Abstract:

Sustainable mobility is a central goal in the evolution of smart cities, with optimized traffic flow playing a critical role in reducing congestion, emissions, and travel time. Reinforcement Learning (RL), with its ability to learn optimal actions through interaction with dynamic environments, offers a promising approach for intelligent traffic management systems. Traditional traffic control methods, including fixed-time signals and rule-based algorithms, often fail to adapt effectively to real-time fluctuations in traffic patterns, leading to inefficiencies and increased environmental impact. These static strategies lack responsiveness and scalability, particularly in complex urban ecosystems. To address these limitations, we propose the Adaptive Reinforcement Learning Traffic Optimization Framework (ARLTOF), which integrates Deep Q-Networks (DQN), Multi-Agent Reinforcement Learning (MARL), and real-time traffic data analytics. ARLTOF dynamically learns and adjusts traffic signal timings, vehicle routing, and congestion mitigation strategies based on current and predictive traffic conditions. The proposed ARLTOF system is implemented using a simulated smart city environment, leveraging sensor inputs, vehicular data, and edge computing for rapid decision-making. Through continual learning and decentralized agent collaboration, the framework effectively reduces idle times, improves traffic fluidity, and enhances commuter experience. Experimental results demonstrate that ARLTOF significantly improves traffic efficiency, with up to a 35% reduction in average vehicle delay and a 28% decrease in carbon emissions compared to conventional systems. These findings underscore the potential of reinforcement learning as a transformative tool in developing eco-friendly and efficient transportation systems in smart cities.