Deep Reinforcement Learning with Vehicle Heterogeneity Based Traffic Light Control for Intelligent Transportation System

Abstract

Existing vehicle traffic light signal control system suffers from many issues such as long waiting time, waste of fuel consumption and increase in carbon emission. This causes the lot of stress to the drivers and delay in arrivals of emergency vehicles and other vehicles including priority vehicles. The above mentioned issues can be limited to an extend through the replacement of the existing inefficient (fixed) traffic control system with an efficient dynamic traffic light control system. The recent advancements in sensor technology enable us to capture the real-time traffic information which can be used to dynamically control the traffic light duration. Existing works either split the traffic signal into fixed duration or just extract very little real-time traffic data which might be inefficient in most of the traffic scenarios. Hence, in this paper, we proposed a deep reinforcement learning model with vehicle heterogeneity (DLVH) for traffic control system. The DLVH model can dynamically change the traffic signal phase and duration based on the traffic information. We have evaluated DLVH via realistic simulation on Gwalior city map of India using an open-source simulator i.e., Simulation of Urban Mobility (SUMO) which proves the effectiveness of the model.