Speeding Up Cloth Simulation

Simulating the motion of cloth is an important component in virtual character animation. Believable animations are now expected in feature films. Games and virtual reality are next, but the computational costs are still high. In this thesis we present a number of methods which reduce these costs, without losing accuracy or generality. To this end, we introduce a novel adaptive implicit-explicit time integration scheme, which takes advantage of simulation parameters — locally in both space and time — to improve the efficiency of the computation. Building upon this technique, we present a decomposition method which attempts to decouple the cloth mesh into multiple components that can be solved separately and in parallel. These techniques are introduced in the context of particle-system models, and include discussions on a variety of modelling and simulation issues. We also significantly improve the efficiency of the modified preconditioned conjugate gradient technique often used in cloth simulation for implicit integration schemes. We present improvements in the form of a preconditioner for the constrained problem and a better initial guess.