Toward a Smart City: Reinforcement Learning for Traffic Light Control

Abstract

AbstractFor decades, the optimization of traffic congestion problem has presented a significant issue for the transportation field. The continuous increase in the number of vehicles and the impossibility of building new high-capacity road infrastructures in big cities, which are densely populated and limited in space, make the alleviation of this risk a challenge for the scientific research community. The transportation researchers are called upon to deal with the complexities of ensuring effective management of intersections, which are in fact the nodes of road traffic congestion. But the real-world decision-making problems always involve uncertainty and indeterminacy, which lead to a paralyzed traffic light control system in some scenarios. Policemen address the traffic directly in an efficient way relying on their knowledge. This motivated the researchers to use Reinforcement Learning to create intelligent intersection traffic light management systems that can learn over time how to handle the intersection traffic flow based on real-time traffic conditions. This paper presents a brief overview of the three most used traffic data sources of a Traffic Management System and the most recently used Reinforcement Learning technics to handle the traffic light control problem. Finally, various experimental parameters that can affect the evaluation of traffic signal management methods are discussed.KeywordsSmart cityTraffic light controlIntelligent urban traffic management systemReinforcement learningArtificial intelligence