Collision avoidance in an agent system for crowd simulation

Aiming at precisely modelling people's ability to avoid collisions in crowded circumstances, as well as reducing the computation load for collision avoidance in large-scale crowd simulation, a collision avoidance method based on a set of rules and strategies for an agent system is proposed. Inspired by the pedestrian movement logic of “perception-decision-action”, a group of collision avoidance rules is designed according to the analysis of obstacle perception and collision detection. When an agent encounters an obstacle, it evaluates its situation and takes action to avoid the collision by looking through the collision avoidance rule tables. Obstacle perception is analyzed by using geometric visual model, and collision detection by using the angle between the observer and obstacle. Simulation results of several scenarios show this rule-based method avoids collisions effectively, and simulation is very similar to the actual crowd behaviors.