The Research Of Mobile Robot Patn Planning Based On Reinforcement Learning

In the study of robot research, navigation technology is the basis of mobile robot research, and path planning is the basis of navigation technology. Therefore, it is important to study the mobile robot path planning technology to improve the adaptability and the navigation ability of mobile robot. In the unknown environment ,the robot is short of experience, which requires the robot to keep learning and increase experiences to improve the ability of optimization capability in the path planning process. Based on the study of reinforcement, this paper takes Q learning as the main mobile robot path planning algorithm,for the Q learning process to balance between exploration and utilization problem,the design of reward and punishment function problem and generalization of Continuous State Space problem,we designed the corresponding solutions.In order to solve the balance between exploration and utilization problem and the problem of reward and punishment function which influence the convergence rate of algorithm in Q learning process,a simulated annealing-Q learning path planning algorithm based on behavioral decomposition is used.The reward and punishment function is designed as a nonuniform reward and punishment function based on behavioral decomposition,That is, the rewards and punishments function is designed as a weighted sum form of Obstacle avoidance reward function and Oriented reward and punishment function,In order to solve the problem of balance between exploration and utilization in learning process, an action selection strategy based on simulated annealing algorithm is designed.SA-Q learning path planning algorithm based on fuzzy control is a good way to improve the generalization of continuous state space issue.we used the fuzzy control algorithm to generalize the continuous state space in a complex environment,and through the fuzzy reasoning to get the output action,and updating the q -valued function by updating fuzzy rule base.In the design of path planning,the improved scheme for reward and punishment function and simulated annealing algorithms is introduced.Finally, the simulation results of MATLAB shows that the algorithm is effective.