Path Planning Of Mobile Robots Based On Improved Reinforcement Learning

As one of the high-tech core technologies in the future,robot technology has become an important indicator to measure the comprehensive strength and technological innovation of country.Therefore,the development of the robot industry has already been paid attention by more and more countries,making the research and application of robotics received unprecedented support.As mobile robot belongs to intelligent robot,path planning is one of the indispensable technologies and it must be a key research content in the robot research.Therefore,the research on the path planning of mobile robot is addressed in the thesis.Firstly,the current research status of path planning at home and abroad is acquainted by referring to the literature,and then a few typical local path planning methods are briefly introduced,meanwhile the methods of superiority and inferiority are summarizedSubsequently,reinforcement learning as the path planning method is introduced,in order to realize mobile robot have adaptive ability and autonomous learning ability in unknown environment.Considering the dilemma of exploration and exploitation in Q-Leaning combined with ?-greedy strategy,an improving method of action strategy is proposed.This method,uses the learning status of robot in various state of environment to accomplish the self-adaptation of parameter ?,and uses the learning experience of robot to reduce the probability of bad action be selected,so that when the algorithm applied to path planning is better to balance exploration and exploitation.Therefore,Q-learning combined with improved action strategy ?-greedy,forming Q-learning based on learning status.Then,under different unknown environment,the Q-learning based learning status and Q-leaming based on other action strategies are applied to mobile robot path planning simulation,which proves that the Q-learning based on learning status has faster convergence speed and has better path than Q-leaming based on other action strategies.Aiming at the situation that the working environment of mobile robots will change in practical applications,a solution is proposed.This solution is a multi-layer path planning by composed the method of rolling window with the Q-learning based on learning status.In the multi-layer path planning method,global path planning is performed by Q-learning based on learning conditions,and the rolling path method is used to perform local path planning for obstacle avoidance.And in the local path planning,in order to improve the quality of local path planning,the heuristic information of the planned global path trajectory is added on the basis of the traditional mapping method.Enable robot can utilize the planned path trajectory when robot performs local path planning,improving the quality of local planning.At last,the effectiveness of the multi-layer path planning is tested by simulation of extensive situation.