One fast RL algorithm and its application in mobile robot navigation

Abstract

Reinforcement learning is a method for search optimal strategy on condition of unknown apriority knowledge. When the learning tasks are complex and work condition dynamically change, learning speed is too slow. For this problem, a kind of speedup reinforcement learning algorithm based on learning experience replay is proposed in this paper. Firstly, the experience sample database is built gradually in the learning process. Secondly, the efficiency of reinforcement learning is improved by experience samples replaying. Finally, the presented method is used to solve the problems of mobile robot navigation, the validity is testified.