Embodied AI Glossary中文

Collision Geometry (Collider)

碰撞体Common

The simplified shape a physics engine uses for collision checks, usually different from what's rendered on screen.

A simulated object usually has two separate geometries: a visual mesh, used for rendering and meant to look good, with a high polygon count; and a collision geometry, handed to the physics engine for collision detection, meant to be computed quickly and stably. Collision geometry is usually a basic shape — sphere, capsule, box — or a convex hull (the smallest convex shape that encloses the object), because these have efficient collision algorithms. MuJoCo automatically converts a non-convex mesh supplied by the user into a convex hull for collision purposes; Isaac Sim also defaults to convex hulls, with convex decomposition (splitting an object into several convex pieces) or an SDF (signed distance field) mesh available when a closer fit is needed. Too coarse a collision shape, and a mug gets treated as solid, with nothing able to grip its handle; too fine, and the simulation slows down. This is exactly the distinction between <visual> and <collision> in a URDF, specified separately for every link.

ExampleIf a coffee mug uses a single convex hull as its collision geometry, the opening gets sealed shut and a spoon can't be placed inside it in simulation; decomposing it into several convex pieces with a tool like CoACD lets the wall, base, and handle collide separately.

Also called
collision mesh, collision shape, Collider
Related
Convex Decomposition · Convex Decomposition · Collision Detection · Unified Robot Description Format · Signed Distance Field / Function · Interpenetration
Sources
Isaac Sim Documentation: Simulation Fundamentals(Collision approximations)
MuJoCo Documentation: Computation - Collision detection

See it in the full glossary →