Real-Time Crowd Simulation for Emergency Planning

We present a system for simulating crowd behavior in emergency scenarios, specifically, the evacuation of buildings. We use a combination of techniques from robot motion planning, collision detection and response, particle systems, and group behavior. Our method initially builds a roadmap of the environment using PRMs, and computes the path to closest exits on the roadmap. The virtual agents must follow these paths to evacuate the scene, under a set of physical constraints and behavior rules. We have modelled the agents as a particle system, moving in a potential field while following the PRM paths.