorsBus(MotorsBusBase):
     ) -> None:
         """Write a value to a single motor's register.
 
-        Contrary to [`~motors.motors_bus.MotorsBus.sync_write`], this expects a response status packet emitted by the motor, which
+        Contrary to :pymeth:`sync_write`, this expects a response status packet emitted by the motor, which
         provides a guarantee that the value was written to the register successfully. In consequence, it is
-        slower than [`~motors.motors_bus.MotorsBus.sync_write`] but it is more reliable. It should typically be used when configuring
+        slower than :pymeth:`sync_write` but it is more reliable. It should typically be used when configuring
         motors.
 
         Args:
@@ -1235,8 +1228,8 @@ class SerialMotorsBus(MotorsBusBase):
     ) -> None:
         """Write the same register on multiple motors.
 
-        Contrary to [`~motors.motors_bus.MotorsBus.write`], this *does not* expects a response status packet emitted by the motor, which
-        can allow for lost packets. It is faster than [`~motors.motors_bus.MotorsBus.write`] and should typically be used when
+        Contrary to :pymeth:`write`, this *does not* expects a response status packet emitted by the motor, which
+        can allow for lost packets. It is faster than :pymeth:`write` and should typically be used when
         frequency matters and losing some packets is acceptable (e.g. teleoperation loops).
 
         Args:

src/lerobot/optim/__init__.py

@@ -20,6 +20,7 @@ from .optimizers import (
     SGDConfig as SGDConfig,
     XVLAAdamWConfig as XVLAAdamWConfig,
     load_optimizer_state,
+    load_optimizer_state_dict,
     save_optimizer_state,
 )
 from .schedulers import (
@@ -50,6 +51,7 @@ __all__ = [
     "VQBeTSchedulerConfig",
     # State management
     "load_optimizer_state",
+    "load_optimizer_state_dict",
     "load_scheduler_state",
     "save_optimizer_state",
     "save_scheduler_state",

src/lerobot/optim/optimizers.py

@@ -27,7 +27,7 @@ from lerobot.utils.constants import (
     OPTIMIZER_PARAM_GROUPS,
     OPTIMIZER_STATE,
 )
-from lerobot.utils.io_utils import deserialize_json_into_object, write_json
+from lerobot.utils.io_utils import deserialize_json_into_object, load_json, write_json
 from lerobot.utils.utils import flatten_dict, unflatten_dict
 
 # Type alias for parameters accepted by optimizer build() methods.
@@ -52,11 +52,6 @@ class OptimizerConfig(draccus.ChoiceRegistry, abc.ABC):
     def type(self) -> str:
         return self.get_choice_name(self.__class__)
 
-    @property
-    def builds_multiple_optimizers(self) -> bool:
-        """True when build() returns a dict of optimizers (unsupported under sharded training)."""
-        return False
-
     @classmethod
     def default_choice_name(cls) -> str | None:
         return "adam"
@@ -250,10 +245,6 @@ class MultiAdamConfig(OptimizerConfig):
     grad_clip_norm: float = 10.0
     optimizer_groups: dict[str, dict[str, Any]] = field(default_factory=dict)
 
-    @property
-    def builds_multiple_optimizers(self) -> bool:
-        return True
-
     def build(self, params: OptimizerParams) -> dict[str, torch.optim.Optimizer]:
         """Build multiple Adam optimizers.
 
@@ -292,27 +283,35 @@ class MultiAdamConfig(OptimizerConfig):
 def save_optimizer_state(
     optimizer: torch.optim.Optimizer | dict[str, torch.optim.Optimizer],
     save_dir: Path,
+    optim_state_dict: dict | None = None,
 ) -> None:
-    """Save optimizer state to disk (non-sharded runs; sharded runs use the DCP channel).
+    """Save optimizer state to disk.
 
     Args:
         optimizer: Either a single optimizer or a dictionary of optimizers.
         save_dir: Directory to save the optimizer state.
+        optim_state_dict: Pre-gathered optimizer state dict (for FSDP, where the sharded state must
+            be gathered across ranks first). If provided, it is saved directly instead of calling
+            ``optimizer.state_dict()``. Only supported for a single optimizer. Defaults to None.
     """
     if isinstance(optimizer, dict):
         # Handle dictionary of optimizers
+        if optim_state_dict is not None:
+            raise ValueError("optim_state_dict is not supported for a dict of optimizers")
         for name, opt in optimizer.items():
             optimizer_dir = save_dir / name
             optimizer_dir.mkdir(exist_ok=True, parents=True)
             _save_single_optimizer_state(opt, optimizer_dir)
     else:
         # Handle single optimizer
-        _save_single_optimizer_state(optimizer, save_dir)
+        _save_single_optimizer_state(optimizer, save_dir, optim_state_dict=optim_state_dict)
 
 
-def _save_single_optimizer_state(optimizer: torch.optim.Optimizer, save_dir: Path) -> None:
+def _save_single_optimizer_state(
+    optimizer: torch.optim.Optimizer, save_dir: Path, optim_state_dict: dict | None = None
+) -> None:
     """Save a single optimizer's state to disk."""
-    state = optimizer.state_dict()
+    state = dict(optim_state_dict) if optim_state_dict is not None else optimizer.state_dict()
     param_groups = state.pop("param_groups")
     flat_state = flatten_dict(state)
     save_file(flat_state, save_dir / OPTIMIZER_STATE)
@@ -366,3 +365,19 @@ def _load_single_optimizer_state(optimizer: torch.optim.Optimizer, save_dir: Pat
 
     optimizer.load_state_dict(loaded_state_dict)
     return optimizer
+
+
+def load_optimizer_state_dict(save_dir: Path) -> dict:
+    """Read a saved optimizer state dict (safetensors + json) back into a plain dict.
+
+    Unlike `load_optimizer_state`, this does not load into an optimizer and preserves the original
+    ``state`` keys verbatim (e.g. FSDP parameter FQNs, which are not integer-castable). It is used by
+    the FSDP resume path, where the full state must be resharded via `FSDP.optim_state_dict_to_load`
+    before being loaded into the (sharded) optimizer.
+    """
+    flat_state = load_file(save_dir / OPTIMIZER_STATE)
+    state = unflatten_dict(flat_state)
+    return {
+        "state": state.get("state", {}),
+        "param_groups": load_json(save_dir / OPTIMIZER_PARAM_GROUPS),
+    }

src/lerobot/policies/act/modeling_act.py

@@ -47,8 +47,6 @@ class ACTPolicy(PreTrainedPolicy):
 
     config_class = ACTConfig
     name = "act"
-    # FSDP2 wrap units: one unit per transformer layer of both stacks.
-    _fsdp_wrap_modules = ["ACTEncoderLayer", "ACTDecoderLayer"]
 
     def __init__(
         self,

src/lerobot/policies/factory.py

@@ -131,16 +131,12 @@ class ProcessorConfigKwargs(TypedDict, total=False):
     This provides type hints for the optional arguments passed to `make_pre_post_processors`,
     improving code clarity and enabling static analysis.
 
-    **Attributes**:
-        - **preprocessor_config_filename** (`str | None`) -- The filename for the preprocessor configuration.
-        - **postprocessor_config_filename** (`str | None`) -- The filename for the postprocessor
-          configuration.
-        - **preprocessor_overrides** (`dict[str, Any] | None`) -- A dictionary of overrides for the
-          preprocessor configuration.
-        - **postprocessor_overrides** (`dict[str, Any] | None`) -- A dictionary of overrides for the
-          postprocessor configuration.
-        - **dataset_stats** (`dict[str, dict[str, torch.Tensor]] | None`) -- Dataset statistics for
-          normalization.
+    Attributes:
+        preprocessor_config_filename: The filename for the preprocessor configuration.
+        postprocessor_config_filename: The filename for the postprocessor configuration.
+        preprocessor_overrides: A dictionary of overrides for the preprocessor configuration.
+        postprocessor_overrides: A dictionary of overrides for the postprocessor configuration.
+        dataset_stats: Dataset statistics for normalization.
     """
 
     preprocessor_config_filename: str | None
@@ -246,7 +242,6 @@ def make_policy(
     ds_meta: LeRobotDatasetMetadata | None = None,
     env_cfg: EnvConfig | None = None,
     rename_map: dict[str, str] | None = None,
-    defer_weight_load: bool = False,
 ) -> PreTrainedPolicy:
     """
     Instantiate a policy model.
@@ -257,27 +252,22 @@ def make_policy(
     can either initialize a new policy from scratch or load a pretrained one.
 
     Args:
-        cfg (PreTrainedConfig): The configuration for the policy to be created. If
-            `cfg.pretrained_path` is set, the policy will be loaded with weights from that path.
-        ds_meta (LeRobotDatasetMetadata | None): Dataset metadata used to infer feature shapes and
-            types. Also provides statistics for normalization layers.
-        env_cfg (EnvConfig | None): Environment configuration used to infer feature shapes and
-            types. One of `ds_meta` or `env_cfg` must be provided.
-        rename_map (dict[str, str] | None): Optional mapping of dataset or environment feature
-            keys to match expected policy feature names (e.g., `"left"` → `"camera1"`).
-        defer_weight_load (bool): Build the exact policy `from_pretrained` would build — same
-            config resolution, same stats-derived buffers, same device placement and eval mode —
-            but skip the safetensors weight load. Used when resuming from a DCP checkpoint, whose
-            sharded weights stream in after `accelerator.prepare()` (the distributed checkpoint
-            engine overwrites the random init).
+        cfg: The configuration for the policy to be created. If `cfg.pretrained_path` is
+             set, the policy will be loaded with weights from that path.
+        ds_meta: Dataset metadata used to infer feature shapes and types. Also provides
+                 statistics for normalization layers.
+        env_cfg: Environment configuration used to infer feature shapes and types.
+                 One of `ds_meta` or `env_cfg` must be provided.
+        rename_map: Optional mapping of dataset or environment feature keys to match
+                 expected policy feature names (e.g., `"left"` → `"camera1"`).
 
     Returns:
-        PreTrainedPolicy: An instantiated and device-placed policy model.
+        An instantiated and device-placed policy model.
 
