Fast Penetration Depth Computation Using Rasterization Hardware and Hierarchical Refinement

We present a novel and fast algorithm to compute penetration depth (PD) between two polyhedral models. Given two overlapping polyhedra, it computes the minimal translation distance to separate them using a combination of object-space and image-space techniques. The algorithm computes pairwise Minkowski sums of decomposed convex pieces, performs closest-point query using rasterization hardware and refines the estimated PD by object-space walking. It uses bounding volume hierarchies, model simplification, object-space and image-space culling algorithms to further accelerate the computation and refines the estimated PD in a hierarchical manner. We highlight its performance on complex models and demonstrate its application to dynamic simulation and tolerance verification.