Wockenfuss, FalcoLürig, ChristophJan Bender and Kenny Erleben and Eric Galin2013-10-312013-10-312011978-3-905673-87-6https://doi.org/10.2312/PE/vriphys/vriphys11/101-110The simulation of larger crowds of peoples at interactive frames rates get more and more important in games and interactive simulations. Crowds contribute significantly to an immersive urban environment. The nature of the problem where one has to simulate a lot of individuals makes it very accessible for parallelization strategies. As graphics hardware today is a very accessible and powerful parallel computation hardware, the problem lends itself to be adapted to graphics hardware. It is implemented in CUDA in our case. CUDA and other parallel implementations of social force models usually have several numerical and parallelization issues that are finally based on too inhomogeneous population distributions. These are often also undesirable from a game developers point of view. In this paper we introduce the concept of congestion avoidance into the crowd simulator that solves those technical issues and makes the whole simulation also look more natural for gaming applications.Categories and Subject Descriptors (according to ACM CCS): I.6.8 [Simulation and Modeling]: Parallel Gaming