A Reinforcement Learning Based Adaptive Traffic Signal Control for Vehicular Networks

The congestion of urban traffic is becoming one of the serious issues with the increase in vehicles and population in cities. The static time traffic controlling system fails to manage traffic and leads to heavy congestion and crashes on roads. To improve traffic safety and efficiency proactively, this study proposes a Adaptive Traffic Signal Control (ATSC) algorithm to optimize efficiency and safety simultaneously. The ATSC works by learning the Optimal Control policy via Double Dueling Deep Q Network (3DQN). The proposed algorithm was trained and evaluated on simulated isolated intersection Simulation of Urban Mobility (SUMO). The results showed that the algorithm improves both traffic efficiency and safety compared with static time traffic control technique by 42%.