GPU-ACCELERATED CROWD SIMULATION WITH USER-GUIDANCE AND MULTI-LEVEL UNCERTAINTY

A long-standing rule for evaluating the realism and robustness of the crowd simulation is whether the simulations are capable of avoiding the congestion as well as maintaining free agent-wise collision. Although a few methods solved this problem in some speci c case, no overall method has been explored. This paper proposes a new approach to simulate large-scale crowd behavior with real-world properties and without obvious artifacts. Specifi cally, we devised a lightweight user-interaction scheme to combine the navigation field and continuum crowd method. Also, the two-level uncertainty model is imposed onto the above crowd dynamic system to simulate non-goal driven behavior. All of the computations are accelerated by using GPU. Through our experiments and comparisons, we demonstrates the advantages of our approach.