A Survey on Reinforcement Learning Models and Algorithms for Traffic Signal Control

Abstract

Traffic congestion has become a vexing and complex issue in many urban areas. Of particular interest are the intersections where traffic bottlenecks are known to occur despite being traditionally signalized. Reinforcement learning (RL), which is an artificial intelligence approach, has been adopted in traffic signal control for monitoring and ameliorating traffic congestion. RL enables autonomous decision makers (e.g., traffic signal controllers) to observe, learn, and select the optimal action (e.g., determining the appropriate traffic phase and its timing) to manage traffic such that system performance is improved. This article reviews various RL models and algorithms applied to traffic signal control in the aspects of the representations of the RL model (i.e., state, action, and reward), performance measures, and complexity to establish a foundation for further investigation in this research field. Open issues are presented toward the end of this article to discover new research areas with the objective to spark new interest in this research field.