Path Planning with Q-Learning

Abstract

As science and technology advance rapidly, the scope of mobile robot applications continues to expand. In mobile robotics, path planning, which has formed a relatively complete theoretical system, is one of the most momentous and challenging tasks, especially in uncertain environments. Path planning has been a popular research problem in recent years, and it has been widely utilized in various fields, mainly including robotic surgery, logistics transportation, and self-driving automobile. To solve path planning, this paper presents an approach that uses Q-learning Algorithm to find multiple feasible paths within obstacle environments. First, the algorithm of Q-learning was pre-trained to make it suitable for path planning. Then an obstacle environment map was modeled and a path planning program was compiled by applying the state action-value function. Finally, the experimental results were collected, including the path planning maps and corresponding graphs of training progress. To guarantee the effectiveness of the final paths, all the parameters in the function are set to be constant. The experimental results show that the Q-learning algorithm can succeed in solving the problem of multi-path planning.