Collision detection and response approaches for computer muscle modelling

Abstract

Computer muscle modelling is used for many pur-poses, from injury recovery and treatment of chronic diseases to disease prediction. These predictions often involve computing the muscle's internal forces to determine further how fast something may happen (e.g. how quickly the muscle joint wears out). During the simulation of such a model, collisions of soft and rigid bodies inevitably occur. This paper tests various state-of-the-art collision handling methods: voxelisation, one using Signed Distance Fields and one based on Bounding Volume Hierarchies. These methods are tested in the context of muscle modelling with the previously proposed position-based dynamics approach. Compared to the other options, using the Discregrid library for Signed Distance Field generation shows the best results, mainly due to its accuracy to the speed of execution ratio. In contrast to the current system, visually pleasant improvements are significant.