     Raises:
         ValueError: If both or neither of `ds_meta` and `env_cfg` are provided.
-        NotImplementedError: If attempting to use an unsupported policy-backend combination
-            (e.g., VQBeT with 'mps').
+        NotImplementedError: If attempting to use an unsupported policy-backend
+                             combination (e.g., VQBeT with 'mps').
     """
     if bool(ds_meta) == bool(env_cfg):
         raise ValueError("Either one of a dataset metadata or a sim env must be provided.")
@@ -342,18 +332,11 @@ def make_policy(
         )
 
     if cfg.pretrained_path and not cfg.use_peft:
-        if defer_weight_load:
-            # Same construction path as from_pretrained (config already resolved from the
-            # checkpoint by the caller; dataset_stats/dataset_meta kwargs identical), minus the
-            # weight load — parity by construction.
-            policy = policy_cls(**kwargs)
-            policy.eval()
-        else:
-            # Load a pretrained policy and override the config if needed (for example, if there
-            # are inference-time hyperparameters that we want to vary).
-            kwargs["pretrained_name_or_path"] = cfg.pretrained_path
-            kwargs["revision"] = cfg.pretrained_revision
-            policy = policy_cls.from_pretrained(**kwargs)
+        # Load a pretrained policy and override the config if needed (for example, if there are inference-time
+        # hyperparameters that we want to vary).
+        kwargs["pretrained_name_or_path"] = cfg.pretrained_path
+        kwargs["revision"] = cfg.pretrained_revision
+        policy = policy_cls.from_pretrained(**kwargs)
     elif cfg.pretrained_path and cfg.use_peft:
         # Load a pretrained PEFT model on top of the policy. The pretrained path points to the folder/repo
         # of the adapter and the adapter's config contains the path to the base policy. So we need the

src/lerobot/policies/fastwam/modeling_fastwam.py

@@ -54,9 +54,6 @@ class FastWAMPolicy(PreTrainedPolicy):
 
     config_class = FastWAMConfig
     name = "fastwam"
-    # FSDP2 wrap units: MoTLayer is the single FSDP owner of each layer's expert blocks
-    # (the blocks are re-parented onto it precisely so sharding has one boundary to hook).
-    _fsdp_wrap_modules = ["MoTLayer"]
 
     def __init__(
         self,

src/lerobot/policies/pi0_fast/modeling_pi0_fast.py

@@ -604,12 +604,6 @@ class PI0FastPytorch(nn.Module):  # see openpi `PI0Pytorch`
     ) -> torch.Tensor:
         """
         Optimized autoregressive decoding for FAST tokens using KV Caching.
-
-        Greedy decoding stops once every sequence emits the end-of-action marker. The
-        returned tensor keeps its fixed shape, with positions not generated after the
-        batch-wide stop left zero-filled. Stochastic decoding always runs to
-        ``max_decoding_steps`` so early stopping does not change the RNG state used by
-        subsequent calls.
         """
         if max_decoding_steps is None:
             max_decoding_steps = self.config.max_action_tokens
@@ -618,12 +612,6 @@ class PI0FastPytorch(nn.Module):  # see openpi `PI0Pytorch`
         device = tokens.device
         lm_head = self.paligemma_with_expert.paligemma.lm_head
 
-        # detokenize_actions() cuts at the first "|", so greedy decoding can stop once
-        # every sequence has emitted it. Keep stochastic decoding unchanged because
-        # skipping multinomial calls would shift the RNG state for subsequent calls.
-        end_of_action_token_id = self._paligemma_tokenizer.convert_tokens_to_ids("|")
-        finished = torch.zeros(bsize, dtype=torch.bool, device=device) if temperature == 0 else None
-
         # --- 1. PREFILL PHASE ---
         # Process Images + Text Prompt + BOS token once to populate the KV cache.
 
@@ -675,10 +663,6 @@ class PI0FastPytorch(nn.Module):  # see openpi `PI0Pytorch`
         # Initialize storage for generated tokens
         generated_action_tokens = torch.zeros((bsize, max_decoding_steps), dtype=torch.long, device=device)
         generated_action_tokens[:, 0] = next_token.squeeze(-1)
-        if finished is not None:
-            finished |= next_token.squeeze(-1) == end_of_action_token_id
-            if bool(finished.all()):
-                return generated_action_tokens
 
         # Track valid tokens mask (0 for pad, 1 for valid)
         # We need this to tell the new token what it can attend to (images + text + past actions)
@@ -729,11 +713,6 @@ class PI0FastPytorch(nn.Module):  # see openpi `PI0Pytorch`
 
             generated_action_tokens[:, t] = next_token.squeeze(-1)
 
-            if finished is not None:
-                finished |= next_token.squeeze(-1) == end_of_action_token_id
-                if bool(finished.all()):
-                    break
-
         return generated_action_tokens
 
 

src/lerobot/policies/pretrained.py

@@ -18,17 +18,20 @@ import builtins
 import dataclasses
 import logging
 import os
-import warnings
+from importlib.resources import files
 from pathlib import Path
-from typing import TYPE_CHECKING, Any, ClassVar, TypedDict, TypeVar, Unpack
+from tempfile import TemporaryDirectory
+from typing import TYPE_CHECKING, TypedDict, TypeVar, Unpack
 
-from huggingface_hub import hf_hub_download, save_torch_state_dict
+from huggingface_hub import HfApi, ModelCard, ModelCardData, hf_hub_download, save_torch_state_dict
 from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE
 from huggingface_hub.errors import HfHubHTTPError
-from safetensors.torch import load_model as load_model_as_safetensor
+from safetensors.torch import load_model as load_model_as_safetensor, save_model as save_model_as_safetensor
 from torch import Tensor, nn
 
+from lerobot.__version__ import __version__
 from lerobot.configs import PreTrainedConfig
+from lerobot.configs.train import TrainPipelineConfig
 from lerobot.utils.device_utils import resolve_safetensors_device
 from lerobot.utils.hub import HubMixin
 from lerobot.utils.import_utils import _peft_available, require_package
@@ -43,14 +46,56 @@ else:
     get_peft_model = None
 
 if TYPE_CHECKING:
-    from lerobot.configs.train import TrainPipelineConfig
     from lerobot.datasets.dataset_metadata import LeRobotDatasetMetadata
 
 T = TypeVar("T", bound="PreTrainedPolicy")
 
-# Pinned far above any policy's total size so save_torch_state_dict always emits exactly one
-# `model.safetensors` (no shards, no index) — a constant, not a computed byte count.
-_SINGLE_FILE_SHARD_SIZE = "1TB"
+
+def _build_card_context(
+    cfg: TrainPipelineConfig | None,
+    dataset_meta: LeRobotDatasetMetadata | None,
+    input_features: dict | None,
+    output_features: dict | None,
+) -> dict:
+    """Collect optional data for the model-card template.
+
+    Returns plain values only (no Markdown) — the template in
+    ``lerobot/templates/lerobot_modelcard_template.md`` decides how and whether to show
+    each one. Everything is best-effort: anything unavailable is left empty/None and the
+    template simply skips that section, so this never breaks a Hub push.
+    """
+    context = {
+        "training": None,
+        "input_features": input_features or {},
+        "output_features": output_features or {},
+        "dataset": None,
+        "robot_type": None,
+        "cameras": [],
+    }
+
+    if cfg is not None:
+        optimizer = getattr(cfg, "optimizer", None)
+        context["training"] = {
+            "steps": cfg.steps,
+            "batch_size": cfg.batch_size,
+            "seed": cfg.seed,
+            "optimizer": getattr(optimizer, "type", None) if optimizer else None,
+            "lr": getattr(optimizer, "lr", None) if optimizer else None,
+            "lerobot_version": __version__,
+        }
+
+    if dataset_meta is not None:
+        context["dataset"] = {
+            "repo_id": dataset_meta.repo_id,
+            "episodes": dataset_meta.total_episodes,
+            "frames": dataset_meta.total_frames,
+            "fps": dataset_meta.fps,
+            "tasks": [str(task) for task in dataset_meta.tasks.index],
+        }
+        context["robot_type"] = dataset_meta.robot_type
+        context["cameras"] = [key.split(".")[-1] for key in dataset_meta.camera_keys]
+
+    return context
 
 
 class ActionSelectKwargs(TypedDict, total=False):
@@ -65,22 +110,6 @@ class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):
     config_class: None
     name: None
 
-    # --- declarative parallelism/acceleration surface ----------------------------------------
-    # Module CLASS names forming the FSDP2 wrap units (and, once wired, the activation-
-    # checkpointing units). Resolved onto the accelerate plugin right before
-    # `accelerator.prepare()` by `lerobot.distributed.set_fsdp_wrap_modules`; sharded training
-    # with no wrap source anywhere fails loudly instead of silently wrapping only the root.
-    _fsdp_wrap_modules: ClassVar[list[str] | None] = None
-    # Non-`forward` entry points that must trigger FSDP2 unshard/reshard hooks when called on a
-    # sharded policy (registered post-prepare via `torch.distributed.fsdp
-    # .register_fsdp_forward_method`); calling them unregistered crashes on mixed Tensor/DTensor.
-    _fsdp_forward_methods: ClassVar[tuple[str, ...]] = ("select_action", "predict_action_chunk")
-    # Capability gate for the (future) activation-checkpointing wiring.
-    supports_gradient_checkpointing: ClassVar[bool] = False
-    # Declarative context-parallel plan (diffusers `ContextParallelModelPlan` semantics:
-    # module FQN -> sequence split/gather spec). Reserved for the CP engine round.
-    _cp_plan: ClassVar[dict[str, Any] | None] = None
-
     def __init__(self, config: PreTrainedConfig, *inputs, **kwargs):
         super().__init__()
         if not isinstance(config, PreTrainedConfig):
@@ -98,33 +127,43 @@ class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):
         if not getattr(cls, "name", None):
             raise TypeError(f"Class {cls.__name__} must define 'name'")
 
-    def _save_pretrained(self, save_directory: Path) -> None:
-        """Serialize this policy's parameters (and config) into `save_directory`.
+    def save_pretrained(
+        self,
+        save_directory: str | Path,
+        *,
+        state_dict: dict[str, Tensor] | None = None,
+        repo_id: str | None = None,
+        push_to_hub: bool = False,
+        card_kwargs: dict | None = None,
+        **push_to_hub_kwargs,
+    ) -> str | None:
+        """Save the policy to a directory (and optionally push to the Hub).
 
-        Sharding is handled internally: under FSDP2 the full state dict is gathered through a
-        COLLECTIVE, so when the policy is sharded this method (via `save_pretrained`) must be
-        called on EVERY rank — a rank-0-gated call deadlocks. File writes happen on the main
-        process only, in all layouts (single, DDP, sharded).
-
-        Args:
-            save_directory (Path): Target directory for the policy config (`config.json`) and the
-                safetensors weight file(s).
+        Overrides `HubMixin.save_pretrained` to add a `state_dict` argument (mirroring
+        `transformers.PreTrainedModel.save_pretrained`). Under FSDP, `self.state_dict()` would
+        return sharded tensors, so the caller gathers the full state dict via a cross-rank
+        collective and passes it here for `_save_pretrained` to write directly.
         """
-        # Lazy imports: the persistence layer pulls in lerobot.distributed only when saving.
-        from lerobot.distributed.checkpoint import full_model_state_dict, is_sharded_module
-        from lerobot.distributed.utils import is_main_process
+        save_directory = Path(save_directory)
+        save_directory.mkdir(parents=True, exist_ok=True)
+        self._save_pretrained(save_directory, state_dict=state_dict)
+        if push_to_hub:
+            if repo_id is None:
+                repo_id = save_directory.name
+            return self.push_to_hub(repo_id=repo_id, card_kwargs=card_kwargs, **push_to_hub_kwargs)
+        return None
 
-        model_to_save = self.module if hasattr(self, "module") else self
-        if is_sharded_module(model_to_save):
-            logging.info("Gathering the full state dict from all ranks (sharded policy).")
-        state_dict = full_model_state_dict(model_to_save)  # collective when sharded; {} off-main
-        if not state_dict or not is_main_process():
-            # Sharded: the gather materializes on the main rank only (emptiness check).
-            # Non-sharded multi-rank (DDP): every rank holds a full dict — the explicit rank
-            # gate prevents N ranks racing on the same files. Single process: never taken.
-            return
+    def _save_pretrained(self, save_directory: Path, state_dict: dict[str, Tensor] | None = None) -> None:
         self.config._save_pretrained(save_directory)
-        save_torch_state_dict(state_dict, str(save_directory), max_shard_size=_SINGLE_FILE_SHARD_SIZE)
+        model_to_save = self.module if hasattr(self, "module") else self
+        if state_dict is None:
+            save_model_as_safetensor(model_to_save, str(save_directory / SAFETENSORS_SINGLE_FILE))
+            return
+        # A pre-gathered (e.g. FSDP full) state dict was supplied: write it directly.
+        # `save_torch_state_dict` discards shared-tensor duplicates just like `save_model` does;
+        # pin `max_shard_size` above the total size so the output stays a single `model.safetensors`
+        total_bytes = sum(t.numel() * t.element_size() for t in state_dict.values())
+        save_torch_state_dict(state_dict, str(save_directory), max_shard_size=max(total_bytes, 1))
 
