[Feature] Introduce mujoco_playground environment wrapper by itwasabhi · Pull Request #3751 · pytorch/rl

itwasabhi · 2026-05-14T12:34:42Z

Introduces a torch-rl wrapper of the mujoco playground environment. Wrapper also supports multi-agent decomposition.

Motivation and Context

Types of changes

What types of changes does your code introduce? Remove all that do not apply:

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds core functionality)
Breaking change (fix or feature that would cause existing functionality to change)
Documentation (update in the documentation)
Example (update in the folder of examples)

Checklist

Go over all the following points, and put an x in all the boxes that apply.
If you are unsure about any of these, don't hesitate to ask. We are here to help!

I have read the CONTRIBUTION guide (required)
My change requires a change to the documentation.
I have updated the tests accordingly (required for a bug fix or a new feature).
I have updated the documentation accordingly.

pytorch-bot · 2026-05-14T12:34:49Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/3751

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 12 New Failures, 1 Cancelled Job, 2 Pending, 1 Unrelated Failure

As of commit b3c41ee with merge base 8f7575d ():

NEW FAILURES - The following jobs have failed:

Build M1 Wheels / pytorch/rl / build-wheel-py3_10-cpu (gh)
Process completed with exit code 1.
Build M1 Wheels / pytorch/rl / upload / upload-wheel-py3_10-cpu (gh)
Unable to download artifact(s): Artifact not found for name: pytorch_rl__3.10_cpu_
Generate documentation / build-docs (3.12, 12.8) / linux-job (gh)
RuntimeError: Command docker exec -t bee270aec9227b3e5f53a37ca4de10e2e42a020d07c5ce1d0719a5919604993f /exec failed with exit code 2
Libs Tests on Linux / unittests-chess (3.10, 12.8) / linux-job (gh)
RuntimeError: Command docker exec -t 21ac64eeea55842f630220664e7adfc7337f950ad7755e8c5183b0a7c7f3b898 /exec failed with exit code 1
Libs Tests on Linux / unittests-jumanji (3.10, 12.8) / linux-job (gh)
RuntimeError: Command docker exec -t 1692b883c4d28110e999c24762885d191c888c4c9990e59a29002f2157373d03 /exec failed with exit code 1
Tutorials Tests on Linux / tests (3.12, 13.0) / linux-job (gh)
RuntimeError: Command docker exec -t f987b3b6200e133855c2fe69838f4c303d9849e0cd6301cb531bbba19f43ff4e /exec failed with exit code 1
Unit-tests on Linux / tests-cpu (3.10) / linux-job (gh)
test/transforms/test_env_transforms.py::TestFrameSkipTransform::test_frame_skip_metaclass_leaves_batched_envs_untouched
Unit-tests on Linux / tests-cpu (3.11) / linux-job (gh)
test/transforms/test_env_transforms.py::TestFrameSkipTransform::test_frame_skip_metaclass_leaves_batched_envs_untouched
Unit-tests on Linux / tests-cpu (3.12) / linux-job (gh)
test/transforms/test_env_transforms.py::TestFrameSkipTransform::test_frame_skip_metaclass_leaves_batched_envs_untouched
Unit-tests on Linux / tests-cpu (3.13) / linux-job (gh)
test/transforms/test_env_transforms.py::TestFrameSkipTransform::test_frame_skip_metaclass_leaves_batched_envs_untouched
Unit-tests on Linux / tests-olddeps (3.10, 11.8) / linux-job (gh)
test/transforms/test_env_transforms.py::TestFrameSkipTransform::test_frame_skip_metaclass_leaves_batched_envs_untouched
Unit-tests on Linux / tests-optdeps (3.12, 13.0) / linux-job (gh)
test/transforms/test_env_transforms.py::TestFrameSkipTransform::test_frame_skip_metaclass_leaves_batched_envs_untouched

CANCELLED JOB - The following job was cancelled. Please retry:

Unit-tests on Linux / tests-cpu (3.14) / linux-job (gh)
##[error]The operation was canceled.

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

Unit-tests on Windows / unittests-cpu (3.10, windows.4xlarge, cpu) / windows-job (gh) (trunk failure)
##[error]The operation was canceled.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

github-actions · 2026-05-14T12:34:54Z

⚠️ PR Title Label Error

PR title must start with a label prefix in brackets (e.g., [BugFix]).

Current title: Introduce mujoco_playground environment wrapper.

