Research On Local Path Planning For The Mobile Robot Based On Reinforcement Learning Algorithm

With the rapid development of artificial intelligence,robots are booming in the fields of industry,agriculture and scientific research.The research fields represented by navigation technologies such as location,path planning and path tracking,have become a hot topic in robot research.With the increasing complexity of the robot's task environment,it has become a new trend for the robot to plan path through interactive learning in real-time with its environment.Based on the reinforcement learning represented by QL algorithm,this paper studies how to obtain the maximum cumulative return and optimal control rules to achieve the local path plan of the mobile robot.This paper proposes a local path planning method for the mobile robot based on the fuzzy-QL algorithm to improve the common problems existed in the traditional method,such as the slow convergence rate,balance of the the dilemma between exploration and exploitation,and avoidance of the hazardous running areas.First,the variables of state and action are designed and fuzzed according to the planning problems;then a Q-value function matrix is designed to store the reinforcement value,and a reward function is constructed according to the requirements of obstacle avoidance and shortest path;using ?-balance strategies between exploration and exploitation,the updated step and the selection algorithm of actions to improve the learning process of QL.After training,the optimal pairs of state and action are obtained,further the optimal fuzzy control rules are achieved.Finally,the fuzzy control rules are utilized to perform local path planning.In order to prevent the deadlock problem,the preventive measure is designed to improve the efficiency of the path planning in the fuzzy control rules.A steering rule table is proposed to solve the visiting inadequate of pairs of state and action.In the process of planning the path with the control rules,the algorithm designs the steering rules according to the measured distance of the three sensors to create more learning opportunities of pairs of state and action,and increase the learning probability of each pair.So the learning efficiency of QL algorithm is improved and the planned path is more stable.Finally,an experimental simulation platform is designed based on MATLAB software.The simulation results show that,the robot based on the fuzzy-QL algorithm can effectively avoid obstacles in a complex environment,thus this algorithm can accelerate the convergence speed of the algorithm and balance the problem of exploration and utilization.Then the robots can jump out of the deadlock area,and plan the optimal or sub-optimal path.The solution with the visits of pairs of state and action being adequate strengthens the traversal learning of state-action pairs and enhances optimization of path planning.