An Algorithm for Path Planning Based on Improved Q-Learning

Abstract

The path planning is a basic research question, where an Agent is going to look for an optimal path from the beginning to the terminal point. The earliest research problem was obstacle avoidance, that is, the Agent could not collide with obstacles or fall into traps in the process of path optimization. With rapid development of the AI (artificial intelligence) technology, the path planning has become an important application field to test intelligent learning algorithms, where reinforcement learning as an active learning method has the obvious advantage in path planning compared with traditional supervised learning and unsupervised learning. As is known to all, Q-Learning is one of the most successful reinforcement learning algorithms, and it surely can help solve the path planning. Aiming at the problems of slower convergence of the existing Q-Learning algorithm, this paper introduces dynamic search factor technology and puts forward a novel ɛ-Q-Learning algorithm. Experiments show that compared with the existing Q Learning algorithm, the ɛ-Q-Learning algorithm can not only present a better optimal solution, but also significantly reduce the number of iterations to quickly complete the optimal path generation.