Visual and collision geometry
A detailed mesh and a useful collision representation serve different jobs.
Research note · November 14, 2026Research question
How can a change to collision geometry alter results without changing the rendered scene?
Method
- Keep the visible mesh fixed and compare two documented collision representations. Record simplification steps and the reason each representation was chosen.
- Run a contact scenario near a feature that differs between the representations. Compare contact location, motion and force traces, including missed contacts and unexpected intersections.
- Expose the collision representation during inspection. A reviewer should be able to see which geometry the solver uses instead of inferring it from a high-resolution render.
Records to retain
- Visual mesh revision
- Collision geometry revision
- Simplification procedure
- Contact and trajectory records
Source basis
glTF describes renderable assets, while MuJoCo documents collision geometry and contact computation. A rendering format and a physics representation need not contain the same information or use the same geometric approximation.
Interpretation & limits
More visual polygons do not automatically improve a physical model. This protocol does not establish the accuracy of a particular mesh or collision simplifier.
Next experiments
- Use one fixture with a narrow gap or concave feature.
- Compare a simplified representation with a declared reference.
References
Technical reference
Implementation details supporting this note, drawn directly from the maintained documentation. These are technical references, not additional papers or new experimental results. Expand a topic to inspect its method, fields and limits.