     @classmethod
     def from_pretrained(
@@ -252,39 +291,92 @@ class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):
         peft_model=None,
         state_dict: dict[str, Tensor] | None = None,
         dataset_meta: LeRobotDatasetMetadata | None = None,
-    ) -> None:
-        """Publish this policy to the Hub.
+    ):
+        api = HfApi()
+        repo_id = api.create_repo(
+            repo_id=self.config.repo_id, private=self.config.private, exist_ok=True
+        ).repo_id
 
-        Deprecated: use :func:`lerobot.common.train_utils.publish_trained_model` instead, which
-        also publishes the pre/post-processors alongside the model.
+        # Push the files to the repo in a single commit
+        with TemporaryDirectory(ignore_cleanup_errors=True) as tmp:
+            saved_path = Path(tmp) / repo_id
 
-        Args:
-            cfg (TrainPipelineConfig): The training config; saved as `train_config.json` and
-                used to render the model card.
-            peft_model: The PEFT wrapper when training adapters, whose weights replace the full
-                model weights in the published repo. Defaults to None.
-            state_dict (dict[str, Tensor] | None): Ignored; weights are now gathered internally
-                when the policy is sharded. Defaults to None.
-            dataset_meta (LeRobotDatasetMetadata | None): Dataset metadata for the model card,
-                if available. Defaults to None.
-        """
-        from lerobot.common.train_utils import publish_trained_model
+            if peft_model is not None:
+                # Since PEFT just forwards calls to `push_model_to_hub`, `self` is not the PeftModel wrapper
+                # but the actual policy which is why we need the PEFT model passed to us to save the adapter.
+                # That also means that we need to store the policy config ourselves since PEFT can't.
+                peft_model.save_pretrained(saved_path)
+                self.config.save_pretrained(saved_path)
+            else:
+                # Calls _save_pretrained and stores model tensors
+                self.save_pretrained(saved_path, state_dict=state_dict)
 
-        warnings.warn(
-            "PreTrainedPolicy.push_model_to_hub is deprecated and will be removed in a future "
-            "version. Use lerobot.common.train_utils.publish_trained_model(cfg, model, "
-            "preprocessor, postprocessor, dataset_meta) instead.",
-            FutureWarning,
-            stacklevel=2,
-        )
-        if state_dict is not None:
-            warnings.warn(
-                "The `state_dict` argument is ignored: sharded weights are gathered internally "
-                "when the policy is saved.",
-                FutureWarning,
-                stacklevel=2,
+            card = self.generate_model_card(
+                cfg.dataset.repo_id,
+                self.config.type,
+                self.config.license,
+                self.config.tags,
+                cfg=cfg,
+                dataset_meta=dataset_meta,
             )
-        publish_trained_model(cfg, self, None, None, dataset_meta, peft_model=peft_model)
+            card.save(str(saved_path / "README.md"))
+
+            cfg.save_pretrained(saved_path)  # Calls _save_pretrained and stores train config
+
+            commit_info = api.upload_folder(
+                repo_id=repo_id,
+                repo_type="model",
+                folder_path=saved_path,
+                commit_message="Upload policy weights, train config and readme",
+                allow_patterns=["*.safetensors", "*.json", "*.yaml", "*.md"],
+                ignore_patterns=["*.tmp", "*.log"],
+            )
+
+            # Contract: lerobot.jobs.hf.submit_to_hf watches for this exact
+            # "Model pushed to <url>" line to end a remote run early. Keep the wording
+            # and URL format in sync (it falls back to status polling if they drift).
+            logging.info(f"Model pushed to {commit_info.repo_url.url}")
+
+    def generate_model_card(
+        self,
+        dataset_repo_id: str,
+        model_type: str,
+        license: str | None,
+        tags: list[str] | None,
+        cfg: TrainPipelineConfig | None = None,
+        dataset_meta: LeRobotDatasetMetadata | None = None,
+    ) -> ModelCard:
+        base_model_mapping = {
+            "smolvla": "lerobot/smolvla_base",
+            "pi0": "lerobot/pi0_base",
+            "pi05": "lerobot/pi05_base",
+            "pi0_fast": "lerobot/pi0fast-base",
+            "xvla": "lerobot/xvla-base",
+        }
+
+        card_data = ModelCardData(
+            license=license or "apache-2.0",
+            library_name="lerobot",
+            pipeline_tag="robotics",
+            tags=list(set(tags or []).union({"robotics", "lerobot", model_type})),
+            model_name=model_type,
+            datasets=dataset_repo_id,
+            base_model=base_model_mapping.get(model_type),
+        )
+
+        context = _build_card_context(
+            cfg, dataset_meta, self.config.input_features, self.config.output_features
+        )
+        # Used by the template to pre-fill commands and the "Fine-tuned from" line.
+        context["policy_repo_id"] = getattr(self.config, "repo_id", None)
+        context["base_model"] = base_model_mapping.get(model_type)
+
+        template_card = (
+            files("lerobot.templates").joinpath("lerobot_modelcard_template.md").read_text(encoding="utf-8")
+        )
+        card = ModelCard.from_template(card_data, template_str=template_card, **context)
+        card.validate()
+        return card
 
     def wrap_with_peft(
         self,

src/lerobot/policies/rtc/action_queue.py

@@ -46,11 +46,10 @@ class ActionQueue:
     Args:
         cfg (RTCConfig): Configuration for Real-Time Chunking behavior.
 
-    **Attributes**:
-        - **queue** (`Tensor | None`) -- Processed actions for robot rollout (time_steps, action_dim).
-        - **original_queue** (`Tensor | None`) -- Original actions for RTC computation (time_steps,
-          action_dim).
-        - **last_index** (`int`) -- Current consumption index in the queue.
+    Attributes:
+        queue (Tensor | None): Processed actions for robot rollout (time_steps, action_dim).
+        original_queue (Tensor | None): Original actions for RTC computation (time_steps, action_dim).
+        last_index (int): Current consumption index in the queue.
     """
 
     def __init__(self, cfg: RTCConfig):

src/lerobot/policies/rtc/debug_tracker.py

@@ -27,19 +27,19 @@ from torch import Tensor
 class DebugStep:
     """Container for debug information from a single denoising step.
 
-    **Attributes**:
-        - **step_idx** (`int`) -- Step index/counter.
-        - **x_t** (`Tensor | None`) -- Current latent/state tensor.
-        - **v_t** (`Tensor | None`) -- Velocity from denoiser.
-        - **x1_t** (`Tensor | None`) -- Denoised prediction (x_t - time * v_t).
-        - **correction** (`Tensor | None`) -- Correction gradient tensor.
-        - **err** (`Tensor | None`) -- Weighted error term.
-        - **weights** (`Tensor | None`) -- Prefix attention weights.
-        - **guidance_weight** (`float | Tensor | None`) -- Applied guidance weight.
-        - **time** (`float | Tensor | None`) -- Time parameter.
-        - **inference_delay** (`int | None`) -- Inference delay parameter.
-        - **execution_horizon** (`int | None`) -- Execution horizon parameter.
-        - **metadata** (`dict[str, Any]`) -- Additional metadata.
+    Attributes:
+        step_idx (int): Step index/counter.
+        x_t (Tensor | None): Current latent/state tensor.
+        v_t (Tensor | None): Velocity from denoiser.
+        x1_t (Tensor | None): Denoised prediction (x_t - time * v_t).
+        correction (Tensor | None): Correction gradient tensor.
+        err (Tensor | None): Weighted error term.
+        weights (Tensor | None): Prefix attention weights.
+        guidance_weight (float | Tensor | None): Applied guidance weight.
+        time (float | Tensor | None): Time parameter.
+        inference_delay (int | None): Inference delay parameter.
+        execution_horizon (int | None): Execution horizon parameter.
+        metadata (dict[str, Any]): Additional metadata.
     """
 
     step_idx: int = 0

src/lerobot/processor/batch_processor.py

@@ -175,6 +175,9 @@ class AddBatchDimensionComplementaryDataStep(ComplementaryDataProcessorStep):
             if isinstance(task_index_value, Tensor) and task_index_value.dim() == 0:
                 complementary_data["task_index"] = task_index_value.unsqueeze(0)
 
+        complementary_data.pop("language_persistent", None)
+        complementary_data.pop("language_events", None)
+
         if "messages" in complementary_data:
             messages = complementary_data["messages"]
             if isinstance(messages, list) and (not messages or isinstance(messages[0], dict)):
@@ -217,12 +220,10 @@ class AddBatchDimensionProcessorStep(ProcessorStep):
     This step combines individual processors for actions, observations, and complementary data
     to create a batched transition (batch size 1) from a single-instance transition.
 
-    **Attributes**:
-        - **to_batch_action_processor** (`AddBatchDimensionActionStep`) -- Processor for the action component.
-        - **to_batch_observation_processor** (`AddBatchDimensionObservationStep`) -- Processor for the
-          observation component.
-        - **to_batch_complementary_data_processor** (`AddBatchDimensionComplementaryDataStep`) -- Processor
-          for the complementary data component.
+    Attributes:
+        to_batch_action_processor: Processor for the action component.
+        to_batch_observation_processor: Processor for the observation component.
+        to_batch_complementary_data_processor: Processor for the complementary data component.
     """
 
     to_batch_action_processor: AddBatchDimensionActionStep = field(

src/lerobot/processor/delta_action_processor.py

@@ -32,8 +32,9 @@ class MapTensorToDeltaActionDictStep(ActionProcessorStep):
     It decomposes the vector into named components for delta movements of the
     end-effector (x, y, z) and optionally the gripper.
 
-    **Attributes**:
-        - **use_gripper** (`bool`) -- If True, assumes the 4th element of the tensor is the gripper action.
+    Attributes:
+        use_gripper: If True, assumes the 4th element of the tensor is the
+                     gripper action.
     """
 
     use_gripper: bool = True
@@ -80,10 +81,10 @@ class MapDeltaActionToRobotActionStep(RobotActionProcessorStep):
     into a target action format that includes an "enabled" flag and target
     end-effector positions. It also handles scaling and noise filtering.
 
-    **Attributes**:
-        - **position_scale** (`float`) -- A factor to scale the delta position inputs.
-        - **noise_threshold** (`float`) -- The magnitude below which delta inputs are considered noise and do
-          not trigger an "enabled" state.
+    Attributes:
+        position_scale: A factor to scale the delta position inputs.
+        noise_threshold: The magnitude below which delta inputs are considered noise
+                         and do not trigger an "enabled" state.
     """
 
     # Scale factors for delta movements

src/lerobot/processor/device_processor.py

@@ -40,10 +40,10 @@ class DeviceProcessorStep(ProcessorStep):
 
     This is crucial for preparing data for model training or inference on hardware like GPUs.
 
