Research On Robot Path Planning Based On Deep Reinforcement Learnin

With the development of the intelligent era,the manipulator has become an irreplaceable thing in human life.Nowadays,people hope manipulator to do smarter things,not just for repetitive mechanical work.So,the research on the path planning algorithm of the manipulator is a matter of research significance.Combining artificial intelligence technology,this thesis studies the intelligent path planning algorithm of manipulator based on deep reinforcement learning.The main work of this thesis is as follows:The Rapidly-exploring Random Tree algorithm(RRT)will make the path planning inefficient when dealing with the local trap environment due to the characteristics of random points being selected with equal probability in the workspace.Combining the idea of reinforcement learning,an improved RRT algorithm based on the negative penalty mechanism is proposed.The comparison simulation results in the U-shaped environment show that the path planning efficiency of the improved RRT algorithm proposed in this paper is better than that of the ordinary RRT algorithm.Due to the poor autonomy and long time consuming for the existing path planning algorithms,this thesis proposes a deep reinforcement learning method for path planning of manipulators.The Deep Deterministic Policy Gradient algorithm(DDPG)and the Twin Delayed Deep Deterministic Policy Gradient algorithm(TD3)are selected as the training algorithm of the model.At the same time,the empirical sample grading mechanism is proposed and the Hindsight Experience Replay(HER)is used to improve the original algorithm and increase the convergence speed of the model network.The improved TD3 algorithm and the original algorithm are used to conduct path planning experiments in a two-link manipulator simulation environment built with the Python.The simulations verify the feasibility of the deep reinforcement learning algorithm in the path planning of the manipulator.For the six-axis manipulator UR5,a simulation environment is built in the V-REP.Under the premise of using only the positive kinematics of the manipulator,simulations were carried out on the improved RRT algorithm and the improved deep reinforcement learning algorithm proposed in this paper.The effectiveness of the algorithm proposed in this paper was verified.The software and hardware experiment platform for the actual manipulator was built,and the algorithm proposed in this paper was successfully applied to the path planning of the actual manipulator.A good foundation for the experimental research on the combination of virtual and real manipulator was laid.