Cherry-picks fixes for visualizers, SKRL template, digit limitations, teleop/mimic dependency

[kellyguo11]

Description

Cherry pick bug fix PRs from develop:

• Document Digit closed-loop articulation limitation on Newton #5821
• Adds isaaclab_teleop as dependency of isaaclab_mimic during mimic only install #5820
• Update Visualizer Integration Tests + Some Visualizer Fixes #5733
• Merges changes from main #5824

[isaaclab-review-bot]

Code Review Summary

This cherry-pick PR brings several bug fixes and improvements from develop to the release/3.0.0-beta2 branch. The changes are well-structured and appropriately scoped for a release stabilization branch.

📋 Changes Overview

Category	Files	Assessment
Dependency Fix	install.py	✅ Correct
Documentation	environments.rst, issues.rst, visualization.rst	✅ Accurate
Visualizer Tests	Split into 4 test files + utils	✅ Improved isolation
SKRL Templates	3 agent config templates	✅ Updated for skrl 2.0.0
Camera/Visualizer	camera_view.py, newton_visualizer.py, physx_manager.py	✅ Sound

---

🔬 Detailed Analysis

1. Mimic/Teleop Dependency (install.py)

OPTIONAL_ISAACLAB_SUBMODULES: dict[str, tuple[str, ...]] = {
    "mimic": ("isaaclab_teleop", "isaaclab_mimic"),  # teleop now installed first
    ...
}

Verdict: ✅ Correct fix. Since isaaclab_mimic depends on isaaclab_teleop, installing teleop first prevents import errors during mimic data generation.

2. Camera View Position Resolution (camera_view.py)

The reordering of fallback methods in prim_world_positions() is more robust:

1. First tries FrameView (PhysX/Fabric-backed transforms)
2. Falls back to scene articulation positions
3. Final fallback uses USD xform cache

Verdict: ✅ Better error isolation - FrameView exceptions now fail fast before attempting scene fallback.

3. PhysX Rigid Body View Wildcard Patterns (physx_manager.py)

patterns.add(re.sub(r"/World/envs/env_\d+", "/World/envs/env_*", prim.GetPath().pathString))

Verdict: ✅ Restores cross-environment compatibility for Newton visualizers updating PhysX transforms.

4. Newton Headless Detection (newton_visualizer.py)

runtime_headless = self.cfg.headless or (
    sys.platform not in ("win32", "darwin") and not os.environ.get("DISPLAY")
)

Verdict: ✅ Fixes false headless detection on Windows/macOS where DISPLAY is not used.

5. SKRL Template Updates

Updates config field names for skrl 2.0.0 compatibility:

• state_preprocessor → observation_preprocessor (for observations)
• amp_state_preprocessor → amp_observation_preprocessor
• task_reward_weight → task_reward_scale
• style_reward_weight → style_reward_scale
• Removed deprecated discriminator_reward_scale

Verdict: ✅ Aligns with skrl 2.0.0 API.

6. Documentation Updates

• Removes newton_mjwarp from Digit environment support tables
• Adds Known Issues section documenting closed-loop articulation limitation
• Improves tiled camera view documentation with UI access instructions

Verdict: ✅ Accurate and helpful for users.

7. Visualizer Test Restructuring

Splits monolithic 609-line test file into:

• 4 separate test files (Newton/PhysX × standard/tiled)
• 1571-line shared utilities module

Verdict: ✅ Better test isolation, reduces flakiness, easier debugging.

---

🧪 Test Coverage

Change Type	Coverage Status
Mimic dependency	✅ test_install_command_parsing.py updated
Visualizer fixes	✅ Integration tests restructured
Documentation	N/A (no tests needed)
SKRL templates	N/A (config templates)

---

📝 Minor Observations

1. visualizer_cfg.py field reordering: The docstring update for tiled_cam_target_prim_path defaulting to "/World/envs/*/Robot" is clear and matches typical robot asset layouts.

2. string.py docstring fix: The corrected preserve_order documentation now accurately describes the behavior (was previously swapped).

---

✅ Verdict: Approve

This is a clean cherry-pick PR with well-scoped bug fixes appropriate for a release branch. All changes are:

• Self-contained with minimal cross-module impact
• Properly tested where applicable
• Well-documented

No blocking issues identified.

[greptile-apps]

Greptile Summary

This PR cherry-picks four bug-fix branches from develop into release/3.0.0-beta2, targeting visualizer correctness, SKRL template API alignment, Agility Digit newton_mjwarp limitations, and the isaaclab_mimic → isaaclab_teleop install dependency.

• Visualizer fixes: prim_world_positions now tries FrameView first for live PhysX/Fabric transforms before falling back to scene articulation and USD; PhysX rigid-body view switches from per-env exact paths to wildcard patterns (env_*) so a single view covers all envs; Newton headless detection is corrected to skip the DISPLAY check on Windows/macOS.
• SKRL templates: state_preprocessor renamed to observation_preprocessor across AMP, IPPO, and MAPPO configs; MAPPO critic input corrected to STATES; AMP reward weight fields renamed to *_scale and unified.
• Install / CI: ./isaaclab.sh -i mimic now co-installs isaaclab_teleop; the nightly changelog workflow is extended to compile against a configurable list of branches using a GitHub App token.

Confidence Score: 4/5

Safe to merge; all changes are targeted bug fixes with no behavioral regressions identified in the main simulation paths.

