Research on Path Planning Based on Q-learning

Abstract

With the advancement of technology, the path planning and navigation technology of robots is more mature than before. However, as the environment becomes increasingly complex, the limitations of traditional navigation methods are becoming more prominent. To overcome this challenge, reinforcement learning was introduced in the path planning problem. As a typical reinforcement learning algorithm, Q-learning does not rely on models. Among them, factors such as learning rate and discount coefficient have a significant impact on Q-learning results. Based on this, this article investigates the influence of Q-learning learning rate and discount factors on robot path planning in certain maps, explores the trend of the number of iterations required for robot training changing with learning rate and discount factors, and reflect the trend of the number of iterations changing with the learning rate and discount factor on the line graph, thus obtaining a universal conclusion, aiming to provide ideas for future research in Q-learning.