Fuzzy Inference Enabled Deep Reinforcement Learning-Based Traffic Light Control for Intelligent Transportation System

Intelligent Transportation System (ITS) has been emerged an important component and widely adopted for the smart city as it overcomes the limitations of the traditional transportation system. Existing fixed traffic light control systems split the traffic light signal into fixed duration and run in an inefficient way, therefore, it suffers from many weaknesses such as long waiting time, waste of fuel and increase in carbon emission. To tackle these issues and increase efficiency of the traffic light control system, in this work, a Dynamic and Intelligent Traffic Light Control System (DITLCS) is proposed which takes real-time traffic information as the input and dynamically adjusts the traffic light duration. Further, the proposed DITLCS runs in three modes namely Fair Mode (FM), Priority Mode (PM) and Emergency Mode (EM) where all the vehicles are considered with equal priority, vehicles of different categories are given different level of priority and emergency vehicles are given at most priority respectively. Furthermore, a deep reinforcement learning model is also proposed to switch the traffic lights in different phases (Red, Green and Yellow), and fuzzy inference system selects one mode among three modes i.e., FM, PM and EM according to the traffic information. We have evaluated DITLCS via realistic simulation on Gwalior city map of India using an open-source simulator i.e., Simulation of Urban MObility (SUMO). The simulation results prove the efficiency of DITLCS in comparison to other state of the art algorithms on various performance parameters.