Impact of Connected Vehicles on Learning based Adaptive Traffic Control Systems

Adaptive Traffic Signal Control (ATSC) systems can be implemented to reduce travel times at urban intersections by changing the signal program according to real-time traffic situations. Modern approaches to ATSC are based on Reinforcement Learning (RL) which can allow the controller to learn the control policy independently. By including the concept of Connected Vehicles (CVs), the RL-based ATSC system can use data gathered from CVs instead of traditional traffic sensors. In this paper, the impact of varying CV penetration rate on RL-based ATSC is implemented and evaluated in a simulated environment. Obtained results show that with a sufficient CVs penetration rate the RL-based ATSC systems can significantly reduce the delay of all vehicles in the traffic network.