A Parallel Block Iterative Method for Interactive Contacting Rigid Multibody Simulations on Multicore PCs

Fast solution methods for complementarity problems (CPs) with several hundred variables are essential for interactive realtime simulation of systems of constrained, contacting rigid bodies. These simulations are commercially relevant for several applications such as virtual environment (VE) heavy machinery operator training systems. We describe a hybrid, asynchronous, block parallel method to approximately solve these CPs in realtime on multicore CPUs. A graph analysis phase identifies components which are weakly coupled using a physically inspired weight function. Each component is then solved in parallel using either a block principal pivot method or a projected block Gauss Seidel method running in separate threads. Couplings which generate forces between the subsystems are handled iteratively using a Gauss-Seidel process which communicates updates between the interacting subsystems asynchronously. Preliminary results show that this approach delivers good performance while keeping overhead small.