Study On Local Path Planning Behavior For The Mobile Robot Based On Reinforcement Q Learning And BP Neural Network

The navigation technology of the mobile robot is a research hot spot in the field of artificial intelligence,including map-building,positioning,path planning,etc.In order to give the learning ability to the intelligent mobile robots and enhance its local path planning capability,it is necessary to study the local path planning behaviors of mobile robots by combining machine learning methods with learning characteristics.This paper studies the local path planning behavior of mobile robots based on the reinforcement Q learning algorithm and BP neural network model.According to the research requirements of path planning behavior,learning strategies and control rules are designed,corresponding control strategies are proposed for environmental perception,and simulation verification is performed.The main research contents cover:Based on the grid map environment,a local path planning algorithm for mobile robots based on CM-Q learning is proposed.Firstly,the algorithm designs the states and actions of the mobile robot based on the reinforcement Q learning algorithm and grid map,and establishes a Q matrix.Secondly,a coordinate matching(CM)obstacle avoidance control rule is designed to improve the mobile robot's obstacle avoidance efficiency.For the evaluation of action execution,a reward function is designed.Finally,experiments verify the effectiveness of the designed algorithm.The planned path of the mobile robot eliminates the redundancy which appears sometimes in the discrete and continuous obstacle environments by increasing the number of learning times and adjusting the learning rate.Based on environmental map in the form of free space and the generalization ability of BP neural network,a local path planning algorithm for mobile robots based on BPNN-Q learning is proposed.The algorithm firstly designs a sensor detection mechanism and action selection strategy based on the state of the mobile robot in the map environment.Secondly,a dynamic reward function is designed.Then according to the obstacle avoidance requirements including special "U" type obstacles,the obstacle avoidance rules for three shocks are designed.Finally,the effectiveness of the BPNN-Q local path planning algorithm is verified.The results show that mobile robot can perform better local path planning behavior with arbitrary starting and ending points in discrete and continuous free space maps,respectively,and the obstacle avoidance effect is good.It also works well in special "U" type obstacles environment.