Skip to content

feat: add visual domain randomization - #6

Open
Hugo-Zhai wants to merge 4 commits into
intelligent-control-lab:mainfrom
Hugo-Zhai:feat/domain-randomization-minimal
Open

feat: add visual domain randomization#6
Hugo-Zhai wants to merge 4 commits into
intelligent-control-lab:mainfrom
Hugo-Zhai:feat/domain-randomization-minimal

Conversation

@Hugo-Zhai

Copy link
Copy Markdown

Summary

  • add current-stage visual randomization for dome lighting, table and background materials, and scene backgrounds
  • add interactive toggles, deterministic assignment tracking, per-episode resampling support, and full stage restoration
  • add starter and expanded asset profiles with locked sources and runtime extras
  • build portable runtime, starter-assets, and expanded-assets wheels with headless launcher support and native RUNPATH cleanup

User impact

Users can enable one or more visual randomization effects from Python or the interactive UI. Asset profiles remain optional: starter provides the smaller catalog, while expanded provides the complete catalog and remains compatible with starter configurations.

Validation

  • focused Ruff and Isaac-aware Ty checks passed for the added runtime and integration files
  • a real GPU smoke test passed environment reset, two task reconstructions, four simultaneous effects, stable 1280x800 Head Camera frames, per-episode resampling, and disabled-resampling retention
  • starter and expanded lifecycle, scene-safety, restore, material-isolation, and four-button UI behavior were validated on the release-equivalent implementation
  • git diff --check passes

This PR contains production code and packaging metadata only. It does not include local test harnesses, binary asset payloads, local paths, or private artifacts.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant