Mas-Based Pursuit-Evasion Algorithm Under Unknown Environment

Abstract

Based on the interaction protocols of Contract Net and Subscribe-publish, a distributive pursuit-evasion algorithm is proposed for multi-agents with limited detection ranges to pursue multi-evaders in an unknown grid map with obstacles. The pursuit agents first explore the map according to the method of Artificial Potential Field. When any obstacle is detected within the neighborhood of a pursuer, it updates the map. If an evader is detected, the pursuer (initiator) sends call-for-proposals messages to other pursuers. The other pursuers use D* algorithm to calculate their steps to reach the evader and submit their proposals. The initiator picks up the proposals and evaluates, chooses the nearest ones to join the pursuit team, sends accept-proposals to the selected agents and manages the team during the process of pursuit. If the evader is within the kill radius of all pursuit team members, the task is done; otherwise, D* algorithm is applied by each pursuit member to plan a new path once the evader moves to a new position. Distributive in nature, it allows a large-scale robot fleet to perform complicated tasks in a coordinated way. The algorithm can be easily extended to 3D cases. Simulation results show that the pursuit agents can dynamically form a pursuit team to catch evaders efficiently.