Reinforcement Learning-Based Multi-Agent Systems for Intelligent Urban Traffic Signal Optimization

Abstract

Modern urban centers face growing challenges in managing traffic congestion, environmental pollution, and vehicular delays. This paper proposes a decentralized reinforcement learning (RL) approach using multiple intelligent agents to optimize traffic signals dynamically in real-time. The proposed system utilizes multi-agent reinforcement learning (MARL) and fuzzy logic rules to handle uncertainties in traffic conditions and provide scalable solutions for smart cities. Key contributions include the development of a real-time adaptable MARL framework and a fuzzy decision layer that enhances coordination between agents. Simulations show that the system outperforms traditional traffic signal algorithms in terms of reduced vehicle wait times, queue lengths, and fuel consumption. This study reinforces the potential of AI-driven decentralized control in transforming urban mobility and sustainable infrastructure.

KEYWORDS: Reinforcement Learning, Multi-Agent Systems, Smart Traffic Management, Fuzzy Logic, Intelligent Transportation, Traffic Signal Control, Decentralized AI, Urban Congestion, Real-Time Optimization, Smart Cities