Deep Reinforcement Learning for Smart Traffic Control

Abstract

Traffic jams in urban areas have become a serious issue in the globe with the consequent impact of the long travelling time, use of fuel and pollution. The traditional traffic control schemes are time-based or rule-based schemes, which is not adaptable to the dynamics in the traffic conditions. The present advancement in Deep Reinforcement Learning (DRL) provides a promising framework of smart and responsive controller of traffic lights. This research paper is a summary of the DRL methods applied to implement smart traffic management including model architecture, training environment, performance measures, and deployment challenges. A provided model of traffic signal control by DRA is developed and experimented using simulation through the application of SUMO (Simulation of Urban MObility). The results indicate that large reduction in the mean waiting time and the queue length is also achieved over the conventional fixed time and actuated time control systems. As noted in the research paper, one of the opportunities the DRL can be used to transform the traffic control of the smart city.