Robot Path Planning Based on Deep Reinforcement Learning

摘要

Deep Q-Network (DQN) algorithm has some problems such as overestimation and poor algorithm stability in the process of robot path planning. To solve the above problems, Double Deep Q-Network (DDQN) algorithm is combined with the Averaged Deep Q-Network (ADQN) algorithm to reduce the overestimation problem and improve the stability of the algorithm. Using priority experience replay instead of average sampling can improve the utilization rate of valuable samples and enhance the learning efficiency of the robot. The improved algorithm is compared with traditional DDQN algorithm in simple and complex simulation environment. Experimental results show that the improved algorithm has higher convergence rate, higher reward value and better path planning.