Real-Time Physics Simulation Forum

Just out of curiousity. What methods can be used to compute the (deep) penetration of two convex polyhedra. Are there any alternatives to EPA and Erwin's "Sampling" method that he uses in Bullet?

Seperating axes works well for convex shapes in 2D, I'm not sure what potential issues there are with extending it to 3D.

This to be precise, Bullet uses EPA by default, it give an exact penetration depth (within a given tolerance). Sampling method is an approximation.

Apart from SAT, which is also exact, other methods to approximate penetration depth include Dual-space Expansion

Benefit of EPA over SAT is that it support non-polyhedral convex objects like implicit cylinders & cones, convex sum of objects, minkowski sum etc.

Hope this helps,
Erwin