Research On Path Planning Strategy Of Mobile Robot Based On Artificial Potential Field And Reinforcement Learning

Mobile robots play an important role in human life,and path planning is its core technology.Therefore,path planning for mobile robots has been the focus of research in the field of autonomous navigation.The working environment of robots in today’s world is complex and variable,and path planning tasks not only need to deal with trap areas composed of multiple static obstacles,but also deal with some unknown dynamic obstacles.Traditional path planning algorithms are difficult to meet path planning tasks in complex environments due to their limitations.Scholars have made a lot of improvements to traditional path planning algorithms,but there are still problems such as long planning paths,large changes in planning steering angles,and slow algorithm convergence.This paper studies the path planning problem of mobile robots based on artificial potential fields and deep reinforcement learning under different maps.The main research work is as follows:(1)An improved artificial potential field path planning algorithm based on variable neighborhood search and reinforcement learning strategy is proposed to solve the problems of local minima,lengthy paths,and path oscillations when robots plan their paths in a trap map in a static environment.Firstly,by adding a variable neighborhood search algorithm to search for sub target points,the robot is prevented from going deep into the trap area,and the problems of local minima and lengthy paths are solved.Secondly,a learning feedback strategy is introduced into the variable neighborhood search algorithm to avoid invalid and repetitive searches,which solves the slow convergence problem of the algorithm.Finally,an angle limiter and corrective force mechanism are designed to solve the path oscillation problem by reducing the degree of course angle variation.The simulation results show that the optimal path length of the improved artificial potential field algorithm is shortened by 3.5%,and the total steering angle change is reduced by 40.6%,verifying the effectiveness and superiority of the improved artificial potential field algorithm in complex environments.(2)Aiming at the problem of robot path planning in unknown dynamic environments,such as falling into local optimization,slow convergence speed,and inability to optimize dynamic obstacle avoidance,so a fusion path planning algorithm based on artificial potential field and deep Q-network(DQN)is proposed.Firstly,an improved DQN algorithm is added for global path planning,and a subobjective point guidance strategy is adopted at the local path planning level to overcome the local optimization problem.Secondly,by adding an initialization position redefinition strategy,the problem of difficult to find early successful experiences of the DQN algorithm in complex environments is solved,and the convergence speed is accelerated.Then,by adding a reward structure based on the global guidance strategy,the problem of slow convergence speed caused by the unreasonable reward structure of the DQN algorithm is solved.Finally,the dynamic obstacle avoidance problem is solved by adding a virtual gravity dynamic obstacle avoidance strategy.Simulation and experimental results show that the optimal path length of the improved fusion algorithm is reduced by 8.2%,and the total heading angle is reduced by 20%,verifying the superiority and effectiveness of the improved fusion algorithm in unknown dynamic environments.