Expert-guided Triple-memory Deep Reinforcement Learning for Robotic Motion Planning
Huashan Liu, Guojing Zhang, Xiangjian Li, Xinjie Xiao · 2023
Off-policy deep reinforcement learning (DRL) mostly use memory to improve the processes of data sampling. However, it is difficult for DRL to obtain positive feedback in the early stage of the training. Although an expert memory can solve the problem of sparse rewards in the early stage of training, it can not ensure to obtain the optima, and even leads to overfitting, which affects the performance of the agent in the later stage. In this paper, an expert-guided triple-memory (ETM) framework is proposed to be applied in off-policy DRL, so as to enhance the convergence performance and enhance the stability of the algorithm. Experimental results show that, the proposed approach is feasible and superior in the obstacle-avoidance motion planning tasks for robots.