MULTI-AGENT REINFORCEMENT LEARNING FOR INTELLIGENT TRAFFIC SIGNAL CONTROL SYSTEM

Abstract

Traffic congestion is a major issue in cities. It causes delays and inefficiencies. This paper introduces a Multi-Agent Reinforcement Learning (MARL) method for adapting traffic signal control. Each intersection works as an intelligent agent trained with the Q-learning algorithm. This training helps optimize when to switch signals through simulations. The model adjusts traffic lights based on queue lengths to reduce waiting time and congestion. A Streamlit interface shows training progress and system performance. Experimental simulations reveal that the MARL-based system outperforms traditional fixed-timing control. It provides a scalable basis for managing traffic in smart cities.