Automated Vehicle Overtaking based on a Multiple-Goal Reinforcement Learning Framework

Abstract

In this paper, we propose a reinforcement learning multiple-goal framework to solve the automated vehicle overtaking problem. Here, the overtaking problem is solved by considering the destination seeking goal and collision avoidance goal simultaneously. The host vehicle uses Double-action Q-Learning for collision avoidance and Q-learning for destination seeking by learning to react with different motions carried out by a leading vehicle. Simulations show that the proposed method performs well disregarding whether the vehicle to be overtaken holds a steady or un-steady course. Given the promising results, better navigation is expected if additional goals such as lane following is introduced in the multiple-goal framework.