Collision Avoidance Dynamic Window Approach in Multi-agent System

Abstract

In multi-agent system formation switching, one common mission is to exchange positions of two agents. This scenario is prone to become a local minimum, since for each agent the direction to target and the direction of another agent movement are col-linear and opposite. Collision will occur if collision avoidance method fails to solve local minimum. Artificial Potential Field (APF) and Dynamic Window Approach (DWA) are classical collision avoidance methods. However, neither can be directly used in this mission. APF easily causes agent trapped in local minimum, which is unpredictable especially in dynamic environment. While DWA has collision avoidance term in objective function, inappropriate weigh parameters may block collision avoidance behavior. To cope with this mission, this paper proposes a new DWA variant called Collision Avoidance Dynamic Window Approach (CADWA). It is divided into collision risk detection part and collision avoidance part. Contributions of CADWA include: i) Neighbor set describes neighboring agents and its derivatives detect potential collision; ii) The objective function only focuses on reducing collision risk, so no local minimum will appear; iii) Besides preserving DWA merits like easiness to understand and fast reaction, CADWA does not need parameter tuning; iv) It is convenient to integrate CADWA into formation control algorithm which does not consider collision between agents. Details of CADWA algorithm are described. Simulation results show that two agents finish formation switching without collision.