A Survey on Deep Reinforcement Learning for Traffic Signal Control

Traffic congestion is one of the most important and complex problems in urban governance for a long time. Although traffic lights are used at intersections, traffic bottlenecks still appear with the increasing number of private cars. In recent years, with the continuous development of related technologies in the field of intelligent transportation, more attention has been paid to the automatic driving scheme with intelligent vehicle infrastructure cooperative systems as the core. As a kind of advanced artificial intelligence method, deep reinforcement learning (DRL) is applied to traffic signal control (TSC) to achieve the purpose of optimizing roadside traffic timing. In this paper, we introduce the background of TSC (i.e., main parameters, methods and simulation tools), and then summarize the representation of DRL model (i.e., state, action and reward) and the application of DRL in TSC. The research scenarios of TSC are divided into single-agent and multi-agent. Finally, according to existing works in this field, the problems to be solved are put forward and the paper is summarized.