Skip to content

DVIDIA OpenUSD workstation v0.1

Pre-release
Pre-release

Choose a tag to compare

View all tags
@Tarzelf Tarzelf released this 30 Oct 15:27
· 8 commits to main since this release
usd-workstation-v0.1
a8e24ec

DVIDIA OpenUSD workstation v0.1 establishes a bounded, reproducible environment step toward installable robot skills. A real USD SDK parser maps one original rigid cube and tabletop into pinned native PiPER/MuJoCo mechanics.

  • 45/45 passive physics checks pass: 27 drops, nine slides, nine disabled-effect controls.
  • Independent goals: 27/27 normal drops; 6/9 normal slides.
  • Read-only audit: 45 cases and 7,845 saved frames checked, zero mismatches.
  • All authored inputs, native poses, traces, source snapshots, environment record and license notices are included.

Download the ZIP, extract it, run python -m http.server 8000 in that folder, then open http://127.0.0.1:8000/. This browser view replays saved native physics; it does not simulate or train. See the included docs/usd-workstation-v01.md for native installation and CLI reproduction from this source tag.

ZIP SHA-256: 037a7fb053c317ed4900771321dc3498a83ed8a92ce83f9a648e20b7f29947e6

Physics ran on CPU with no dedicated GPU. The tested environment used Python 3.12.9, NumPy 2.5.3, MuJoCo 3.15.0 and OpenUSD 26.11, with an existing native environment made available through a local .pth file. Clean installation, minimum hardware and hard real-time operation are not qualified.

This does not establish arbitrary USD/SimReady import, calibrated shoe mechanics, learned manipulation, human-video motor learning or physical transfer. Original code and scene are MIT; upstream assets retain their notices.

Research: https://research.dvidia.org/papers/openusd-workstation-v01/
Hugging Face: https://huggingface.co/datasets/Dvidia/usd-workstation-v01

DVIDIA research · @iammrriver · hello@dvidia.org

Assets 5
Loading