-    **Attributes**:
-        - **device** (`str`) -- The target device for tensors (e.g., "cpu", "cuda", "cuda:0").
-        - **float_dtype** (`str | None`) -- The target floating-point dtype as a string (e.g., "float32",
-          "float16", "bfloat16"). If None, the dtype is not changed.
+    Attributes:
+        device: The target device for tensors (e.g., "cpu", "cuda", "cuda:0").
+        float_dtype: The target floating-point dtype as a string (e.g., "float32", "float16", "bfloat16").
+                     If None, the dtype is not changed.
     """
 
     device: str = "cpu"

src/lerobot/processor/gym_action_processor.py

@@ -33,9 +33,10 @@ class Torch2NumpyActionProcessorStep(ActionProcessorStep):
     This step is useful when the output of a policy (typically a torch.Tensor)
     needs to be passed to an environment or component that expects a NumPy array.
 
-    **Attributes**:
-        - **squeeze_batch_dim** (`bool`) -- If True, removes the first dimension of the array if it is of size
-          1. This is useful for converting a batched action of size (1, D) to a single action of size (D,).
+    Attributes:
+        squeeze_batch_dim: If True, removes the first dimension of the array
+                           if it is of size 1. This is useful for converting a
+                           batched action of size (1, D) to a single action of size (D,).
     """
 
     squeeze_batch_dim: bool = True

src/lerobot/processor/hil_processor.py

@@ -101,8 +101,8 @@ class AddTeleopActionAsComplimentaryDataStep(ComplementaryDataProcessorStep):
     be available to downstream processors, for example, to override a policy's action
     during an intervention.
 
-    **Attributes**:
-        - **teleop_device** (`Teleoperator`) -- The teleoperator instance to get the action from.
+    Attributes:
+        teleop_device: The teleoperator instance to get the action from.
     """
 
     teleop_device: "Teleoperator"
@@ -137,9 +137,9 @@ class AddTeleopEventsAsInfoStep(InfoProcessorStep):
     This step extracts control events from teleoperators that support event-based
     interaction, making these signals available to other parts of the system.
 
-    **Attributes**:
-        - **teleop_device** (`TeleopWithEvents`) -- An instance of a teleoperator that implements the
-          `HasTeleopEvents` protocol.
+    Attributes:
+        teleop_device: An instance of a teleoperator that implements the
+                       `HasTeleopEvents` protocol.
     """
 
     teleop_device: TeleopWithEvents
@@ -180,10 +180,10 @@ class ImageCropResizeProcessorStep(ObservationProcessorStep):
     the specified transformations. It handles device placement, moving tensors to the
     CPU if necessary for operations not supported on certain accelerators like MPS.
 
-    **Attributes**:
-        - **crop_params_dict** (`dict[str, tuple[int, int, int, int]] | None`) -- A dictionary mapping image
-          keys to cropping parameters (top, left, height, width).
-        - **resize_size** (`tuple[int, int] | None`) -- A tuple (height, width) to resize all images to.
+    Attributes:
+        crop_params_dict: A dictionary mapping image keys to cropping parameters
+                          (top, left, height, width).
+        resize_size: A tuple (height, width) to resize all images to.
     """
 
     crop_params_dict: dict[str, tuple[int, int, int, int]] | None = None
@@ -267,9 +267,9 @@ class TimeLimitProcessorStep(TruncatedProcessorStep):
     """
     Tracks episode steps and enforces a time limit by truncating the episode.
 
-    **Attributes**:
-        - **max_episode_steps** (`int`) -- The maximum number of steps allowed per episode.
-        - **current_step** (`int`) -- The current step count for the active episode.
+    Attributes:
+        max_episode_steps: The maximum number of steps allowed per episode.
+        current_step: The current step count for the active episode.
     """
 
     max_episode_steps: int
@@ -358,11 +358,11 @@ class GripperPenaltyProcessorStep(ProcessorStep):
     This discourages gripper oscillation while leaving "stay" and saturating-further
     commands unpenalized.
 
-    **Attributes**:
-        - **penalty** (`float`) -- The negative reward value to apply.
-        - **max_gripper_pos** (`float`) -- The maximum position value for the gripper, used for normalization.
-        - **open_threshold** (`float`) -- Normalized state below which the gripper is considered "open".
-        - **closed_threshold** (`float`) -- Normalized state above which the gripper is considered "closed".
+    Attributes:
+        penalty: The negative reward value to apply.
+        max_gripper_pos: The maximum position value for the gripper, used for normalization.
+        open_threshold: Normalized state below which the gripper is considered "open".
+        closed_threshold: Normalized state above which the gripper is considered "closed".
     """
 
     penalty: float = -0.02
@@ -456,10 +456,10 @@ class InterventionActionProcessorStep(ProcessorStep):
     this step replaces the policy's action with the human's teleoperated action.
     It also processes signals to terminate the episode or flag success.
 
-    **Attributes**:
-        - **use_gripper** (`bool`) -- Whether to include the gripper in the teleoperated action.
-        - **terminate_on_success** (`bool`) -- If True, automatically sets the `done` flag when a `success`
-          event is received.
+    Attributes:
+        use_gripper: Whether to include the gripper in the teleoperated action.
+        terminate_on_success: If True, automatically sets the `done` flag when a
+                              `success` event is received.
     """
 
     use_gripper: bool = False
@@ -557,13 +557,13 @@ class RewardClassifierProcessorStep(ProcessorStep):
     This step uses a model to determine if the current state is successful, updating
     the reward and potentially terminating the episode.
 
-    **Attributes**:
-        - **pretrained_path** (`str | None`) -- Path to the pretrained reward classifier model.
-        - **device** (`str`) -- The device to run the classifier on.
-        - **success_threshold** (`float`) -- The probability threshold to consider a prediction as successful.
-        - **success_reward** (`float`) -- The reward value to assign on success.
-        - **terminate_on_success** (`bool`) -- If True, terminates the episode upon successful classification.
-        - **reward_classifier** (`Any`) -- The loaded classifier model instance.
+    Attributes:
+        pretrained_path: Path to the pretrained reward classifier model.
+        device: The device to run the classifier on.
+        success_threshold: The probability threshold to consider a prediction as successful.
+        success_reward: The reward value to assign on success.
+        terminate_on_success: If True, terminates the episode upon successful classification.
+        reward_classifier: The loaded classifier model instance.
     """
 
     pretrained_path: str | None = None

src/lerobot/processor/migrate_policy_normalization.py

@@ -647,15 +647,10 @@ def main():
     tags = set(tags).union({"robotics", "lerobot", policy_type})
     tags = list(tags)
 
-    # Generate model card through the free helper (PreTrainedPolicy.generate_model_card was
-    # removed with the publisher redesign), then apply the metadata recovered above — the
-    # migrated policy config does not carry the original repo's card fields.
-    from lerobot.common.train_utils import generate_model_card
-
-    card = generate_model_card(policy.config)
-    card.data.datasets = dataset_repo_id
-    card.data.license = license
-    card.data.tags = sorted(tags)
+    # Generate model card
+    card = policy.generate_model_card(
+        dataset_repo_id=dataset_repo_id, model_type=policy_type, license=license, tags=tags
+    )
 
     # Save model card locally
     card.save(str(output_dir / "README.md"))

src/lerobot/processor/normalize_processor.py

@@ -71,23 +71,22 @@ class _NormalizationMixin:
         )
         ```
 
-    **Attributes**:
-        - **features** (`dict[str, PolicyFeature]`) -- A dictionary mapping feature names to `PolicyFeature`
-          objects, defining the data structure to be processed.
-        - **norm_map** (`dict[FeatureType, NormalizationMode]`) -- A dictionary mapping `FeatureType` to
-          `NormalizationMode`, specifying which normalization method to use for each type of feature.
-        - **stats** (`dict[str, dict[str, Any]] | None`) -- A dictionary containing the normalization
-          statistics (e.g., mean, std, min, max) for each feature.
-        - **device** (`torch.device | str | None`) -- The PyTorch device on which to store and perform tensor
-          operations.
-        - **eps** (`float`) -- A small epsilon value to prevent division by zero in normalization
-          calculations.
-        - **normalize_observation_keys** (`set[str] | None`) -- An optional set of keys to selectively apply
-          normalization to specific observation features.
-        - **_tensor_stats** (`dict[str, dict[str, Tensor]]`) -- An internal dictionary holding the
-          normalization statistics as PyTorch tensors.
-        - **_stats_explicitly_provided** (`bool`) -- Internal flag tracking whether stats were explicitly
-          provided during construction (used for override preservation).
+    Attributes:
+        features: A dictionary mapping feature names to `PolicyFeature` objects, defining
+            the data structure to be processed.
+        norm_map: A dictionary mapping `FeatureType` to `NormalizationMode`, specifying
+            which normalization method to use for each type of feature.
+        stats: A dictionary containing the normalization statistics (e.g., mean, std,
+            min, max) for each feature.
+        device: The PyTorch device on which to store and perform tensor operations.
+        eps: A small epsilon value to prevent division by zero in normalization
+            calculations.
+        normalize_observation_keys: An optional set of keys to selectively apply
+            normalization to specific observation features.
+        _tensor_stats: An internal dictionary holding the normalization statistics as
+            PyTorch tensors.
+        _stats_explicitly_provided: Internal flag tracking whether stats were explicitly
+            provided during construction (used for override preservation).
     """
 
     features: dict[str, PolicyFeature]

src/lerobot/processor/pipeline.py

@@ -41,7 +41,7 @@ from pathlib import Path
 from typing import Any, TypedDict, TypeVar, cast
 
 import torch
-from huggingface_hub import hf_hub_download, snapshot_download
+from huggingface_hub import hf_hub_download
 from safetensors.torch import load_file, save_file
 
 from lerobot.configs import FeatureType, PipelineFeatureType, PolicyFeature
@@ -212,10 +212,6 @@ class ProcessorStep(ABC):
         """
         return None
 
-    def save_artifacts(self, save_directory: Path) -> dict[str, str]:
-        """Save non-tensor assets and map constructor arguments to relative paths."""
-        return {}
-
     def reset(self) -> None:
         """Resets the internal state of the processor step, if any."""
         return None
@@ -269,18 +265,13 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
     data processing workflow. It's generic, allowing for custom input and output types,
     which are handled by the `to_transition` and `to_output` converters.
 
