Towards Foresighted and Cooperative Vehicular Lane Changes: An Edge AI Approach

Connected and Automated Vehicles (CAVs) will revolutionize the future intelligent transportation system by enabling fully automated vehicular control such as accelerating, steering and lane changes, etc. Among them, lane changes are vital aspects since they require not only local CAV control, but also interactions with surrounding CAVs to guarantee the overall traffic efficiency. Nevertheless, the Line-of-Sight (LoS) sensing range adds significant limitations on CAVs to efficiently make lane change decisions. Moreover, the self-interested CAV lane change decision ignores its impact on the surrounding CAVs as well as the entire traffic flow. In this article, we propose an edge Artificial intelligence (AI) empowered CAV framework to enhance the sensing and decision making abilities of CAVs and thus assist in efficient lane changes with foresighted and cooperative vehicular intelligence. Under the proposed framework, Non-Line-of-Sight (NLoS) lane environment data can be collected by the edge AI node and aggregated with the LoS data collected by the onboard CAV sensors. The aggregated data can be utilized to train a deep reinforcement learning model that enables CAVs to make foresighted lane change decisions by learning from the NLoS environment data. Besides, we incorporate cooperative game theory with the deep reinforcement learning model to help CAVs evaluate their impact on the surrounding CAVs and make cooperative lane change decisions to improve the overall traffic efficiency. Simulations are conducted to prove the effectiveness of the proposed edge AI empowered CAV lane changes.