fix(rsl_rl): default obs_groups for new runners

[saime428]

Summary:

• Populate a default obs_groups mapping for OnPolicyRunner configs when running with rsl-rl >= 4.0.0.
• Preserve any explicitly configured obs_groups values.
• Add regression coverage for missing and existing obs_groups in the rsl-rl deprecation helper.

Why:

• rsl-rl 4.x expects obs_groups in runner configs. Existing play.py already routes configs through handle_deprecated_rsl_rl_cfg() before calling OnPolicyRunner, so filling the missing default there keeps play/eval paths compatible without changing each script separately.

Validation:

• python -m compileall source\isaaclab_rl\isaaclab_rl\rsl_rl\utils.py
• python -m compileall source\isaaclab_rl\test\test_rsl_rl_cfg_deprecation.py
• python -m ruff check source\isaaclab_rl\isaaclab_rl\rsl_rl\utils.py source\isaaclab_rl\test\test_rsl_rl_cfg_deprecation.py
• lightweight handle_deprecated_rsl_rl_cfg() dummy-runner check for default and preserved obs_groups
• python -m pytest source\isaaclab_rl\test\test_rsl_rl_cfg_deprecation.py -q (collection is blocked in this local environment by missing Isaac Sim / Warp packages)

Notes for reviewers:

• Closes [Bug Report] play.py hangs in OnPolicyRunner init with rsl-rl 4.0.0+ due to missing obs_groups #5363.
• This does not change public APIs or training logic; it only fills a compatibility default when the runner config omits obs_groups.

[greptile-apps]

Greptile Summary

This PR fixes compatibility with rsl-rl >= 4.0.0 by automatically populating a default obs_groups mapping on OnPolicyRunner configs when the field is left at its MISSING sentinel. The logic is placed inside handle_deprecated_rsl_rl_cfg, which is already called on every play/eval path, so existing scripts get the fix without modification.

• _set_default_obs_groups is added to utils.py and called at the top of the >= 4.0.0 branch; it only mutates configs whose class_name is \"OnPolicyRunner\" and whose obs_groups is still MISSING, preserving any explicitly set value.
• Two parametrized tests in TestObsGroups verify the default-fill and the no-overwrite behavior across versions 4.x and 5.x.

Confidence Score: 4/5

Safe to merge — the change is a small, well-guarded default-fill that only mutates configs missing obs_groups, and explicitly preserves any value the user already set.

The production logic in utils.py is straightforward and correctly gated by three independent guards (class_name, hasattr, and _is_missing). The only gaps are in the test file: there is no assertion that the rsl-rl < 4.0.0 path leaves obs_groups untouched, and there is a minor PEP 8 formatting issue. Neither affects correctness of the shipped code.

The test file test_rsl_rl_cfg_deprecation.py would benefit from a small additional boundary test for the < 4.0.0 path; utils.py itself needs no further attention.

Important Files Changed

Filename	Overview
source/isaaclab_rl/isaaclab_rl/rsl_rl/utils.py	Adds _set_default_obs_groups helper and calls it at the start of the rsl-rl >= 4.0.0 branch; logic is guarded correctly by class_name, hasattr, and _is_missing checks.
source/isaaclab_rl/test/test_rsl_rl_cfg_deprecation.py	Adds TestObsGroups with two parametrized tests covering default-fill and preserve cases; missing a blank line before the following section separator, and no negative test confirming obs_groups is not set for rsl-rl < 4.0.0.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["handle_deprecated_rsl_rl_cfg(agent_cfg, version)"] --> B{version < 4.0.0?}
    B -- Yes --> C[Validate policy present\nClear new model cfgs]
    B -- No --> D["_set_default_obs_groups(agent_cfg)"]
    D --> E{class_name == 'OnPolicyRunner'\nAND hasattr obs_groups\nAND obs_groups is MISSING?}
    E -- Yes --> F["Set obs_groups = {'actor': ['policy'], 'critic': ['policy']}"]
    E -- No --> G[Leave obs_groups unchanged]
    F --> H{policy present?}
    G --> H
    H -- Yes --> I[Infer actor/critic/student/teacher from deprecated policy]
    H -- No --> J{version < 5.0.0?}
    I --> J
    J -- Yes --> K[Validate old stochastic cfg]
    J -- No --> L[Migrate to distribution_cfg]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A["handle_deprecated_rsl_rl_cfg(agent_cfg, version)"] --> B{version < 4.0.0?}
    B -- Yes --> C[Validate policy present\nClear new model cfgs]
    B -- No --> D["_set_default_obs_groups(agent_cfg)"]
    D --> E{class_name == 'OnPolicyRunner'\nAND hasattr obs_groups\nAND obs_groups is MISSING?}
    E -- Yes --> F["Set obs_groups = {'actor': ['policy'], 'critic': ['policy']}"]
    E -- No --> G[Leave obs_groups unchanged]
    F --> H{policy present?}
    G --> H
    H -- Yes --> I[Infer actor/critic/student/teacher from deprecated policy]
    H -- No --> J{version < 5.0.0?}
    I --> J
    J -- Yes --> K[Validate old stochastic cfg]
    J -- No --> L[Migrate to distribution_cfg]

Loading

_{Reviews (1): Last reviewed commit: "fix(rsl_rl): default obs_groups for new ..." | Re-trigger Greptile}

[saime428]

Hi, checking in on this fix for #5363. The PR remains conflict-free, though the branch is now behind main. I can update it before review if you prefer, or make any requested changes.

[saime428]

Following up on this PR. The branch is currently behind main but remains mergeable. I can update it if preferred; otherwise I will avoid an unnecessary rebase. Please let me know if any changes are needed.

[saime428]

Retargeted this PR from main to develop and rebased onto current develop. The fix and test commits are unchanged; I also added the changelog fragment that develop requires (source/isaaclab_rl/changelog.d/fix-rsl-rl-play-obs-groups.rst, patch tier) — that mechanism does not exist on main, so the PR was missing it.

I originally opened this against main, which is likely why it never went anywhere: in a month the only check that ever ran was labeler. docs/source/refs/contributing.rst (L53 on develop) still says "Submit a pull request to the main branch", which contradicts the PR template's "Target develop for all contributions" — that is what I followed. I've opened #6940 to fix that one line so the next contributor doesn't hit it.

The fix still applies on develop: both insertion points are unchanged there and obs_groups is still never defaulted, so #5363 still reproduces.

Verified locally against develop (source/isaaclab_rl/test/test_rsl_rl_cfg_deprecation.py):

• with the fix — 42 passed
• reverting only the utils.py hunk while keeping the tests — 2 failed, 40 passed, both being TestObsGroups::test_defaults_missing_obs_groups_for_on_policy_runner[4.0.0/5.0.0]

So the new tests do fail without the fix rather than passing vacuously. Happy to make any changes you'd like.