-    **Attributes**:
-        - **steps** (`Sequence[ProcessorStep]`) -- A sequence of `ProcessorStep` objects that make up the
-          pipeline.
-        - **name** (`str`) -- A descriptive name for the pipeline.
-        - **to_transition** (`Callable[[TInput], EnvTransition]`) -- A function to convert raw input data into
-          the standardized `EnvTransition` format.
-        - **to_output** (`Callable[[EnvTransition], TOutput]`) -- A function to convert the final
-          `EnvTransition` into the desired output format.
-        - **before_step_hooks** (`list[Callable[[int, EnvTransition], None]]`) -- A list of functions to be
-          called before each step is executed.
-        - **after_step_hooks** (`list[Callable[[int, EnvTransition], None]]`) -- A list of functions to be
-          called after each step is executed.
+    Attributes:
+        steps: A sequence of `ProcessorStep` objects that make up the pipeline.
+        name: A descriptive name for the pipeline.
+        to_transition: A function to convert raw input data into the standardized `EnvTransition` format.
+        to_output: A function to convert the final `EnvTransition` into the desired output format.
+        before_step_hooks: A list of functions to be called before each step is executed.
+        after_step_hooks: A list of functions to be called after each step is executed.
     """
 
     steps: Sequence[ProcessorStep] = field(default_factory=list)
@@ -565,22 +556,6 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
         pipeline_config = self.get_config()
         pipeline_state_dict = self.state_dict()
 
-        for processor_step, step_entry in zip(self.steps, pipeline_config["steps"], strict=True):
-            artifacts = processor_step.save_artifacts(save_directory)
-            if artifacts:
-                for config_key, relative_path in artifacts.items():
-                    artifact_path = Path(relative_path)
-                    if artifact_path.is_absolute() or ".." in artifact_path.parts:
-                        raise ValueError(
-                            f"Processor artifact path must be relative to the checkpoint: {relative_path!r}"
-                        )
-                    if not (save_directory / artifact_path).exists():
-                        raise FileNotFoundError(
-                            f"Processor step did not save declared artifact '{relative_path}'"
-                        )
-                    step_entry["config"][config_key] = artifact_path.as_posix()
-                step_entry["artifacts"] = artifacts
-
         for state_key, step_state_dict in pipeline_state_dict.items():
             state_filename = f"{state_key}.safetensors"
             save_file(step_state_dict, save_directory / state_filename)
@@ -765,13 +740,7 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
 
         # 3. Build steps with overrides
         steps, validated_overrides = cls._build_steps_with_overrides(
-            loaded_config,
-            overrides or {},
-            model_id,
-            base_path,
-            config_filename,
-            hub_download_kwargs,
-            is_local_source,
+            loaded_config, overrides or {}, model_id, base_path, hub_download_kwargs, is_local_source
         )
 
         # 4. Validate that all overrides were used
@@ -967,7 +936,6 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
         overrides: dict[str, Any],
         model_id: str,
         base_path: Path | None,
-        config_filename: str,
         hub_download_kwargs: dict[str, Any],
         is_local_source: bool = False,
     ) -> tuple[list[ProcessorStep], set[str]]:
@@ -977,11 +945,6 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
 
         **For each step in loaded_config["steps"]**:
 
-        0. **Artifact Resolution** (via _resolve_artifact_paths):
-           - Resolve declared relative artifact paths against a local checkpoint
-           - Download declared artifacts when loading the pipeline from the Hub
-           - Reject absolute paths and path traversal before step construction
-
         1. **Class Resolution** (via _resolve_step_class):
            - **If "registry_name" exists**: Look up in ProcessorStepRegistry
              Example: {"registry_name": "normalize_step"} -> Get registered class
@@ -1015,8 +978,6 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
             overrides: User-provided parameter overrides (keyed by class/registry name)
             model_id: The model identifier (needed for Hub state file downloads)
             base_path: Local directory path for finding state files
-            config_filename: Processor config path, used as the repository-relative
-                base for state files and declared artifacts.
             hub_download_kwargs: Parameters for hf_hub_download (tokens, cache, etc.)
             is_local_source: Whether model_id resolved to a local directory or config file.
 
@@ -1029,80 +990,15 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
             ImportError: If a step class cannot be imported or found in registry
             ValueError: If a step cannot be instantiated with its configuration
         """
-        loaded_config = deepcopy(loaded_config)
-        cls._resolve_artifact_paths(
-            loaded_config,
-            model_id,
-            base_path,
-            config_filename,
-            hub_download_kwargs,
-        )
         steps, remaining_override_keys = cls._build_steps_from_config(loaded_config, overrides)
 
         for step_instance, step_entry in zip(steps, loaded_config["steps"], strict=True):
             cls._load_step_state(
-                step_instance,
-                step_entry,
-                model_id,
-                base_path,
-                config_filename,
-                hub_download_kwargs,
-                is_local_source,
+                step_instance, step_entry, model_id, base_path, hub_download_kwargs, is_local_source
             )
 
         return steps, remaining_override_keys
 
-    @classmethod
-    def _resolve_artifact_paths(
-        cls,
-        loaded_config: dict[str, Any],
-        model_id: str,
-        base_path: Path | None,
-        config_filename: str,
-        hub_download_kwargs: dict[str, Any],
-    ) -> None:
-        """Resolve declared relative processor artifacts before step construction.
-
-        Args:
-            loaded_config: Mutable processor configuration containing step artifact declarations.
-            model_id: Local checkpoint path or Hub model identifier.
-            base_path: Local directory containing the resolved processor configuration.
-            config_filename: Processor config path, whose parent is the artifact root on the Hub.
-            hub_download_kwargs: Authentication, revision, and cache arguments for Hub downloads.
-
-        Raises:
-            ValueError: If a declared artifact path is absolute or escapes the checkpoint.
-            FileNotFoundError: If a declared artifact cannot be found locally or downloaded.
-        """
-        is_local = Path(model_id).is_dir() or Path(model_id).is_file()
-
-        for step_entry in loaded_config["steps"]:
-            artifacts = step_entry.get("artifacts", {})
-            for config_key, relative_path in artifacts.items():
-                artifact_path = Path(relative_path)
-                if artifact_path.is_absolute() or ".." in artifact_path.parts:
-                    raise ValueError(
-                        f"Processor artifact path must be relative to the checkpoint: {relative_path!r}"
-                    )
-
-                resolved_path = base_path / artifact_path if base_path is not None else artifact_path
-                if not resolved_path.exists() and not is_local:
-                    repository_path = Path(config_filename).parent / artifact_path
-                    snapshot_download(
-                        repo_id=model_id,
-                        repo_type="model",
-                        allow_patterns=f"{repository_path.as_posix()}/**",
-                        **hub_download_kwargs,
-                    )
-
-                if not resolved_path.exists():
-                    step_name = step_entry.get("registry_name", step_entry.get("class", "unknown"))
-                    raise FileNotFoundError(
-                        f"Missing processor artifact '{relative_path}' for step '{step_name}' "
-                        f"next to '{config_filename}'. Checkpoint artifacts are incomplete."
-                    )
-                step_entry["config"][config_key] = str(resolved_path)
-
     @classmethod
     def _build_steps_from_config(
         cls,
@@ -1262,7 +1158,6 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
         step_entry: dict[str, Any],
         model_id: str,
         base_path: Path | None,
-        config_filename: str,
         hub_download_kwargs: dict[str, Any],
         is_local_source: bool = False,
     ) -> None:
@@ -1303,8 +1198,6 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
             step_entry: The step configuration dictionary (may contain "state_file")
             model_id: The model identifier (used for Hub downloads if needed)
             base_path: Local directory path for finding state files (None for Hub-only)
-            config_filename: Processor config path, whose parent is used to resolve
-                repository-relative state files on the Hub.
             hub_download_kwargs: Parameters for hf_hub_download (tokens, cache, etc.)
             is_local_source: Whether model_id resolved to a local directory or config file.
 
@@ -1330,7 +1223,7 @@ class DataProcessorPipeline[TInput, TOutput](HubMixin):
             # Download from Hub
             state_path = hf_hub_download(
                 repo_id=model_id,
-                filename=(Path(config_filename).parent / state_filename).as_posix(),
+                filename=state_filename,
                 repo_type="model",
                 **hub_download_kwargs,
             )

src/lerobot/processor/relative_action_processor.py

@@ -91,11 +91,11 @@ class RelativeActionsProcessorStep(ProcessorStep):
     Caches the last seen state so a paired AbsoluteActionsProcessorStep can reverse
     the conversion during postprocessing.
 
-    **Attributes**:
-        - **enabled** (`bool`) -- Whether to apply the relative conversion.
-        - **exclude_joints** (`list[str]`) -- Joint names to keep absolute (not converted to relative).
-        - **action_names** (`list[str] | None`) -- Action dimension names from dataset metadata, used to build
-          the mask from exclude_joints. If None, all dims are converted.
+    Attributes:
+        enabled: Whether to apply the relative conversion.
+        exclude_joints: Joint names to keep absolute (not converted to relative).
+        action_names: Action dimension names from dataset metadata, used to build
+            the mask from exclude_joints. If None, all dims are converted.
     """
 
     enabled: bool = False
@@ -168,10 +168,9 @@ class AbsoluteActionsProcessorStep(ProcessorStep):
     predicted relative offsets are converted back to absolute positions for execution.
     Reads the cached state from its paired RelativeActionsProcessorStep.
 
-    **Attributes**:
-        - **enabled** (`bool`) -- Whether to apply the absolute conversion.
-        - **relative_step** (`RelativeActionsProcessorStep | None`) -- Reference to the paired
-          RelativeActionsProcessorStep that caches state.
+    Attributes:
+        enabled: Whether to apply the absolute conversion.
+        relative_step: Reference to the paired RelativeActionsProcessorStep that caches state.
     """
 
     enabled: bool = False

src/lerobot/processor/rename_processor.py

@@ -32,9 +32,10 @@ class RenameObservationsProcessorStep(ObservationProcessorStep):
     from an environment's format to the format expected by a LeRobot policy or
     other downstream components.
 
-    **Attributes**:
-        - **rename_map** (`dict[str, str]`) -- A dictionary mapping from old key names to new key names. Keys
-          present in an observation that are not in this map will be kept with their original names.
+    Attributes:
+        rename_map: A dictionary mapping from old key names to new key names.
+                    Keys present in an observation that are not in this map will
+                    be kept with their original names.
     """
 
     rename_map: dict[str, str] = field(default_factory=dict)

src/lerobot/processor/render_messages_processor.py

@@ -16,11 +16,9 @@
 
 from __future__ import annotations
 
-from dataclasses import asdict, dataclass
+from dataclasses import dataclass
 from typing import Any
 
-import numpy as np
-
 from lerobot.configs import PipelineFeatureType, PolicyFeature
 from lerobot.configs.recipe import TrainingRecipe
 from lerobot.datasets.language import LANGUAGE_EVENTS, LANGUAGE_PERSISTENT
@@ -34,46 +32,25 @@ from .pipeline import ProcessorStep, ProcessorStepRegistry
 @dataclass
 @ProcessorStepRegistry.register(name="render_messages_processor")
 class RenderMessagesStep(ProcessorStep):
-    """Turn raw language columns into recipe-defined messages and supervision.
+    """Processor step that turns raw language columns into rendered chat messages.
 