The core logic changes (FrameView-first fallback chain, wildcard rigid-body view, platform-aware headless detection) are well-scoped and covered by the restructured integration tests. The pyarrow==22.0.0 hard pin and the silent exception swallow in camera_view.py are minor quality concerns that do not affect correctness under normal operation.

source/isaaclab/isaaclab/envs/utils/camera_view.py (exception swallowing) and source/isaaclab_visualizers/setup.py (pyarrow hard pin) warrant a second look.

Important Files Changed

Filename	Overview
source/isaaclab/isaaclab/envs/utils/camera_view.py	Refactored prim_world_positions: FrameView is now tried first for all envs atomically; if any env fails, the entire batch falls back to scene articulation then USD XformCache. Silent exception swallow loses diagnostic information.
source/isaaclab_physx/isaaclab_physx/physics/physx_manager.py	Rigid-body view now deduplicates per-env paths into wildcard patterns (/World/envs/env_*) using a set, so a single PhysX view covers all envs; fixes Newton visualizer live-transform updates.
source/isaaclab_visualizers/isaaclab_visualizers/newton/newton_visualizer.py	Headless detection now skips DISPLAY check on non-Linux platforms (Windows/macOS), and the tiled camera UI control is renamed from Newton ImageLogger to Tiled Camera View.
source/isaaclab/isaaclab/cli/commands/install.py	mimic install now includes isaaclab_teleop as a prerequisite since isaaclab_mimic declares it as a required dependency.
source/isaaclab/isaaclab/visualizers/visualizer_cfg.py	Field reordering only; tiled_cam_target_prim_path default changed from /World/envs//Robot/base to /World/envs//Robot (matching the articulation root, not a child link).
tools/template/templates/agents/skrl_amp_cfg	Updated to SKRL renamed API: state_preprocessor to observation_preprocessor, amp_state_preprocessor to amp_observation_preprocessor, task/style reward keys renamed to *_scale, discriminator_reward_scale removed.
tools/template/templates/agents/skrl_mappo_cfg	Critic model input corrected to STATES (centralized critic), shared_state_preprocessor removed, observation_preprocessor added alongside state_preprocessor per updated SKRL MAPPO API.
source/isaaclab/isaaclab/utils/string.py	Docstring corrections only: preserve_order=True and preserve_order=False descriptions were swapped in both resolve_matching_names and resolve_matching_names_values; now correctly documented.
source/isaaclab_visualizers/setup.py	Added pyarrow==22.0.0 as a hard-pinned dependency in the rerun extra to fix a runtime compatibility issue.
.github/workflows/nightly-changelog.yml	Expanded from single-branch (develop) to matrix over a configurable CRON_BRANCHES list; migrated auth from PAT to GitHub App token; concurrency scoped per-branch; manual dispatch requires exactly one branch.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["prim_world_positions(stage, path, env_indices, scene)"] --> B["Try FrameView for ALL envs"]
    B -->|"Success for all"| C["Return tensor"]
    B -->|"Any exception"| D["positions.clear()"]
    D --> E{scene is not None?}
    E -->|"Yes"| F["_scene_articulation_positions()"]
    F -->|"Returns tensor"| G["Return tensor"]
    F -->|"Returns None"| H["USD XformCache fallback per env"]
    E -->|"No"| H
    H --> I["Return tensor"]

Loading

Comments Outside Diff (2)

1. source/isaaclab_visualizers/setup.py, line 754 (link)

   Hard-pinned transitive dependency may cause conflicts

   pyarrow==22.0.0 is pinned with == in the rerun extra. Any consumer that also pulls in pyarrow (e.g., from a data-science stack) will hit an unsatisfiable constraint if their requirement is not exactly 22.0.0. The same pin appears in tools/wheel_builder/res/python_packages.toml. If the pin is needed to work around a specific rerun-sdk/pyarrow incompatibility, a >=22.0.0,<23 floor-style bound would be more permissive while still excluding the broken version.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

2. source/isaaclab_visualizers/isaaclab_visualizers/newton/newton_visualizer.py, line 711-727 (link)

   Accessing private _images and _selected on Newton's ImageLogger

   _draw_tiled_camera_view_controls reads image_logger._images and writes image_logger._selected directly. These are private attributes of Newton's upstream ImageLogger; if that class renames or restructures them in a future Newton release, this method will raise AttributeError at runtime with no clear failure message. It would be safer to access them through whatever public API Newton exposes, or at minimum document the Newton version these attributes were verified against.

_{Reviews (1): Last reviewed commit: "Merges changes from main (#5824)" | Re-trigger Greptile}

[greptile-apps]

The bare except Exception that clears positions and falls through to the next strategy swallows the original error completely. If FrameView fails for a reason other than "this prim type isn't supported" (e.g., a wrong import, a GPU OOM, a partial iteration where 8 of 10 envs succeeded), there is no log entry to explain why the slower fallback path was taken. Adding a logging.debug here would make production triage much easier.

Suggested change

	except Exception:
	positions.clear()
	if scene is not None:
	except Exception as _e:
	import logging as _logging
	_logging.getLogger(__name__).debug(
	"FrameView path failed for prim_path_template=%r; falling back to scene/USD path. Reason: %s",
	prim_path_template,
	_e,
	)
	positions.clear()
	if scene is not None: