Coordinated Ramp Metering Considering the Dynamics of Mixed-Autonomy Traffic

The introduction of connected autonomous vehicles may bring opportunities and challenges to traditional traffic control instruments, like ramp metering. This paper starts by constructing the fundamental diagram for mixed-autonomy traffic based on the car-following behaviors of both connected autonomous vehicles and human-driven vehicles. Then, analyses are performed on the main factors that influence the critical velocity, critical density, and road capacity under mixed-autonomy traffic. Two methods named COE-HERO and TRLCRM are developed to support the implementations of coordinated ramp metering for freeways with mixed-autonomy traffic. The COE-HERO method enhances the HERO method by incorporating a critical occupancy estimation module. Both COE-HERO and TRLCRM consider dynamic traffic flow parameters of mixed-autonomy traffic. The TRLCRM method is a reinforcement learning-based approach with a two-stage training framework, enabling it to adapt to varying mixed-autonomy demand scenarios. Extensive microscopic simulations show that the learning-based TRLCRM method can effectively alleviate bottleneck congestion and is robust to deal with various traffic scenarios. The COE-HERO method performs better than the HERO method, indicating the necessity of critical occupancy estimation in the implementations of coordinated ramp metering.