Multi-Agent Reinforcement Learning for Smart City Automated Traffic Light Control

Traffic congestion has emerged as a significant issue for road users, particularly in densely populated urban areas. Many individuals endure extended travel times while moving from one location to another, a direct result of the excessive traffic volume caused by ineffective traffic light control. Consequently, we propose the implementation of an automated traffic light control (TLC) system using Artificial Intelligence (AI) principles to manage two interconnected intersections within this case study. This framework is designed based on Multi-Agent Reinforcement Learning (MARL) with the Q-Learning (QL) algorithm, enabling the TLC system to assign appropriate green signal timing to each junction at both intersections. The primary objective is to alleviate traffic congestion at these intersections by minimizing vehicle wait times at red signals. The traffic model we have developed is implemented and simulated using the MATLAB simulator. We evaluate the system's performance using three distinct configurations with varying learning rates and record the results. The outcomes of our proposed model, presented using the designed methodology, effectively reduce weighted wait times and vehicle queue lengths through the QL algorithm.

Paper

Full text

PDF

Multi-Agent Reinforcement Learning for Smart City Automated Traffic Light Control

Semantic Scholar · Engineering · 2023

Abstract

Traffic congestion has emerged as a significant issue for road users, particularly in densely populated urban areas. Many individuals endure extended travel times while moving from one location to another, a direct result of the excessive traffic volume caused by ineffective traffic light control. Consequently, we propose the implementation of an automated traffic light control (TLC) system using Artificial Intelligence (AI) principles to manage two interconnected intersections within this case study. This framework is designed based on Multi-Agent Reinforcement Learning (MARL) with the Q-Learning (QL) algorithm, enabling the TLC system to assign appropriate green signal timing to each junction at both intersections. The primary objective is to alleviate traffic congestion at these intersections by minimizing vehicle wait times at red signals. The traffic model we have developed is implemented and simulated using the MATLAB simulator. We evaluate the system's performance using three distinct configurations with varying learning rates and record the results. The outcomes of our proposed model, presented using the designed methodology, effectively reduce weighted wait times and vehicle queue lengths through the QL algorithm.

Similar papers

© 2026 NYSGPT2525 LLC