Lane change is an important behavior of autonomous vehicles. Safe and reasonable lane change behavior can increase the speeds of autonomous vehicles on the road and improve the traffic efficiency. Before lane changing, the autonomous vehicle needs to obtain the status of the surrounding vehicles and judge whether the lane change requirements are met. The traditional rule-based vehicle lane change model can perform well in the pre-defined scenarios, but they may be prone to failure when unexpected situations are encountered. To solve this problem, we proposed a Reinforcement Learning based approach to train the autonomous vehicle agent to learn an automated lane change behavior. Firstly, we defined a continuous action space and a continuous state space to enhance the feasibility of the solution in real-world problems. Then, a quadratic function was applied as Q-function approximator to learn autonomous vehicle lane change behavior by estimating the total return in the lane change process. The experiment results show that the average speed of the vehicle is faster by using the proposed method compared with the common Reinforcement Learning algorithm and the proposed algorithm can control the autonomous vehicle to change lanes safely and efficiently.

CICTP