Developing Smart Lane-changing Strategies for CAVs on Freeways based on MOBIL and Reinforcement Learning

Abstract

This paper is concerned with the impacts of smart lane changes of connected and automated vehicles (CAVs) on their own travel performance as well as the entire traffic flow. Based on MOBIL and reinforcement learning, two ego-efficient lane-changing strategies were developed in this work to enable lane-changing decisions for CAV s to improve their travel efficiency. The MOBIL approach intends to establish such a lane-changing strategy by optimizing MOBIL's two parameters, while the reinforcement learning approach tries to develop such a strategy from scratch using Q-learning with sufficient traffic environmental information. The lane-changing strategies were developed and compared on the basis of intensive microscopic traffic simulation. In addition, the information impact on the performance of the reinforcement learning approach was examined to determine the essential amount of environmental information required.