GPU-based Collision Detection and Response for Particles on 3D Models

This paper presents a GPU-based approach to handle collision detection and response for particles on 3D models. Signed Hausdorff distance fields were used and stored in volume textures. The voxels in a distance field stored the Hausdorff distance to the 3D model and the normal of the corresponding nearest surface. Object-based collision detection and response were processed using a pixel shader with texture sampling on the Hausdorff distance field. Particles were updated by rendering them using programmable graphics hardware acceleration. No data had to be transferred to the CPU between frames. The power of parallel GPU computation made the simulation of the particle system on a GPU at least twice as fast as that on a CPU.