Research on the Local Path Planning for Mobile Robots based on PRO-Dueling Deep Q-Network (DQN) Algorithm

摘要

This paper proposes a Pro-Dueling DQN algorithm to solve the problems of slow convergence speed and waste of effective experience of the traditional DQN (Deep Q-Network) algorithm for the local path planning of mobile robot. The new algorithm introduces a priority experience playback mechanism based on SumTree to avoid forgetting the learning effective experiences as the number of samples in the experience pool increases. A more detailed reward and punishment function is designed for the new algorithm to reduce the blindness of extracting experience in the early stages of algorithm training. The feasibility of the algorithm is verified by comparative verification on ROS simulation platform and real scene, respectively. The results show that the designed Pro-Dueling DQN algorithm converges faster and the length of planned path is shorter than that of the original DQN algorithm.