A Study on Deep Reinforcement Learning Based Traffic Signal Control for Mitigating Traffic Congestion

Abstract

Traffic Congestion (TC) is a crucial problem in many urban areas, leading to extreme delays in daily lives. Intelligent traffic light control is a solution that dynamically balances the traffic duration based on different road traffic flows. Deep reinforcement learning (Deep RL) is a prevalent method to adapt traffic signal control in active environments nowadays. Many researchers recently developed deep RL-based intelligent and adaptive traffic signal control systems based on different states (e.g., speed and the position of vehicles) in a particular environment. However, deep RL has not employed the control of the traffic signal by observing traffic flow in any busy road. In this paper, we proposed a deep Q-network (DQN) method based on different traffic flow information to control the traffic signal dynamically. The simulation results show improved results in terms of the average delays of the vehicles.