Fast, Interactive Soft-Body Animation A Real-Time Physics Engine using Position-Based Dynamics

Abstract

Physics simulations for realistic computer graphics is a subject that has been around for a long time. This paper covers the implementation of a framework called Position-Based Dynamics with the goal to simulate real-time soft-body dynamics in an interactive setting. The implemented simulator can simulate a number of different models, including cloth, in a number of different scenes. The paper also compares two techniques for solving linear systems, Jacobi and Gauss-Seidel, as well as looks at the advantages and problems with using parallelization with several CPU cores. For a CPU implementation it is found that the octree performs better than the uniform grid. A parallel Gauss-Seidel which ignores order of computations is evaluated to show that the incurred overhead error is outweighed by the time gained from parallelization, at least for a low amount of parallel threads.