Collision model uses a voxel-based convex-hull decomposition that is relatively fast but may not be highly accurate, especially for holes, doorways, and cavities in general.

Collision model uses the convex hull of the visual mesh.

Collision model uses a bounding box that encompasses the visual mesh.

Collison model uses a convex-hull decomposition of the visual mesh which is computed by a variation of an algorithm called HACD. This option is the most accurate representation of collision geometry for complex objects with built-in tolerances to reduce the total number of convex-hulls by compromising on accuracy when justified.