Design Of Path Planning Method For Mobile Robot Based On Reinforcement Learning

With the development of mobile robots,more and more experts pay attention to the path planning technology applied to mobile robots.Mobile robots lack a priori knowledge of the environment in dynamic and unknown static environments,which requires strong flexibility and adaptability to cope with various situations.Reinforcement learning is an important branch of machine learning,which acquires knowledge through agent's exploration in the environment and learns in the process of trial and error.Mobile robots will inevitably encounter various obstacles in the process of path planning,which requires that the path planning method of mobile robots designed can be flexible and adaptable to the environment.Therefore,it is of great practical significance to endow robots with autonomous learning ability.Firstly,the research background and significance of path planning method for mobile robots based on reinforcement learning are described.The current research status of path planning method and reinforcement learning method for mobile robots at home and abroad are analyzed,and the main work and framework of this paper are briefly introduced.Secondly,the basic principle of reinforcement learning method is introduced in detail,and the reinforcement method is further analyzed from the system composition.By introducing three classical reinforcement learning algorithms in detail,the strategy iteration method and the optimal strategy acquisition method of reinforcement learning are explained.The application of reinforcement learning method in the field of path planning is summarized,and the existing problems are analyzed.Thirdly,taking the path planning of mobile robots in static unknown environment as the research background,the reinforcement learning method is used to design the path planning method based on Q-learning algorithm,and the static grid environment is built for simulation experiments.The experimental results show that mobile robots can find an optimal path from the starting position to the target position through autonomous learning in unknown environment.In order to solve the deficiency of table-valued learning method in data processing ability in large-scale state space,the potential field method is used to optimize DQN algorithm.On the one hand,the advantage of neural network in data processing is utilized,On the other hand,the potential field method is used to guide the accelerated training process.The path planning method is designed and the complex static environment is built to carry out the simulation experiment,the experimental results show that the improved algorithm can be complex.Effective path planning in complex static environment.Fourthly,aiming at the problem of path planning for mobile robots in dynamic environment,a simulation environment with dynamic obstacle elements is built,and the simulation experiment is carried out by using A3 C algorithm to design the path planning method.The experimental results show that the mobile robot can effectively plan the path in the environment,and can deal with sudden dynamic obstacles.Then the simulation environment is built based on the background of UAV automatic driving.Double DQN algorithm is used to reduce the overestimation of the objective function to reduce the influence factors in the dynamic obstacle environment.A local path planning method for UAV is set up to carry out simulation experiments.The experimental results show that the collision-free driving distance of UAV in the environment increases with the training process.Fifthly,according to the kinematics model and constraints of Pioneer 3 mobile robot,the simulation environment is built,and the path planning method based on DDPG algorithm is designed for simulation experiment.Find the best action in the continuous action space,and get an optimal path through training.The experimental results show that this method can select actions in the continuous motion space of mobile robots and plan an optimal path without collision.Finally,the work done in this paper is summarized and the future research work is prospected.