OpenUSD workstation · v0.1

A scene with testable physics.

One reusable recipe. A pinned PiPER arm. Object properties parsed from USD and checked in native simulation.

Opening the benchmark…
RECORDED NATIVE PHYSICS · ARM HELD
Opening verified native geometry…
0.00 s
———

USD scene ↓Scene recipe ↓Native trace ↓All 45 results ↓Protocol ↓

What this proves

The bridge reads a deliberately small USD scene: a rigid object, a tabletop and explicit physical properties. The native simulator keeps the pinned robot mechanics. This experiment checks falling, resting contact and sliding across mass, friction and height variations. Removing gravity, support or friction supplies negative controls.

The browser replays saved native body poses. It does not run physics, train a policy or connect to a robot. Properties remain assumptions until measured on real objects. No shoe grasp, video-to-action learning or physical transfer is qualified by these checks.

Run it locally

python -m dvidia_training.usd_scene validate examples/usd-workstation-v01/recipe.json
python -m dvidia_training.usd_scene_benchmark prepare --example examples/usd-workstation-v01/recipe.json --output runs/my-usd-check
python -m dvidia_training.usd_scene_benchmark run --output runs/my-usd-check --assets ./robot-assets