Conflict Resolution for Multi-Agent Hybrid Systems

A conflict resolution architecture for multi-agent hybrid systems is presented. The agents are considered as players in a noncooperative, zero-sum dynamic game where the actions of other agents are modeled as disturbances. If the saddle solution to the game results in no collision for the worst possible disturbance then coordination among agents is not necessary. In case the saddle solution is not safe, then predefined, coordinated maneuvers are used to avoid possible collisions. The architecture is illustrated with a simple example from air traffic systems. Keywords: Hybrid systems, game theory, collision avoidance. 1 Introduction One of the features of multi-agent systems is that agents compete for usage of a common resource. Examples are highway systems [1], in which many cars drive on the same highway, and air traffic management systems [2], in which the aircraft must land on the same runway. In such systems, the optimal policies between different agents may be conflicting. Furthe..