Research on Path Planning for Mobile Robots Based on Improved Q-learning Algorithm
Yubo Ma, Xiuqin Pan, Ruixiang Li, Jiayun Li · 2024
Path planning, as an important direction in robotics research, is the first condition to ensure that the robot accomplishes the task. This study presents an optimized version of the Q-learning algorithm for robot path planning, taking into account the limitations of the Q-learning algorithm's sluggish search speed and lengthy planning time. Firstly, the original four-neighborhood exploration is changed to eight-neighborhood exploration, and the robot's action space is dynamically adjusted according to the real-time relative locations of the objective point and the robot to reduce the exploration of invalid actions; Secondly, based on the attributes that call for more exploration in the initial phase and quicker convergence in the subsequent phase, a method with gradually increasing learning rate is proposed. Finally, the outcomes of the experiment indicate that the improved Q-learning algorithm is not only able to find the shortest path, but also has a faster convergence rate.