Research On Multi-Robot Cooperation Control Method Based On Reinforcement Learning

Multi-robot cooperation is an important subject in the field of robot, which can finish task individual robot can't do, so more scholars research on robot cooperation. But for multi-robot system, if only controlling parameters are provided, effective cooperation will be difficultly finished, it is a waste of resource. To make multi-robot cooperation effective in the complex and unknown environment, researchers pay more attention to the learning ability of multi-robot, aiming to optimize performance and enhance the ability of robot cooperation and effectively complete the task.Reinforcement learning developed in recent years is an important machine learning method. It is an optimal strategy through the perception of environmental information. Robot has a self-learning ability through constant trial and error, and interaction with the environment to improve their behavior. So the main job of this paper is as follows:(1) This paper systematically analyzed the concept and organizational behavior of multi-robot system, which uses the reinforcement learning algorithm and researches on the path planning. Through trial and error and evaluation of online learning, it makes the robot move from the starting point to the target point by selecting the optimal path to obtain the required planning rules for motive behavior, thus lays the theoretical basis for the further study control methods of t cooperation behavior of multi-robot system.(2) A mixed reinforcement learning method of multi-robot cooperation hunting is proposed in this paper,which divides reinforcement learning system into learn split subsystems to solve the complex problem of cooperative hunting, and uses fuzzy logic to solve the big state-spac problemof multi-robot system, simultaneously, it introduces heuristic reward and punishment function to improve the efficiency of searching target robot, and it uses the blackboard and coordination based on the consultation to solve behavior of conflict, simulation test by object-oriented programming technique verifies the effectiveness of the above methods.