Resistant Q-Learning for Path Planning of Mobile Robot in Complex Environment

This study proposes a Resistant Q-Learning (RQL) algorithm to address the problems of low efficiency and slow convergence when mobile robots perform optimal path planning in unknown complex environments. The algorithm learns by utilizing the estimation deviation of the Q-values in Q-Learning. According to the number of updates of the Q-values estimation, the estimation deviation and learning rate are dynamically adjusted. By comparing this algorithm to Q-learning, the adapt-ability and convergence speed are greatly improved. The RQL algorithm is implemented on a mobile robot for path planning, and related experiments are conducted on simulated maps. The effectiveness and utility of the RQL are demonstrated by the simulation results.