-    Reads ``language_persistent`` and ``language_events`` from complementary
-    data, renders them at each sample timestamp, and replaces the raw columns
-    with ``messages``, ``message_streams``, and ``target_message_indices``.
-    Batched inputs are filtered to samples with applicable supervision; samples
-    without language annotations use their task string as low-level supervision
-    when one is available.
+    Reads ``language_persistent`` and ``language_events`` from the transition's
+    complementary data, renders them through ``recipe`` at the sample timestamp,
+    and replaces the raw columns with the resulting ``messages`` /
+    ``message_streams`` / ``target_message_indices`` keys.
     """
 
     recipe: TrainingRecipe
     dataset_ctx: Any | None = None
 
-    def __post_init__(self) -> None:
-        if isinstance(self.recipe, dict):
-            self.recipe = TrainingRecipe.from_dict(self.recipe)
-
-    def get_config(self) -> dict[str, Any]:
-        return {"recipe": asdict(self.recipe)}
-
     def __call__(self, transition: EnvTransition) -> EnvTransition | None:
-        """Render messages, preserving unannotated samples and dropping unmatched annotated ones."""
+        """Render messages for a single transition; return ``None`` to drop it."""
         complementary_data = transition.get(TransitionKey.COMPLEMENTARY_DATA) or {}
         persistent = complementary_data.get(LANGUAGE_PERSISTENT) or []
         events = complementary_data.get(LANGUAGE_EVENTS) or []
 
         if not persistent and not events:
-            # A dataset without language annotations remains usable: render its
-            # task as low-level supervision, or pass it through when no task exists.
-            rendered = _fallback_low_level_render(complementary_data.get("task"))
-            if rendered is None:
-                return transition
-            new_transition = transition.copy()
-            new_complementary_data = dict(new_transition.get(TransitionKey.COMPLEMENTARY_DATA) or {})
-            new_complementary_data.update(rendered)
-            new_transition[TransitionKey.COMPLEMENTARY_DATA] = new_complementary_data
-            return new_transition
-
-        if _is_batched_language(persistent) or _is_batched_language(events):
-            return self._call_batch(transition, complementary_data, persistent, events)
+            return transition
 
         timestamp = complementary_data.get("timestamp")
         if timestamp is None:
@@ -90,171 +67,18 @@ class RenderMessagesStep(ProcessorStep):
             dataset_ctx=self.dataset_ctx,
         )
         if rendered is None:
-            # Language is present but this sparse frame has no applicable recipe
-            # branch. Keep it only when task-level action supervision is possible.
-            rendered = _fallback_low_level_render(complementary_data.get("task"))
-            if rendered is None:
-                return None
+            return None
 
         new_transition = transition.copy()
-        new_complementary_data = dict(new_transition.get(TransitionKey.COMPLEMENTARY_DATA) or {})
+        new_complementary_data = dict(complementary_data)
         new_complementary_data.pop(LANGUAGE_PERSISTENT, None)
         new_complementary_data.pop(LANGUAGE_EVENTS, None)
         new_complementary_data.update(rendered)
         new_transition[TransitionKey.COMPLEMENTARY_DATA] = new_complementary_data
         return new_transition
 
-    def _call_batch(
-        self,
-        transition: EnvTransition,
-        complementary_data: dict[str, Any],
-        persistent_batch: list,
-        events_batch: list,
-    ) -> EnvTransition | None:
-        """Render a language batch.
-
-        Non-empty persistent and event batches must have the same size. Either
-        list may be empty when that language column is absent from the batch.
-        """
-        timestamp = complementary_data.get("timestamp")
-        if timestamp is None:
-            raise KeyError("RenderMessagesStep requires sample timestamp in complementary data.")
-
-        non_empty_batch_sizes = {len(batch) for batch in (persistent_batch, events_batch) if batch}
-        if len(non_empty_batch_sizes) > 1:
-            raise ValueError(
-                "Batched language columns must have equal lengths when both are non-empty, "
-                f"got persistent={len(persistent_batch)} and events={len(events_batch)}."
-            )
-        batch_size = next(iter(non_empty_batch_sizes), 0)
-        messages: list[list[dict[str, Any]]] = []
-        message_streams: list[list[str | None]] = []
-        target_message_indices: list[list[int]] = []
-        keep_indices: list[int] = []
-
-        for i in range(batch_size):
-            rendered = render_sample(
-                recipe=self.recipe,
-                persistent=persistent_batch[i] if i < len(persistent_batch) else [],
-                events=events_batch[i] if i < len(events_batch) else [],
-                t=_batch_value(timestamp, i),
-                sample_idx=int(_batch_value(complementary_data.get("index", 0), i)),
-                task=_batch_value(complementary_data.get("task"), i),
-                dataset_ctx=self.dataset_ctx,
-            )
-            if rendered is None:
-                rendered = _fallback_low_level_render(_batch_value(complementary_data.get("task"), i))
-                if rendered is None:
-                    continue
-            keep_indices.append(i)
-            messages.append(rendered["messages"])
-            message_streams.append(rendered["message_streams"])
-            target_message_indices.append(rendered["target_message_indices"])
-
-        if not messages:
-            return None
-
-        new_transition = (
-            _select_batch_indices(transition, keep_indices, batch_size)
-            if len(keep_indices) != batch_size
-            else transition.copy()
-        )
-        new_complementary_data = dict(new_transition.get(TransitionKey.COMPLEMENTARY_DATA) or {})
-        new_complementary_data.pop(LANGUAGE_PERSISTENT, None)
-        new_complementary_data.pop(LANGUAGE_EVENTS, None)
-        new_complementary_data["messages"] = messages
-        new_complementary_data["message_streams"] = message_streams
-        new_complementary_data["target_message_indices"] = target_message_indices
-        new_transition[TransitionKey.COMPLEMENTARY_DATA] = new_complementary_data
-        return new_transition
-
     def transform_features(
         self, features: dict[PipelineFeatureType, dict[str, PolicyFeature]]
     ) -> dict[PipelineFeatureType, dict[str, PolicyFeature]]:
         """Pass features through unchanged; rendering only touches complementary data."""
         return features
-
-
-def _is_batched_language(value: Any) -> bool:
-    return isinstance(value, list) and bool(value) and isinstance(value[0], list)
-
-
-def _batch_value(value: Any, index: int) -> Any:
-    if value is None:
-        return None
-    if isinstance(value, list):
-        return value[index]
-    if hasattr(value, "ndim") and value.ndim > 0:
-        return unwrap_scalar(value[index])
-    return unwrap_scalar(value)
-
-
-def _select_batch_indices(transition: EnvTransition, indices: list[int], batch_size: int) -> EnvTransition:
-    selected = transition.copy()
-    for key in (TransitionKey.OBSERVATION, TransitionKey.COMPLEMENTARY_DATA):
-        data = selected.get(key)
-        if isinstance(data, dict):
-            selected[key] = {
-                name: _select_value(value, indices, batch_size, f"{key}.{name}")
-                for name, value in data.items()
-            }
-    action = selected.get(TransitionKey.ACTION)
-    if action is not None:
-        selected[TransitionKey.ACTION] = _select_value(action, indices, batch_size, str(TransitionKey.ACTION))
-    return selected
-
-
-def _select_value(value: Any, indices: list[int], batch_size: int, path: str) -> Any:
-    if isinstance(value, dict):
-        return {key: _select_value(item, indices, batch_size, f"{path}.{key}") for key, item in value.items()}
-    if isinstance(value, list):
-        if len(value) != batch_size:
-            raise ValueError(
-                f"Cannot filter batched field {path!r}: expected {batch_size} values, got {len(value)}."
-            )
-        return [value[i] for i in indices]
-    if isinstance(value, np.ndarray) and value.ndim > 0:
-        return value[indices]
-    if hasattr(value, "index_select") and hasattr(value, "new_tensor") and getattr(value, "ndim", 0) > 0:
-        return value.index_select(0, value.new_tensor(indices).long())
-    return value
-
-
-def _fallback_low_level_render(task: Any) -> dict[str, Any] | None:
-    """Keep action-only samples trainable when no recipe branch matches."""
-    if hasattr(task, "item"):
-        task = task.item()
-    if isinstance(task, list):
-        if not task:
-            return None
-        messages = []
-        message_streams = []
-        target_message_indices = []
-        missing_indices = []
-        for index, t in enumerate(task):
-            rendered = _fallback_low_level_render(t)
-            if rendered is None:
-                missing_indices.append(index)
-                continue
-            messages.append(rendered["messages"])
-            message_streams.append(rendered["message_streams"])
-            target_message_indices.append(rendered["target_message_indices"])
-        if missing_indices:
-            if len(missing_indices) == len(task):
-                return None
-            raise ValueError(
-                "Batched low-level fallback requires a non-empty task for every sample; "
-                f"missing task at indices {missing_indices}."
-            )
-        return {
-            "messages": messages,
-            "message_streams": message_streams,
-            "target_message_indices": target_message_indices,
-        }
-    if not isinstance(task, str) or not task:
-        return None
-    return {
-        "messages": [{"role": "user", "content": task}],
-        "message_streams": ["low_level"],
-        "target_message_indices": [],
-    }

src/lerobot/processor/tokenizer_processor.py

@@ -25,7 +25,6 @@ from __future__ import annotations
 
 import logging
 from dataclasses import dataclass, field
-from pathlib import Path
 from typing import TYPE_CHECKING, Any
 
 import torch
@@ -33,7 +32,6 @@ import torch
 from lerobot.configs import FeatureType, PipelineFeatureType, PolicyFeature
 from lerobot.lerobot_types import EnvTransition, RobotObservation, TransitionKey
 from lerobot.utils.constants import (
-    ACTION_CODE_TOKEN_MASK,
     ACTION_TOKEN_MASK,
     ACTION_TOKENS,
     OBS_LANGUAGE_ATTENTION_MASK,
@@ -65,17 +63,15 @@ class TokenizerProcessorStep(ObservationProcessorStep):
 
     Requires the `transformers` library to be installed.
 
-    **Attributes**:
-        - **tokenizer_name** (`str | None`) -- The name of a pretrained tokenizer from the Hugging Face Hub
-          (e.g., "bert-base-uncased").
-        - **tokenizer** (`Any | None`) -- A pre-initialized tokenizer object. If provided, `tokenizer_name` is
-          ignored.
-        - **max_length** (`int`) -- The maximum length to pad or truncate sequences to.
-        - **task_key** (`str`) -- The key in `complementary_data` where the task string is stored.
-        - **padding_side** (`str`) -- The side to pad on ('left' or 'right').
-        - **padding** (`str`) -- The padding strategy ('max_length', 'longest', etc.).
-        - **truncation** (`bool`) -- Whether to truncate sequences longer than `max_length`.
-        - **input_tokenizer** (`Any`) -- The internal tokenizer instance, loaded during initialization.
+    Attributes:
+        tokenizer_name: The name of a pretrained tokenizer from the Hugging Face Hub (e.g., "bert-base-uncased").
+        tokenizer: A pre-initialized tokenizer object. If provided, `tokenizer_name` is ignored.
+        max_length: The maximum length to pad or truncate sequences to.
+        task_key: The key in `complementary_data` where the task string is stored.
+        padding_side: The side to pad on ('left' or 'right').
+        padding: The padding strategy ('max_length', 'longest', etc.).
+        truncation: Whether to truncate sequences longer than `max_length`.
+        input_tokenizer: The internal tokenizer instance, loaded during initialization.
     """
 
     tokenizer_name: str | None = None
@@ -140,7 +136,7 @@ class TokenizerProcessorStep(ObservationProcessorStep):
         # Standardize to a list of strings for the tokenizer
         if isinstance(task, str):
             return [task]
-        elif isinstance(task, list | tuple) and all(isinstance(t, str) for t in task):
+        elif isinstance(task, (list, tuple)) and all(isinstance(t, str) for t in task):
             return list(task)
 
         return None
@@ -297,15 +293,6 @@ class TokenizerProcessorStep(ObservationProcessorStep):
 
         return config
 
-    def save_artifacts(self, save_directory: Path) -> dict[str, str]:
-        """Save the tokenizer so object-provided instances reload without overrides."""
-        artifact_path = Path("tokenizer")
-        save_pretrained = getattr(self.input_tokenizer, "save_pretrained", None)
-        if save_pretrained is None:
-            raise TypeError("Tokenizer must implement save_pretrained() to save a portable pipeline.")
-        save_pretrained(save_directory / artifact_path)
-        return {"tokenizer_name": artifact_path.as_posix()}
-
     def transform_features(
         self, features: dict[PipelineFeatureType, dict[str, PolicyFeature]]
     ) -> dict[PipelineFeatureType, dict[str, PolicyFeature]]:
@@ -348,17 +335,12 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
 
     Requires the `transformers` library to be installed.
 
