Connected Automated Vehicles as Emergent Control Agents: Reinforcement Learning for Cooperative Traffic Control in Urban Networks

Connected Automated Vehicles as Emergent Control Agents: Reinforcement Learning for Cooperative Traffic Control in Urban Networks

Kennesaw, Georgia

Traffic congestion continues to pose a significant challenge, particularly in urban areas characterized by large populations and increasing vehicle ownership. Despite the implementation of various traffic management systems to address this issue, there remains substantial potential to introduce more advanced, intelligent and data-driven traffic management systems. These systems could further mitigate traffic congestion when integrated with existing measures. In this paper, we propose a reinforcement learning-based system (CAVRLS) that uses connected automated vehicles (CAVs) in mixed traffic environments to enable real-time centralized cooperative traffic control with the aim of reducing congestion in an inner congested network encircled by an outer uncongested network. To demonstrate CAVRLS integrated with Adaptive Traffic Signal Control Systems (ATCS) effectiveness, we simulate a traffic scenario characterized by a congested inner traffic network encircled by an uncongested outer traffic network. Subsequently, we utilize a Reinforce algorithm to control the following distance (spacing) for CAVs within the congested inner network based on the traffic state of the general network. Our experiment results show that our CAVRLS integrated with ATCS reduces traffic density, increases vehicular speed and traffic flow during the course of the centralized cooperative control implementation in the Boston traffic inner network compared to only using ATCS.