Urban traffic management and adaptive traffic signal control are fundamental to improving mobility, reducing congestion, and enhancing the efficiency of modern smart cities. Coordinating multiple signalized intersections under dynamic and uncertain traffic conditions remains a significant challenge, particularly when scalability and adaptability are required. Reinforcement Learning (RL) has shown strong potential for multi-agent task allocation in such environments; however, existing multi-agent RL approaches for multi-intersection traffic signal control often struggle with unstable convergence, limited scalability, and inefficient coordination as network size and traffic variability increase. To address these limitations, this paper proposes a Dueling Reinforcement-based Adaptive Graph Optimization Network (DR-AGON) for scalable and adaptive multi-agent task allocation in urban traffic signal control systems. The proposed framework employs a Dueling Double Deep Q-Network (D3QN) architecture to decouple state-value and advantage estimations, enabling more stable learning and robust decision-making under unpredictable traffic flows. A centralized training and decentralized execution paradigm is integrated with Graph-based Attention Networks (GAT) to explicitly model inter- and intra-interaction dynamics and dynamically adjust the importance of communication. In addition, a meta-learning mechanism allows agents to rapidly adapt their policies to non-stationary traffic patterns and evolving network-level objectives. Within the traffic signal control context, DR-AGON facilitates cooperative signal-phase allocation among intersections, thereby improving global traffic flow efficiency while maintaining local responsiveness. Experimental results in the SUMO traffic simulator demonstrate improvements of 22% in average vehicle delay, 19% in queue length reduction, 27% in throughput, and 15% in travel time compared with baseline traffic control methodsThese findings confirm that the offered DR-AGON framework is an efficient simulation-based approach using which big cities can optimize traffic signals, confirmed by a complex of experiments in the SUMO simulator environment.
Alanazi et al. (Mon,) studied this question.