A leading vehicle-aware car-following control strategy for autonomous vehicles and its impact on heterogeneous traffic flow

Car-following control is a fundamental component of autonomous driving systems and directly affects traffic safety, efficiency, and ride comfort. However, most existing studies fail to account for the driving behavior characteristics of the leading vehicle, which vary substantially with human driving styles and vehicle types. To address this gap, this study proposes a leading vehicle-aware car-following control strategy (denoted as LVA-TD3) for autonomous vehicles (AVs) based on deep reinforcement learning. Specifically, human driving style and vehicle type are incorporated to characterize the car-following behavior of human-driven vehicles (HDVs) and are further embedded into the state representation perceived by the AV in the simulation environment. In addition, a reward function incorporating safety, efficiency, comfort, and environmental impact is formulated to guide the policy learning process based on the twin delayed deep deterministic policy gradient (TD3) algorithm. Finally, a three-lane freeway on-ramp scenario is constructed to examine the impact of the proposed LVA-TD3 controller on heterogeneous traffic flow. The results demonstrate that the LVA-TD3 controller effectively mitigates traffic oscillations and enhances traffic safety, operational efficiency, and environmental performance by increasing fuel economy and reducing emissions. These findings suggest that the proposed approach has considerable potential for facilitating the design and implementation of advanced driver assistance systems.