Decentralized Multi-Robot Motion Planning Applicable to Dynamic Environment
Abstract Multi-robots navigation in dynamic environment is a promising topic in intelligent robotics with motion planning being one of the fundamental problems. However, in practicel, multi-robots motion planning is challenging with traditional centralized approach since computational demand makes it less practical and robust for the motion planning of a large number of robots. In this paper, a decentralized distribute robots motion planning framework (DDRMPF) is discussed which addresses the specific issue. DDRMPF directly maps raw sensor data to steering command to generate optimal paths for each constituent robot. Unlike centralized method which needs a complete observation along with a center agent which processes heavy data collected from all the robots, DDRMPF allows each agent to generate an optimal local path needing only partial observation, thus rendering motion planning involving large numbers of robots more practical and robust. DDRMPF trains the policy for each robot in the complex and dynamic environment simultaneously based on the reinforcement algorithm.