Research On Path Planning Of Mobile Robot Based On Reinforcement Learning

In recent years,with the development of modern science and technology,especially computer technology,electronic communication technology and control technology,the performance of mobile robot has been constantly improved,and it has been applied to medical and health,national defense and military,aerospace,machinery manufacturing,education entertainment and other fields in various degrees.The premise for mobile robot to complete specific tasks in each application is to plan an effective path from the starting point to the target point,which makes the path planning technology has always been a research hotspot in the field of mobile robot research.With the gradual maturity of artificial intelligence technology,intelligent algorithms for path planning of mobile robots have been widely studied.Among them,reinforcement learning theory does not need to predict the environmental model,but interactive learning with the environment through "trial-and-error" method,which provides theoretical support for robots to understand environmental information.In this paper,the traditional reinforcement learning algorithm is used to solve the problems such as low convergence efficiency,insufficient convergence accuracy,non-convergence of large state space and inapplicability of continuous state space in the path planning of mobile robots.The research work is mainly carried out in the following aspects:Firstly,a path planning method of mobile robot based on DBQ algorithm is proposed.The environment model was defined by the formal description of the environment information,and the action selector BPAS based on BP neural network was constructed according to the environment information and action rules.The action selection mechanism in dyna-q algorithm was improved according to the way of BPAS to select actions,and the random strategy was combined to avoid the robot falling into a dead corner area.At the same time,the reward function is designed according to the characteristics of DBQ algorithm,the convergence condition is defined,and the effectiveness of the algorithm in planning effect,convergence,average cumulative reward value,execution efficiency,balance between learning and planning are discussed through simulation experiment.Secondly,on the basis of DBQ path planning algorithm,a path planning method based on RDBQ algorithm is proposed according to the idea of approximate reinforcement learning.RBFNN is used to construct the value function approximator to fit the value function table in DBQ algorithm,which makes the algorithm suitable for large state space and continuous state space.At the same time,a hierarchical planning strategy based on RDBQ algorithm is proposed on the basis of global path planning by DBQ algorithm.The strategy adopts the method of "offline" and "online" learning to conduct real-time monitoring of dynamic obstacles in the static obstacle avoidance environment of the robot,so as to realize the path planning process of the robot in the dynamic environment.Simulation experiments were conducted to observe the planning effect of RDBQ algorithm in large state space,analyze the algorithm performance,and verify the effectiveness of hierarchical strategy in dynamic obstacle environment.Finally,verification experiments and analysis are carried out in the real environment.The Pioneer3-DX hardware platform and the experimental scene of the teaching building corridor are selected.The effect and operation efficiency of the proposed algorithm in the real environment are analyzed,and the effectiveness and stability of the algorithm are further proved.