-    **Attributes**:
-        - **tokenizer_name** -- The name of a pretrained processor from the Hugging Face Hub (e.g.,
-          "lerobot/fast-action-tokenizer").
-        - **tokenizer** -- A pre-initialized processor/tokenizer object. If provided, `tokenizer_name` is
-          ignored.
-        - **trust_remote_code** (`bool`) -- Whether to trust remote code when loading the tokenizer (required
-          for some tokenizers).
-        - **action_tokenizer** (`Any`) -- The internal tokenizer/processor instance, loaded during
-          initialization.
-        - **paligemma_tokenizer_name** (`str`) -- The name of a pretrained PaliGemma tokenizer from the
-          Hugging Face Hub (e.g., "google/paligemma-3b-pt-224").
+    Attributes:
+        tokenizer_name: The name of a pretrained processor from the Hugging Face Hub (e.g., "lerobot/fast-action-tokenizer").
+        tokenizer: A pre-initialized processor/tokenizer object. If provided, `tokenizer_name` is ignored.
+        trust_remote_code: Whether to trust remote code when loading the tokenizer (required for some tokenizers).
+        action_tokenizer: The internal tokenizer/processor instance, loaded during initialization.
+        paligemma_tokenizer_name: The name of a pretrained PaliGemma tokenizer from the Hugging Face Hub (e.g., "google/paligemma-3b-pt-224").
     """
 
     action_tokenizer_name: str | None = None
@@ -367,7 +349,6 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
     max_action_tokens: int = 256
     fast_skip_tokens: int = 128
     paligemma_tokenizer_name: str = "google/paligemma-3b-pt-224"
-    allow_truncation: bool = True
     # Internal tokenizer instance (not part of the config)
     action_tokenizer: Any = field(default=None, init=False, repr=False)
     _paligemma_tokenizer: Any = field(default=None, init=False, repr=False)
@@ -431,15 +412,14 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
             # During inference, no action is available, skip tokenization
             return new_transition
 
-        # Tokenize and get masks for the full formatted sequence and the discrete action codes.
-        tokens, mask, code_mask = self._tokenize_action(action)
+        # Tokenize and get both tokens and mask
+        tokens, mask = self._tokenize_action(action)
 
         # Store mask in complementary data
         complementary_data = new_transition.get(TransitionKey.COMPLEMENTARY_DATA, {})
         if complementary_data is None:
             complementary_data = {}
         complementary_data[ACTION_TOKEN_MASK] = mask
-        complementary_data[ACTION_CODE_TOKEN_MASK] = code_mask
         complementary_data[ACTION_TOKENS] = tokens
         new_transition[TransitionKey.COMPLEMENTARY_DATA] = complementary_data
         return new_transition
@@ -450,7 +430,7 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
         """
         return self._paligemma_tokenizer.vocab_size - 1 - self.fast_skip_tokens - tokens
 
-    def _tokenize_action(self, action: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    def _tokenize_action(self, action: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Tokenizes the action tensor and creates a mask.
 
@@ -479,7 +459,6 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
         # The fast tokenizer expects action data and returns token IDs
         tokens_list = []
         masks_list = []
-        code_masks_list = []
 
         for i in range(batch_size):
             # Tokenize single action (move to CPU first as tokenizer uses scipy which requires numpy)
@@ -497,82 +476,65 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
             if tokens.dim() > 1:
                 tokens = tokens.flatten()
 
-            action_code_tokens = self._act_tokens_to_paligemma_tokens(tokens)
             bos_id = self._paligemma_tokenizer.bos_token_id
-            prompt_tokens = torch.tensor(
-                self._paligemma_tokenizer.encode("Action: ", add_special_tokens=False),
-                device=action.device,
-            )
-            end_tokens = torch.tensor(self._paligemma_tokenizer.encode("|"), device=action.device)
-
-            code_start = 1 + len(prompt_tokens)
-            code_end = code_start + len(action_code_tokens)
+            # add bos
             tokens = torch.cat(
                 [
                     torch.tensor([bos_id], device=action.device),
-                    prompt_tokens,
-                    action_code_tokens,
-                    end_tokens,
+                    torch.tensor(
+                        self._paligemma_tokenizer.encode("Action: ", add_special_tokens=False),
+                        device=action.device,
+                    ),
+                    self._act_tokens_to_paligemma_tokens(tokens),
+                    torch.tensor(self._paligemma_tokenizer.encode("|"), device=action.device),
                 ]
             )
-            code_mask = torch.zeros(len(tokens), dtype=torch.bool, device=action.device)
-            code_mask[code_start:code_end] = True
 
             # Truncate or pad to max_action_tokens
             if len(tokens) > self.max_action_tokens:
-                if not self.allow_truncation:
-                    raise ValueError(
-                        f"FAST action sequence has {len(tokens)} tokens, exceeding "
-                        f"max_action_tokens={self.max_action_tokens}."
-                    )
                 logging.warning(
                     f"Token length ({len(tokens)}) exceeds max length ({self.max_action_tokens}), truncating. "
                     "Consider increasing the `max_action_tokens` in your model config if this happens frequently."
                 )
                 tokens = tokens[: self.max_action_tokens]
-                code_mask = code_mask[: self.max_action_tokens]
                 mask = torch.ones(self.max_action_tokens, dtype=torch.bool, device=action.device)
             else:
-                pad_len = self.max_action_tokens - len(tokens)
                 mask = torch.cat(
                     [
                         torch.ones(len(tokens), dtype=torch.bool, device=action.device),
-                        torch.zeros(pad_len, dtype=torch.bool, device=action.device),
+                        torch.zeros(
+                            self.max_action_tokens - len(tokens), dtype=torch.bool, device=action.device
+                        ),
                     ]
                 )
-                code_mask = torch.nn.functional.pad(code_mask, (0, pad_len), value=False)
                 # Pad tokens with zeros
-                tokens = torch.nn.functional.pad(tokens, (0, pad_len), value=0)
+                tokens = torch.nn.functional.pad(tokens, (0, self.max_action_tokens - len(tokens)), value=0)
 
             tokens_list.append(tokens)
             masks_list.append(mask)
-            code_masks_list.append(code_mask)
 
         # Stack into batched tensors
         tokens_batch = torch.stack(tokens_list, dim=0)  # (B, max_action_tokens)
         masks_batch = torch.stack(masks_list, dim=0)  # (B, max_action_tokens)
-        code_masks_batch = torch.stack(code_masks_list, dim=0)  # (B, max_action_tokens)
 
         # Remove batch dimension if input was single sample
         if single_sample:
             tokens_batch = tokens_batch.squeeze(0)
             masks_batch = masks_batch.squeeze(0)
-            code_masks_batch = code_masks_batch.squeeze(0)
 
         # Move to the same device as the input
         if device is not None:
             tokens_batch = tokens_batch.to(device)
             masks_batch = masks_batch.to(device)
-            code_masks_batch = code_masks_batch.to(device)
 
-        return tokens_batch, masks_batch, code_masks_batch
+        return tokens_batch, masks_batch
 
     def action(self, action: torch.Tensor) -> torch.Tensor:
         """
         This method is not used since we override __call__.
         Required by ActionProcessorStep ABC.
         """
-        tokens, _, _ = self._tokenize_action(action)
+        tokens, _ = self._tokenize_action(action)
         return tokens
 
     def get_config(self) -> dict[str, Any]:
@@ -588,9 +550,6 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
         config = {
             "trust_remote_code": self.trust_remote_code,
             "max_action_tokens": self.max_action_tokens,
-            "fast_skip_tokens": self.fast_skip_tokens,
-            "paligemma_tokenizer_name": self.paligemma_tokenizer_name,
-            "allow_truncation": self.allow_truncation,
         }
 
         # Only save tokenizer_name if it was used to create the tokenizer
@@ -599,14 +558,6 @@ class ActionTokenizerProcessorStep(ActionProcessorStep):
 
         return config
 
-    def save_artifacts(self, save_directory: Path) -> dict[str, str]:
-        artifact_path = Path("action_tokenizer")
-        save_pretrained = getattr(self.action_tokenizer, "save_pretrained", None)
-        if save_pretrained is None:
-            raise TypeError("Action tokenizer must implement save_pretrained() to save a portable pipeline.")
-        save_pretrained(save_directory / artifact_path)
-        return {"action_tokenizer_name": artifact_path.as_posix()}
-
     def transform_features(
         self, features: dict[PipelineFeatureType, dict[str, PolicyFeature]]
     ) -> dict[PipelineFeatureType, dict[str, PolicyFeature]]:

src/lerobot/rewards/pretrained.py

@@ -16,11 +16,12 @@ import abc
 import builtins
 import logging
 import os
-import warnings
+from importlib.resources import files
 from pathlib import Path
+from tempfile import TemporaryDirectory
 from typing import TYPE_CHECKING, Any, TypeVar
 
-from huggingface_hub import hf_hub_download
+from huggingface_hub import HfApi, ModelCard, ModelCardData, hf_hub_download
 from huggingface_hub.constants import SAFETENSORS_SINGLE_FILE
 from huggingface_hub.errors import HfHubHTTPError
 from safetensors.torch import load_model as load_model_as_safetensor, save_model as save_model_as_safetensor
@@ -60,22 +61,6 @@ class PreTrainedRewardModel(nn.Module, HubMixin, abc.ABC):
             raise TypeError(f"Class {cls.__name__} must define 'name'")
 
     def _save_pretrained(self, save_directory: Path) -> None:
-        """Serialize this reward model's parameters (and config) into `save_directory`.
-
-        Safe to call on every rank: replicas carry identical weights, so only the main process
-        writes (sharded reward models are rejected at config validation — no collective gather).
-
-        Args:
-            save_directory (Path): Target directory for the reward model config (`config.json`)
-                and `model.safetensors`.
-        """
-        from lerobot.distributed.utils import is_main_process
-
-        # save_checkpoint calls this on every rank; replicas carry identical
-        # weights, so the main process is the only writer. Sharded reward models are rejected
-        # at config validation, so no collective gather is needed here.
-        if not is_main_process():
-            return
         self.config._save_pretrained(save_directory)
         model_to_save = self.module if hasattr(self, "module") else self
         save_model_as_safetensor(model_to_save, str(save_directory / SAFETENSORS_SINGLE_FILE))
@@ -190,22 +175,53 @@ class PreTrainedRewardModel(nn.Module, HubMixin, abc.ABC):
         """
         return type(self).forward is not PreTrainedRewardModel.forward
 
-    def push_model_to_hub(self, cfg: "TrainPipelineConfig") -> None:
-        """Publish this reward model to the Hub.
+    def push_model_to_hub(self, cfg: "TrainPipelineConfig"):
+        api = HfApi()
+        repo_id = api.create_repo(
+            repo_id=self.config.repo_id, private=self.config.private, exist_ok=True
+        ).repo_id
 
-        Deprecated: use :func:`lerobot.common.train_utils.publish_trained_model` instead.
+        # Push the files to the repo in a single commit
+        with TemporaryDirectory(ignore_cleanup_errors=True) as tmp:
+            saved_path = Path(tmp) / repo_id
 
