The rapid growth of vehicle-to-infrastructure (V2I) enables the efficient and timely communication at the signalized intersection for the multi-mode traffic participants: such as emergency vehicles, transit bus, commercial trucks, and passenger cars. This study proposes a framework for adaptive signal control to address the multi-mode traffic control problem. We develop a method to extract the aggregate information for the intersection traffic state representation from massive data collected by the central controller. This information extraction technique can achieve a balance between information loss and heavy computation burden. Then an offline adaptive signalized strategy using reinforcement learning is formulated to provide the real-time signal control plans using online traffic data as input. This algorithm enables the signal controller to respond timely to priority requests from different modes of traffic. Finally, the simulations are implemented under both under-saturated and oversaturated traffic conditions to validate the algorithm efficiency under stable and unstable traffic conditions. Finally, the numerical experiments show that, compared to the optimized fixed time plan (Synchro), the proposed method can reduce delay for different vehicle modes.

CICTP