Real-time Simulation of Heterogeneous Crowds

We present a real-time algorithm for simulating heterogeneous crowds with potentially distinct individual behavior characteristics and goals. Our approach combines global motion planning with a generalized pedestrian dynamics model to synthesize motion for numerous groups of people in complex dynamic environments. In- spired by self-organization phenomena in crowds, we introduce "Pedestrian Levels of Detail" to accelerate large-scale simulation of individual agents, while preserving natural collective behaviors observed in real crowds. We highlight the performance of our algo- rithm on heterogeneous crowds in urban environments.