Environment builder / v0.2
A reusable workstation for physical skills. Place the objects, run native physics locally, and inspect whether the whole object is inside the bin.
Rigid objects and an open bin made from five separate colliders. The shoe is an original three-box surrogate, with unmeasured physical properties. This is an environment and verification experiment; a learned robot placement is not demonstrated.
The recipe describes dimensions, poses, mass and friction alongside a pinned arm profile. Native simulation checks the resulting contacts and motion. Drafts with invalid geometry are rejected before a run.
These checks help build reusable training environments. They do not establish real-object calibration, deformable shoe mechanics, learned grasping or physical robot transfer.
Use the same recipe from the editor or the Terminal. Once the Python environment and pinned robot assets are installed, local simulation runs on the CPU.
dvidia-environment studio \ --assets ./robot-assets --output ./local-scenes --port 8346
The browser only draws geometry and saved poses. “Validate & simulate” is available when served by the local Python application.