Research On Path Planning Algorithm For Mobile Robots In Unknown Environment

As a crucial part of the automatic navigation and obstacles avoidance of mobile robots,the research of path planning algorithms has always been the focus and significant research direction in the field of robotics.Most of the existing robot path planning studies are based on the known environment,but in many cases,the environment of the robot is unknown.Therefore,how to realize navigation and obstacle avoidance of mobile robot in unknown environment is the hot and difficult point of path planning research,which has important research value.This paper not only proposes a novel path planning method which improves and integrates the traditional algorithms,but also put forward another path planning method which is based on the Q learning algorithm of reinforcement learning.The main contents of this paper are as follows:1.The path planning method of AAPF-RRT* algorithm with dynamic step is proposed.Aiming at solving the problem of excessive attractive and repulsive forces in the artificial potential field method with relative distances,this paper first proposes an Adaptive Artificial Potential Field(AAPF)method by improving the attractive and repulsive functions.At the same time,in order to solve the problems of blind expansion of nodes in randomly-exploring trees and insufficient smoothness of the planned path in the improved Rapidly-exploring Random Tree(RRT*)algorithm,this paper combines the AAPF algorithm with the RRT* algorithm,and original fixed step of the combined algorithm is changed in to dynamic steps.As a result,AAPFRRT* combined algorithm with dynamic step is proposed.Simulation experiments verify that the novel algorithm not only overcomes the problem of blind expansion of nodes in the RRT*algorithm,but also significantly improves the path planning efficiency and flexibility of obstacle avoidance of mobile robots in the unknown environment.Meanwhile smoothness of the planned path is guaranteed.2.The path planning method of RBF-Q learning algorithm based on behavior decomposition algorithm is proposed.In order to solve the "dimensionality problem" Qlearning system facing in complex and continuous environment,considering the local approximation ability of RBF neural network,RBF-Q learning path planning algorithm of mobile robots in unknown environments is proposed in this paper.The algorithm uses a threelayer RBF neural network to approximate the Q value of Q learning algorithm.Aiming at the problem that Q learning algorithm has low autonomous learning ability in path planning,the dynamic clustering method is used to train the input sample to determine the center and width parameters of the hidden layer in the RBF neural network,and the least mean square algorithm is used to update the weight between hidden layer and output layer in the RBF neural network.Aiming at the problem that the design of reward function is too simple in the application of Qlearning algorithm in path planning field,a new reward function based on behavior decomposition is designed in this paper,which decomposes the mobile robot’s navigation behavior into obstacle avoidance behavior and guiding target point behavior,and reward functions of different behaviors are designed respectively.Finally,RBF-Q learning path planning algorithm of mobile robots in unknown environments based on behavior decomposition is proposed in this paper.The algorithm not only significantly improves the mobile robot’s ability to generate better paths,greatly boosts the collision avoidance ability,but also accelerates the learning process of obstacle avoidance,thereby effectively enhances the mobile robot’s ability to adapt to unknown complex environments.