Real-Time Path Planning Through Q-learning's Exploration Strategy Adjustment

Abstract

As the usefulness of reinforcement learning has been confirmed through AlphaGo, research on applying reinforcement learning to the field of autonomous driving is actively progressing. In this paper, we implemented path planning using Q-learning, one of the reinforcement learning algorithms. Path planning using Q-learning has a limitation in that it is difficult to store tables for all environments. To overcome this limitation, we used real-time Q-learning that does not store tables in advance. We Adjustment our exploration strategy to the learning speed required for real-time Q-learning. It was confirmed that Q-learning using an appropriate exploration strategy enables real-time path planning.