MADDPG Algorithm for Coordinated Welding of Multiple Robots

This paper deals with the coordinated welding problem of multi-robot systems by applying a deep reinforcement learning algorithm which is called multi-agent deep deterministic policy gradient (MADDPG). It is assumed that the states and actions of robots are continuous and each robot can only get local information of its neighbors. A novel reward composed of trajectory optimization, coordinated welding and collision avoidance is designed, which yield multi-robot systems to arrive the welding targets quickly, achieve the simultaneous welding for a weld line and avoid collision among robots. Simulation results are provided to demonstrate the effectiveness of the proposed control algorithm.