A Mobile Robot Path Planning Method Based on DQN with Hierarchical Empirical Storage Structure

Mobile robots take on the role of service and are widely used in daily life, such as dish-serving robots and floor-sweeping robots. Path planning and its optimization are critical and difficult for mobile robots in obstacle environments. In this paper, we propose a deep Q-network (DQN) algorithm with hierarchical empirical storage structure (HM-DQN) for exploring paths with optimal length and smoothness. First, the DQN algorithm utilizes a hierarchical experience storage structure to retain experience and classifies experience by judging the reward value, which ensures the diversity of stored experience and improves the efficiency of network parameter sampling. Second, sampling through a Bias-Free strategy when extracting experience for updating the network parameters eliminates the overfitting that may result from the bias to take high reward value experience for updating the network parameters. Finally, simulation experiments are carried out using the Turtlebot3 simulated mobile robot in a customized scenario, and the experimental results validate the performance of the proposed work.