Participatory traffic control: Leveraging connected and automated vehicles to enhance network efficiency

This paper aims to establish a framework of participatory traffic control, wherein connected and automated vehicles (CAVs) subtly influence the day-to-day adjustment process of human drivers, strategically redistributing traffic demand to enhance overall system efficiency. To address this complex challenge, we adopt the mean-field control framework, which enables us to model macroscopic interactions between CAVs and other travelers. After theoretically establishing the existence of the optimal policy, we leverage reinforcement learning algorithms to numerically solve the control problem. Distinct from existing approaches, our proposed method is scalable, model-free, distributed, and does not rely on the convergence properties of the underlying day-to-day traffic dynamics. It helps pave the way for the practical implementation of participatory traffic control.