Deep Reinforcement Learning based Traffic Signal optimization for Multiple Intersections in ITS

Abstract

The number of vehicles is drastically increasing worldwide, especially in large cities. Thus there is a need to model and enhance the traffic management to help meet this rising requirement. The primary purpose of traffic management is to reduce traffic congestion by optimizing traffic signal, which is currently one of the main concerns. Reinforcement Learning (RL) approaches in Intelligent Transportation System (ITS) are infeasible for traffic management of large road networks. However, Deep Reinforcement Learning (DRL) is capable of handling this enlarged problem. In order to manage the traffic flow of a large road network, there is a strong need for coordination between traffic signals of the intersections, enabling vehicles to pass through intersections more easily. In this paper, a single DRL agent manages the traffic signal of multiple intersections using policy gradient algorithm. In particular, the agent is trained with spatio-temporal data of the environment that allows it to perform action in different deep neural network models. The simulation experiment is studied in terms of three different simulation metrics. The proposed system outperforms while comparing it with the baseline i.e. fixed signal duration systems.