Supported Prefixes (case-sensitive)

Your PR title must start with exactly one of these prefixes:

Prefix	Label Applied	Example
`[BugFix]`	BugFix	`[BugFix] Fix memory leak in collector`
`[Feature]`	Feature	`[Feature] Add new optimizer`
`[Doc]` or `[Docs]`	Documentation	`[Doc] Update installation guide`
`[Refactor]`	Refactoring	`[Refactor] Clean up module imports`
`[CI]`	CI	`[CI] Fix workflow permissions`
`[Test]` or `[Tests]`	Tests	`[Tests] Add unit tests for buffer`
`[Environment]` or `[Environments]`	Environments	`[Environments] Add Gymnasium support`
`[Data]`	Data	`[Data] Fix replay buffer sampling`
`[Performance]` or `[Perf]`	Performance	`[Performance] Optimize tensor ops`
`[BC-Breaking]`	bc breaking	`[BC-Breaking] Remove deprecated API`
`[Deprecation]`	Deprecation	`[Deprecation] Mark old function`
`[Quality]`	Quality	`[Quality] Fix typos and add codespell`

Note: Common variations like singular/plural are supported (e.g., [Doc] or [Docs]).

itwasabhi · 2026-05-19T14:46:20Z

Thanks for the fixes @vmoens

itwasabhi · 2026-05-21T21:45:03Z

failing mujoco_playground tests should now pass.

itwasabhi · 2026-05-26T08:05:40Z

@vmoens any other requested changes? I believe the remaining test failures are unrelated.

Introduces a torch-rl wrapper of the mujoco playground environment. Wrapper also supports multi-agent decomposition.

- CI: install `playground` (PyPI name) instead of import name `mujoco_playground`; the previous `pip install mujoco_playground` was the root cause of the current `unittests-mujoco-playground` CI failure. - CI: drop the broken try/except JAX-init fallback in run_test.sh — it ran in a subprocess that exited immediately, so `JAX_PLATFORM_NAME` never reached pytest and any real "GPU not visible" error was hidden. - Wrapper: docstrings no longer advertise a `state` field that the env never emits; the JAX state is intentionally kept on `self._current_state` rather than round-tripped through TensorDict (this is now documented as a `.. note::` block). - Wrapper: freeze `MujocoPlaygroundAgentSpec` / `MujocoPlaygroundAgentMapping` dataclasses and deep-copy `KNOWN_MARL_MAPPINGS` entries on string lookup so users cannot mutate the module-level mapping by accident. - Wrapper: emit a `UserWarning` when resolving a string against `KNOWN_MARL_MAPPINGS`, since those indices target Brax's observation layout and may not be semantically equivalent for mujoco_playground envs. - Wrapper: document the policy contract for `homogenization_mode='max'` and `'concat'` (which action/obs entries are real vs padding/discarded). - Wrapper: align `_MujocoPlaygroundMeta` num_workers handling with `_BraxMeta`; make `agent_mapping` and `config`/`config_overrides` keyword-only; accept `seed=None` in `_set_seed` (defaults to 0, matching the `_reset` fallback) instead of raising bare `Exception`; document `_listerize`'s inclusive-range semantics; drop unused `pixels_only`/`camera_id`/`render_kwargs` from `_build_env`. - Example: rewrite `save_visualization` in `profile_mujoco_playground_collector.py` to snapshot `env._current_state` during a manual rollout, removing dependencies on non-existent `env._state_example` and `td["state"]`. - Config: drop stale `categorical_action_encoding` field on `MujocoPlaygroundEnvConfig`; add `agent_mapping` and `num_workers` so the MARL and parallel-process knobs are reachable from the config system. - Tests: replace three duplicated `_setup_jax` fixtures with a single module-level autouse fixture; introduce a session-scoped `marl_env_sizes` fixture to avoid constructing a throwaway env in every MARL test; expect the new MABrax warning on string lookups; add a new `TestMujocoPlaygroundDictObs` class covering reset / `check_env_specs` and a negative test that `homogenization_mode != 'none'` raises `NotImplementedError` for dict-obs envs; make `test_no_mapping_regression` actually exercise the MARL env it constructs. - Docs: keep the `MOGymEnv` / `MOGymWrapper` pair adjacent in `envs_libraries.rst`; `__repr__` now uses `env_name=` to match the constructor kwarg.

itwasabhi · 2026-06-04T11:26:51Z

@vmoens OK with these changes?

