CVLight: Decentralized learning for adaptive traffic signal control with connected vehicles

This paper develops a decentralized reinforcement learning (RL) scheme for multi-intersection adaptive traffic signal control (TSC), called “CVLight”, that leverages data collected from connected vehicles (CVs). The state and reward design facilitates coordination among agents considers travel delays by CVs. A novel algorithm, Asymmetric Advantage Actor-critic (Asym-A2C), is proposed where both CV non-CV information used to train the critic network, while only execute optimal timing. Comprehensive experiments show superiority of CVLight over state-of-the-art algorithms under 2-by-2 synthetic road network with various demand patterns penetration rates. learned policy then visualized further demonstrate advantage Asym-A2C. pre-train technique applied improve scalability CVLight, which significantly shortens training time shows in performance 5-by-5 network. case study performed on located State College, Pennsylvania, USA, effectiveness algorithm real-world scenarios. Compared other baseline models, trained agent can efficiently multiple intersections solely based achieve best performance, especially low