Multi-Agent Optimizing Traffic Light Signals Using Deep Reinforcement Learning

In recent years, rapid urbanization has led to increased traffic congestion, rendering traditional traffic light control methods ineffective. Deep Reinforcement Learning (DRL) has emerged as a promising approach to sequential decision-making, offering adaptive and efficient solutions for traffic management. This paper aims to develop an optimal traffic light planning strategy that integrates seamlessly with urban transportation systems, including trams and Bus Rapid Transit Systems (BRTS). The study explores three DRL-based approaches: Single-Agent Deep Reinforcement Learning (SADRL), Multi-Agent Deep Reinforcement Learning (MADRL) with fixed traffic lights, and an actuated control approach. System for Managing Actuated and Real-Time Traffic, referred to as SMART, dynamically adjusts traffic signals based on real-time conditions to enhance traffic flow efficiency. The proposed methods are evaluated and compared against the Webster method, Simulation of Urban Mobility (SUMO)-based control, and a genetic algorithm-based multi-objective traffic light optimization method (MamoTLO). The results demonstrate that DRL-based solutions improve traffic flow and reduce congestion.