Conceptual Study of Reinforcement Learning-Based Intelligent Traffic Management System for Congestion Mitigation

Abstract

Urban traffic congestion is a complex problem that impacts mobility efficiency, energy consumption, increased exhaust emissions, and the quality of life of the community. The development of a Reinforcement Learning (RL)-based Intelligent Traffic Management System offers an adaptive approach to traffic signal regulation that is able to respond to the dynamics of vehicle flow in real-time. This study aims to formulate a conceptual framework for a simple, integrated, and potentially implementable RL-based traffic management system at the local government level, without involving primary data collection or empirical testing. The method used is a conceptual study through analysis and synthesis of the latest scientific literature discussing traffic signal control, multi-agent reinforcement learning, and intelligent transportation systems. The results of the study are a conceptual framework that integrates sensor-based and Internet of Things (IoT) traffic data acquisition, a traffic condition prediction module, a multi-agent RL decision-making mechanism, coordination between intersections, and performance monitoring based on key indicators. Based on the literature review, the RL approach shows potential in improving traffic flow, reducing waiting times and queue lengths, and reducing emissions, especially in dynamic traffic conditions. This conceptual framework is expected to be an initial reference in the development of simulations, limited trials, and further research to support the design of an effective and sustainable adaptive traffic management system.