…e_skip Several env wrappers (Brax, Jumanji, MuJoCo Playground, ...) accept a ``frame_skip`` argument but silently ignore it because they don't loop over it in their own ``_step``. Capture this once in the ``EnvBase`` metaclass: when ``frame_skip > 1`` and the env does not implement it natively, wrap the env in a ``FrameSkipTransform``. A new ``EnvBase._has_frame_skip`` class attribute (default ``False``) gates the behaviour. Envs that implement frame-skipping in their own ``_step`` opt out by setting it to ``True``: - ``GymLikeEnv`` (loops over ``wrapper_frame_skip``) -> True, which also covers Gym, DMControl, RoboHive, Habitat, IsaacGym, SafetyGymnasium. - ``GenesisWrapper`` (loops over ``_frame_skip``) -> True. - ``JumanjiWrapper`` overrides ``_step`` without the loop, so it resets the inherited flag back to False and now gets the auto-transform. The default construction path is unchanged (the block only triggers for ``frame_skip > 1``), so there is no impact on existing envs. Building the ``TransformedEnv`` does not recurse: ``TransformedEnv`` is created via ``_TEnvPostInit`` which bypasses ``_EnvPostInit.__call__``. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- Cache the registry env listing (`available_envs`) once per process instead of re-enumerating the suites on every classproperty access. - Precompute per-agent obs/action dims and index tensors in `_make_marl_specs` so the hot-path `_split_obs_for_agents` / `_reconstruct_global_action` no longer recompute `max_obs`/`raw_sizes` on every step. Behaviour is unchanged (identical shapes/values). - Drop the redundant `wrapper_frame_skip = 1` line: `frame_skip > 1` is now handled by the auto-appended `FrameSkipTransform`. - Document the two limitations surfaced in review: partial resets are not supported for vmapped batches (the JAX state lives on the instance), and `terminated` is aliased to `done` (no separate truncation signal). - Document the public `MujocoPlaygroundAgentMapping` / `MujocoPlaygroundAgentSpec` dataclasses in the reference docs. - Add a `frame_skip` test class and the missing `if __name__ == "__main__"` pytest entrypoint. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…attr The auto-FrameSkipTransform block in `_EnvPostInit.__call__` read `getattr(instance, "frame_skip", 1)` on every env construction. On a batched env (`ParallelEnv`/`SerialEnv`) `frame_skip` is not a real attribute, so this fell through to `BatchedEnv.__getattr__`, which for a lazy (non-started) env calls `self.start()` and raises `RuntimeError` ("Trying to access attributes of closed/non started environments"). `getattr`'s default only swallows `AttributeError`, so the RuntimeError propagated and broke construction of every lazy batched env (observed across brax, jumanji, mujoco_playground num_workers tests and any plain `ParallelEnv(...)` in the suite), in addition to spuriously starting the workers. Read `frame_skip` straight from `instance.__dict__` instead: it is set there by `_EnvWrapper.__init__`, and batched/native envs that don't set it simply default to 1 and are skipped without ever touching `__getattr__`. Adds a regression test constructing lazy `SerialEnv` / `ParallelEnv` of `CountingEnv`. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label May 14, 2026

github-actions Bot added Documentation Improvements or additions to documentation CI Has to do with CI setup (e.g. wheels & builds, tests...) Environments Adds or modifies an environment wrapper Examples Trainers labels May 14, 2026

itwasabhi changed the title ~~Introduce mujoco_playground environment wrapper.~~ [Feature] Introduce mujoco_playground environment wrapper. May 14, 2026

github-actions Bot added the Feature New feature label May 14, 2026

vmoens force-pushed the mujoco_playground0 branch from c7a92d5 to 15de4dd Compare May 19, 2026 10:21

itwasabhi force-pushed the mujoco_playground0 branch from 15de4dd to 916baf0 Compare May 21, 2026 21:26

itwasabhi and others added 4 commits June 4, 2026 10:07

Introduce mujoco_playground environment wrapper.

a4ccad2

Introduces a torch-rl wrapper of the mujoco playground environment. Wrapper also supports multi-agent decomposition.

Minor cleanup.

f445a5a

fix failing tests

cc6c461

itwasabhi force-pushed the mujoco_playground0 branch from 916baf0 to cc6c461 Compare June 4, 2026 09:10

vmoens changed the title ~~[Feature] Introduce mujoco_playground environment wrapper.~~ [Feature] Introduce mujoco_playground environment wrapper Jun 8, 2026

vmoens and others added 2 commits June 8, 2026 10:27

github-actions Bot added Environments/jumanji Triggers jumanji environment tests only Transforms labels Jun 8, 2026

vmoens and others added 3 commits June 8, 2026 10:29

Merge remote-tracking branch 'origin/main' into pr3751-fixes

2ed4ed3

linter

b3c41ee

github-actions Bot added the tutorials/ label Jun 8, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Feature] Introduce mujoco_playground environment wrapper#3751

[Feature] Introduce mujoco_playground environment wrapper#3751
itwasabhi wants to merge 9 commits into
pytorch:mainfrom
itwasabhi:mujoco_playground0

itwasabhi commented May 14, 2026

Uh oh!

pytorch-bot Bot commented May 14, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented May 14, 2026

Uh oh!

itwasabhi commented May 19, 2026

Uh oh!

itwasabhi commented May 21, 2026

Uh oh!

itwasabhi commented May 26, 2026

Uh oh!

itwasabhi commented Jun 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

itwasabhi commented May 14, 2026

Motivation and Context

Types of changes

Checklist

Uh oh!

pytorch-bot Bot commented May 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/3751

❌ 12 New Failures, 1 Cancelled Job, 2 Pending, 1 Unrelated Failure

Uh oh!

github-actions Bot commented May 14, 2026

⚠️ PR Title Label Error

Supported Prefixes (case-sensitive)

Uh oh!

itwasabhi commented May 19, 2026

Uh oh!

itwasabhi commented May 21, 2026

Uh oh!

itwasabhi commented May 26, 2026

Uh oh!

itwasabhi commented Jun 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

pytorch-bot Bot commented May 14, 2026 •

edited

Loading