Position-based real-time simulation of large crowds

We introduce a crowd simulation method that runs at interactive rates for on the order of a hundred thousand agents, making it particularly suitable for use in games. Our new method is inspired by Position-Based Dynamics (PBD), a fast physics-based animation technique in widespread use. Individual agents in crowds are abstracted by particles, whose motions are controlled by intuitive position constraints and planning velocities, which can be readily integrated into a standard PBD solver, and agent positions are projected onto constraint manifolds to eliminate colliding configurations. A variety of constraints are presented, enabling complex collective behaviors with robust agent collision avoidance in both sparse and dense crowd scenarios.