A Novel Multi-robot Coordination Method Based on Reinforcement Learning

摘要

Focusing on multi-robot coordination, role transformation and reinforcement learning method are combined in this paper. Under centralize control framework, the distance nearest rule which means that the nearest robot ranges from obstacles is selected to be the master robot for controlling salve robots is presented. Meanwhile, different from traditional way which reinforcement learning is applied in online learning of multi-robot coordination, this paper proposed a novel behavior weight method based on reinforcement learning, the robot behavior weights are optimized through interacting with environment and the coordination policy based on maximum behavior value is presented to plan the collision avoidance behavior of robot. The learning method proposed in this paper is applied to the application related to collaboration movement of mobile robots and demonstrated by the simulation results presented in this paper.