Start with the behavior you need to reproduce
A box used as a fixed background obstacle has different requirements from one the robot must push, lift or stack. Before tuning values, define the action and observable outcome: sliding distance, lifting response, settling behavior or another task-specific measure.
Treat a supplied physical value as data to inspect. Ask whether it came from a measurement, manufacturer reference, approximation or simulator default. Those are different levels of evidence.
Mass is more than a single number
The UsdPhysics mass API describes mass, density, center of mass and inertia properties. Check units and how the target simulator resolves missing or combined values. A plausible total mass does not establish that the mass distribution matches the intended object.
For an asset that tips or rotates during the task, include orientation changes in your test. Document the reference object and what was measured. Avoid assigning an apparently precise number when its source is only a visual guess.
Friction and restitution need contact tests
UsdPhysics materials include static friction, dynamic friction and restitution. A visual material such as metal or rubber does not automatically determine those parameters. Contact behavior also depends on the other surface and runtime settings.
Useful small tests include sliding an object on the task surface, pushing it from rest and dropping it from a known height. Record both surfaces, the initial state and simulation configuration. Compare results with the behavior needed for your task before expanding to a full scene.
Keep a reproducible record
| Record | Why it matters |
|---|---|
| Asset and scene revision | Identifies the exact content tested |
| Parameter source | Separates measurements from assumptions |
| Simulator and physics settings | Makes the runtime conditions explicit |
| Initial state and action | Enables a repeatable test |
| Acceptance range | Defines what counts as useful behavior |
Our environment quality checklist applies this approach to a complete scene. Inspect individual packages in the library; do not assume every item includes every physical property or a measurement report.