Research And Implementation Of Unmanned Vehicle Path Planning Based On Reinforcement Learning

With the progress of various fields of science and technology,intelligent robots began to infiltrate into various industries,in which the research and development of intelligent vehicle autopilot has been paid more and more attention.In the research of unmanned vehicle self-driving,path planning and autonomous obstacle avoidance in dynamic environment are the main problems of current research.The principle of obstacle avoidance is that the unmanned vehicle first obtains its own position and surrounding environment information through the installed on-board perceptron,especially the distance from the surrounding obstacles,and then calculates the minimum safe distance between the current unmanned vehicle and the obstacle,and generates a safe obstacle avoidance strategy.Automatic control of unmanned vehicles to avoid obstacles safely.Path planning also needs to rely on perceptron to sense the surrounding obstacle information,as well as to detect the location of the target.The grid is used to divide the driving path of the unmanned vehicle,the Q-learning algorithm is used to generate the Q-table list,the value of each position of the unmanned vehicle is calculated,and the greedy strategy is used to generate the optimal path to the direct destination.Make the unmanned vehicle automatically follow the optimal path to the target position.In this paper,the theory of neural network algorithm and Q-learning algorithm in reinforcement learning and their solutions to obstacle avoidance and path planning of unmanned vehicles in unknown dynamic environment are studied.the specific work is as follows:Firstly,according to the real vehicle driving path,the driverless intelligent vehicle system is built on the PyCharm platform by using Python language,which mainly includes the setting of dynamic and static obstacles,the setting of target point and the setting of unmanned vehicle driving mode.These are the basic conditions for the self-driving experiment of unmanned vehicles.Secondly,by analyzing the obstacle avoidance method of unmanned vehicle and the analysis of possible obstacle avoidance methods,the constraint conditions of unmanned vehicle safety obstacle avoidance are studied,and the BP neural network algorithm is combined with the constraint conditions.The safety obstacle avoidance strategy of unmanned vehicle in unknown dynamic environment is generated,and the unmanned vehicle is successfully controlled to leave the dangerous area safely.Thirdly,the grid environment is established and simulated by Q-learning algorithm.it is pointed out that the planned path is not optimal because of the premature reduction of the exploration factor in the training process.Therefore,the traditional autopilot path explorationmethod is improved,so that all the action strategies are fully explored and utilized,and the global optimal planning path is obtained.The maximum number of training times is required to ensure the training efficiency,so as to achieve the double optimization of the experimental effect and the experimental time from two aspects.Fourthly,in order to optimize the path planning strategy and improve the experimental efficiency,a new training and learning method,hierarchical training,is proposed.The whole system environment is divided into several independent areas,so that unmanned vehicles can train and learn in valuable areas,carry out local planning,and save exploration and training time in worthless areas.after the training is completed,The information of each region is aggregated into Agent to realize the overall planning.finally,the optimal path to the target position is generated in the whole environment to improve the learning efficiency.The simulation results show that the optimization method proposed in this paper plays an active role in autonomous control obstacle avoidance and autonomous path planning of unmanned vehicles.It is of great significance to promote the development of unmanned vehicle obstacle avoidance and path planning research for the practical application of unmanned vehicle self-driving.