-        Args:
-            cfg (TrainPipelineConfig): The training config; saved as `train_config.json` and
-                used to render the model card.
-        """
-        from lerobot.common.train_utils import publish_trained_model
+            self.save_pretrained(saved_path)  # Calls _save_pretrained and stores model tensors
 
-        warnings.warn(
-            "PreTrainedRewardModel.push_model_to_hub is deprecated and will be removed in a "
-            "future version. Use lerobot.common.train_utils.publish_trained_model(cfg, model, "
-            "preprocessor, postprocessor, dataset_meta) instead.",
-            FutureWarning,
-            stacklevel=2,
+            card = self.generate_model_card(
+                cfg.dataset.repo_id, self.config.type, self.config.license, self.config.tags
+            )
+            card.save(str(saved_path / "README.md"))
+
+            cfg.save_pretrained(saved_path)  # Calls _save_pretrained and stores train config
+
+            commit_info = api.upload_folder(
+                repo_id=repo_id,
+                repo_type="model",
+                folder_path=saved_path,
+                commit_message="Upload reward model weights, train config and readme",
+                allow_patterns=["*.safetensors", "*.json", "*.yaml", "*.md"],
+                ignore_patterns=["*.tmp", "*.log"],
+            )
+
+            logging.info(f"Model pushed to {commit_info.repo_url.url}")
+
+    def generate_model_card(
+        self, dataset_repo_id: str, model_type: str, license: str | None, tags: list[str] | None
+    ) -> ModelCard:
+        card_data = ModelCardData(
+            license=license or "apache-2.0",
+            library_name="lerobot",
+            pipeline_tag="robotics",
+            tags=list(set(tags or []).union({"robotics", "lerobot", "reward-model", model_type})),
+            model_name=model_type,
+            datasets=dataset_repo_id,
         )
-        publish_trained_model(cfg, self, None, None, None)
+
+        template_card = (
+            files("lerobot.templates")
+            .joinpath("lerobot_rewardmodel_modelcard_template.md")
+            .read_text(encoding="utf-8")
+        )
+        card = ModelCard.from_template(card_data, template_str=template_card)
+        card.validate()
+        return card

src/lerobot/rewards/sarm/processor_sarm.py

@@ -16,7 +16,6 @@
 
 from __future__ import annotations
 
-import logging
 import random
 from typing import TYPE_CHECKING, Any
 
@@ -70,8 +69,6 @@ from .sarm_utils import (
     pad_state_to_max_dim,
 )
 
-logger = logging.getLogger(__name__)
-
 
 class SARMEncodingProcessorStep(ProcessorStep):
     """ProcessorStep that encodes images and text with CLIP and generates stage and progress labels for SARM."""
@@ -111,8 +108,6 @@ class SARMEncodingProcessorStep(ProcessorStep):
             else None
         )
 
-        self._validate_annotation_columns()
-
         self.device = torch.device(
             self.config.device if self.config.device else "cuda" if torch.cuda.is_available() else "cpu"
         )
@@ -125,78 +120,6 @@ class SARMEncodingProcessorStep(ProcessorStep):
         self.verbs = ["move", "grasp", "rotate", "push", "pull", "slide", "lift", "place"]
         self.fake = Faker()
 
-    @staticmethod
-    def _resolve_annotation_column(episodes_df: pd.DataFrame, annotation_type: str, suffix: str) -> str:
-        """Resolve a mode-specific annotation column, falling back to the legacy unprefixed name."""
-        prefixed = f"{annotation_type}_{suffix}"
-        return prefixed if prefixed in episodes_df.columns else suffix
-
-    @staticmethod
-    def _annotations_are_usable(names: Any, starts: Any, ends: Any) -> bool:
-        """Return whether an episode has non-empty, aligned annotation arrays."""
-        values = (names, starts, ends)
-        if not all(isinstance(value, (list, tuple, np.ndarray)) for value in values):
-            return False
-
-        lengths = {len(value) for value in values}
-        return len(lengths) == 1 and next(iter(lengths)) > 0
-
-    def _validate_annotation_columns(self) -> None:
-        """Validate annotation coverage before loading models or generating training targets.
-
-        A multi-stage head with no usable episode annotations would otherwise train entirely
-        against all-zero targets. Reject that configuration and warn when only part of the
-        dataset is usable.
-        """
-        if self.dataset_meta is None:
-            return
-        episodes_df = self.dataset_meta.episodes.to_pandas()
-        num_episodes = len(episodes_df)
-
-        modes = []
-        if self.dense_subtask_names and len(self.dense_subtask_names) > 1:
-            modes.append(("dense", self.dense_subtask_names))
-        if self.sparse_subtask_names and len(self.sparse_subtask_names) > 1:
-            modes.append(("sparse", self.sparse_subtask_names))
-
-        for annotation_type, names in modes:
-            columns = [
-                self._resolve_annotation_column(episodes_df, annotation_type, suffix)
-                for suffix in ("subtask_names", "subtask_start_frames", "subtask_end_frames")
-            ]
-            missing_columns = [column for column in columns if column not in episodes_df.columns]
-            if missing_columns:
-                num_usable = 0
-            else:
-                num_usable = sum(
-                    self._annotations_are_usable(*(episodes_df.loc[ep_idx, column] for column in columns))
-                    for ep_idx in episodes_df.index
-                )
-
-            if num_usable == 0:
-                missing_columns_message = (
-                    f" Missing required columns: {', '.join(missing_columns)}." if missing_columns else ""
-                )
-                raise ValueError(
-                    f"SARM {annotation_type} head is configured with {len(names)} stages, but none of "
-                    f"the {num_episodes} episodes have usable annotations in meta/episodes/*.parquet. "
-                    f"Required columns: {', '.join(columns)}.{missing_columns_message} "
-                    "Training would produce all-zero "
-                    "targets. Materialize the annotations into the episodes metadata before training."
-                )
-
-            num_unusable = num_episodes - num_usable
-            if num_unusable:
-                logger.warning(
-                    "SARM %s head: %d/%d episodes have unusable annotations in columns %s; "
-                    "their targets will be 0 and only the %d annotated episodes will train the head.",
-                    annotation_type,
-                    num_unusable,
-                    num_episodes,
-                    ", ".join(columns),
-                    num_usable,
-                )
-
     def _find_episode_for_frame(self, frame_idx: int) -> int:
         """Find the episode index for a given frame index."""
         for ep_idx in range(len(self.dataset_meta.episodes)):
@@ -244,18 +167,24 @@ class SARMEncodingProcessorStep(ProcessorStep):
         if episodes_df is None or len(global_names) == 1:
             return None, None, None
 
-        columns = [
-            self._resolve_annotation_column(episodes_df, annotation_type, suffix)
-            for suffix in ("subtask_names", "subtask_start_frames", "subtask_end_frames")
-        ]
-        if any(column not in episodes_df.columns for column in columns) or ep_idx >= len(episodes_df):
+        # Resolve column name with fallback
+        def col(suffix):
+            prefixed = f"{annotation_type}_{suffix}"
+            return prefixed if prefixed in episodes_df.columns else suffix
+
+        col_names = col("subtask_names")
+        if col_names not in episodes_df.columns or ep_idx >= len(episodes_df):
             return None, None, None
 
-        annotations = tuple(episodes_df.loc[ep_idx, column] for column in columns)
-        if not self._annotations_are_usable(*annotations):
+        subtask_names = episodes_df.loc[ep_idx, col_names]
+        if subtask_names is None or (isinstance(subtask_names, float) and pd.isna(subtask_names)):
             return None, None, None
 
-        return annotations
+        return (
+            subtask_names,
+            episodes_df.loc[ep_idx, col("subtask_start_frames")],
+            episodes_df.loc[ep_idx, col("subtask_end_frames")],
+        )
 
     def __call__(self, transition: EnvTransition) -> EnvTransition:
         """

src/lerobot/rewards/topreward/modeling_topreward.py

@@ -58,11 +58,12 @@ import builtins
 import logging
 import os
 from pathlib import Path
+from tempfile import TemporaryDirectory
 from typing import TYPE_CHECKING, Any, TypeVar
 
 import numpy as np
 import torch
-from huggingface_hub import hf_hub_download
+from huggingface_hub import HfApi, hf_hub_download
 from huggingface_hub.constants import CONFIG_NAME
 from huggingface_hub.errors import HfHubHTTPError
 from torch import Tensor
@@ -74,6 +75,9 @@ from lerobot.rewards.topreward.configuration_topreward import TOPRewardConfig
 from lerobot.rewards.topreward.processor_topreward import TOPREWARD_FEATURE_PREFIX, TOPREWARD_INPUT_KEYS
 from lerobot.utils.import_utils import _transformers_available, require_package
 
+if TYPE_CHECKING:
+    from lerobot.configs.train import TrainPipelineConfig
+
 if TYPE_CHECKING or _transformers_available:
     from transformers import Qwen3VLForConditionalGeneration
 else:
@@ -201,3 +205,34 @@ class TOPRewardModel(PreTrainedRewardModel):
         instance.to(config.device)
         instance.eval()
         return instance
+
+    def push_model_to_hub(self, cfg: TrainPipelineConfig):
+        """Push the TOPReward ``config.json`` + model card to the Hub."""
+        api = HfApi()
+        repo_id = api.create_repo(
+            repo_id=self.config.repo_id, private=self.config.private, exist_ok=True
+        ).repo_id
+
+        with TemporaryDirectory(ignore_cleanup_errors=True) as tmp:
+            saved_path = Path(tmp) / repo_id
+            saved_path.mkdir(parents=True, exist_ok=True)
+
+            self.config._save_pretrained(saved_path)
+
+            card = self.generate_model_card(
+                cfg.dataset.repo_id, self.config.type, self.config.license, self.config.tags
+            )
+            card.save(str(saved_path / "README.md"))
+
+            cfg.save_pretrained(saved_path)
+
+            commit_info = api.upload_folder(
+                repo_id=repo_id,
+                repo_type="model",
+                folder_path=saved_path,
+                commit_message="Upload TOPReward config and readme",
+                allow_patterns=["*.json", "*.yaml", "*.md"],
+                ignore_patterns=["*.tmp", "*.log", "*.safetensors"],
+            )
+
+            logger.info(f"Model pushed to {commit_info.repo_url.url}")

src/lerobot/rl/joint_observations_processor.py

@@ -38,11 +38,11 @@ class JointVelocityProcessorStep(ObservationProcessorStep):
     difference between the current and the last observed joint positions. The
     resulting velocity vector is then concatenated to the original state vector.
 
-    **Attributes**:
-        - **dt** (`float`) -- The time step (delta time) in seconds between observations, used for calculating
-          velocity.
-        - **last_joint_positions** (`torch.Tensor | None`) -- Stores the joint positions from the previous
-          step to enable velocity calculation.
+    Attributes:
+        dt: The time step (delta time) in seconds between observations, used for
+            calculating velocity.
+        last_joint_positions: Stores the joint positions from the previous step
+                              to enable velocity calculation.
     """
 
     dt: float = 0.1
@@ -138,9 +138,9 @@ class MotorCurrentProcessorStep(ObservationProcessorStep):
     This step queries the robot's hardware interface to get the present current
     for each motor and concatenates this information to the existing state vector.
 
-    **Attributes**:
-        - **robot** (`Robot | None`) -- An instance of a `lerobot` Robot class that provides access to the
-          hardware bus.
+    Attributes:
+        robot: An instance of a `lerobot` Robot class that provides access to
+               the hardware bus.
     """
 
     robot: Robot | None = None