A New Fast and Robust Collision Detection and Force Computation Algorithm Applied to the Physics Engine Bullet: Method, Integration, and Evaluation

Abstract

We present a collision detection and force computation algorithm based on the Voxelmap-Pointshell Algorithm which was integrated and evaluated in the physics engine Bullet. Our algorithm uses signed distance fields and point-sphere trees and it is able to compute collision forces between arbitrary complex shapes at simulation frequencies smaller than 1 msec. Utilizing sphere hierarchies, we are able to rapidly detect likely colliding areas, while the point trees can be used for processing colliding regions in a level-of-detail manner. The integration into the physics engine Bullet was performed inheriting interface classes provided in that framework. We compared our algorithm with Bullet's native GJK, GJK with convex decomposition, and GImpact, varying the resolution and the scenarios. Our experiments show that our integrated algorithm performs with similar computation times as the standard collision detection algorithms in Bullet if low resolutions are chosen. With high resolutions, our algorithm outperforms Bullet's native implementations and objects behave realistically.