CI Configuration Review

This is a clean infrastructure change that removes the temporary SHA256 digest pin from the Isaac Sim Docker image tag.

Summary

The change reverts from a pinned image (latest-develop@sha256:0dd49a...) to the floating latest-develop tag. According to the PR description, this is appropriate because the upstream Isaac Sim issue that necessitated the pin has been resolved.

Analysis

What this changes:

• All CI workflows that consume .github/workflows/config.yaml (build, docs, publish-images, etc.) will now pull whatever image latest-develop resolves to at execution time
• Different CI runs may use different underlying images if NVIDIA pushes updates between runs

Trade-offs (informational, not blocking):

• ✅ Benefit: CI stays current with Isaac Sim nightly improvements and fixes automatically
• ⚠️ Trade-off: If a broken Isaac Sim nightly is pushed, CI could fail for reasons outside this repo's control. This is a known and accepted trade-off for projects tracking upstream development branches.

Verdict

Ship it. This is a straightforward, correct change with no code-level concerns. The decision to use floating tags vs pinned digests is a policy choice, and the team has clearly opted for staying current.

No action items required.

Update (186c08f):

⚠️ Scope change: This PR has grown significantly from a 1-line CI config change to a large feature/refactoring commit. The new changes include:

• URDF/MJCF converter refactoring: Delegation to Isaac Sim's importer with new config options
• Camera/sensor warp math helpers: New warp_math.py with orientation conversion and depth processing kernels
• Import path standardization: from isaaclab.utils.configclass import configclass across ~80 files
• New observation terms: stacked_image for frame-stacking camera observations
• Newton renderer updates: ProxyArray integration for set_outputs/update_camera
• Extensive new test coverage: MJCF, stacked image observations, etc.

Previous inline comment status:

• ✅ The original CI config change remains valid; my suggestion about re-pinning is informational, not blocking.

New changes assessment:

• The import path changes are mechanical and correct.
• The converter refactoring delegates to upstream Isaac Sim APIs — cleaner architecture.
• The warp math kernels are well-documented with clear derivations.
• New tests exercise the added functionality.

No new blocking issues identified in the incremental diff.

Update (c06d99c):

This commit focuses on CI refactoring and OVRTX renderer memory optimization:

CI Changes

• Removed extra-pip-packages: "ovrtx" from the 3 generic isaaclab_tasks test shards — OVRTX is no longer needed there
• Added 2 new dedicated CI jobs:
  • test-rendering-correctness: Runs in-kit rendering tests (test_rendering_*.py, non-kitless)
  • test-rendering-correctness-kitless: Runs kitless OVRTX tests with extra-pip-packages: "ovrtx"
• Both new jobs have continue-on-error: true — appropriate for experimental/non-blocking tests
• Updated tools/test_settings.py with RENDERING_CORRECTNESS_TESTS and RENDERING_CORRECTNESS_KITLESS_TESTS lists to exclude these from generic shards

OVRTX Renderer Memory Optimization

• Eliminated disk I/O: Stage export now uses ExportToString() → open_usd_from_string() instead of writing temp files
• Removed temp_usd_suffix config option (breaking change, documented in changelog)
• Changed temp_usd_dir default from temp dir to None — disk write only when explicitly configured for debugging
• Backward compatibility: OVRTX < 0.3.0 still writes to disk (falls back to temp dir)

Assessment

✅ Clean, well-structured changes
✅ Proper changelog entries added
✅ Memory optimization is a clear improvement for production use
✅ CI job separation makes rendering tests more maintainable

No blocking issues.

Update (0a9b9af):

This is a large commit covering several areas:

1. Install CI Test Suite Expansion

• Added test_isaaclabx_i_mimic.py, test_isaaclabx_i_none.py, test_isaaclabx_i_rl.py, test_isaaclabx_i_visualizer.py, test_isaaclabx_integrated.py for comprehensive install scenario testing
• Added Conda_Mixin in utils.py for conda environment testing support
• Updated existing install tests to use new -i none (core only) and feature-selector patterns

2. Ray Caster Backend Refactoring (PhysX + Newton)

• New PhysX backend: isaaclab_physx/sensors/ray_caster/ with direct RigidBodyView pose tracking
• New Newton backend: isaaclab_newton/sensors/ray_caster/ with site-based pose tracking via NewtonManager.cl_register_site
• Fixed sensor pose tracking regression when parented under rigid bodies
• Mesh transform updates now write directly to Warp arrays (bypassing torch staging)

3. Test Cleanup

• Removed AppLauncher boilerplate from ~20 unit test files that don't require Isaac Sim runtime
• Test files for utils, configclass, math, noise, etc. now run standalone with just Warp

4. Docker CI Improvements (tools/run_install_ci.py)

• Added --conda flag for layered conda image builds
• Improved JUnit XML result extraction via docker cp instead of bind mounts
• Better container lifecycle management with named containers

5. Package Updates

• rsl-rl-lib upgraded from 3.1.2 to 5.0.1 in wheel builder config
• robomimic moved from opt-in extra to required dependency for isaaclab_mimic on Linux
• Version bumps: isaaclab_newton 0.11.0, isaaclab_physx 0.9.0, isaaclab_ov 0.2.1, isaaclab_ovphysx 2.0.0, isaaclab_tasks 1.8.0

6. Task Config Additions

• Added raycaster-camera depth presets (raycaster_depth64/128/256) for Dexsuite Kuka-Allegro manipulation task

Assessment

✅ Ray caster refactoring addresses the pose tracking regression correctly
✅ Test cleanup is appropriate — these tests don't need Isaac Sim
✅ Install CI expansion provides better coverage for the modular install system
✅ Docker CI improvements are sensible for CI reliability
✅ Changelog entries present for all version bumps

No new blocking issues.