Scalable Traffic Signal Control with Shared Policy Framework

Learning-based Traffic Signal Control (TSC) achieves satisfactory performance in small networks, but its effectiveness often deteriorates in larger networks under dynamic traffic patterns and intersection heterogeneity. In this work, we propose SLight, a policy-aware grouped MARL-TSC framework that enables scalability and efficiency balance under dynamic and heterogeneous traffic conditions. SLight captures policy-influenced traffic patterns with a policy-aware traffic pattern encoder, learns explicit group-level shared control principles from state–action trajectories, and matches each intersection’s traffic pattern embedding to principle prototypes flexibly through a compatibility-based adaptive assignment module. Experiments on real-world and synthetic networks demonstrate that SLight sustains performance gains as scale increases and outperforms existing rule-based, reinforcement learning, and grouping-based baselines. Code is available at \url{https://github.com/MaHaoLun/Slight-code.git}

City traffic lights are difficult to coordinate because each intersection faces different traffic, and conditions can change quickly during the day. A timing strategy that works for a busy avenue may be a poor fit for a small side street, while training a separate AI controller for every light can become expensive and unstable in large road networks. We introduce SLight, an RL framework that lets traffic lights share a small set of reusable control strategies while still choosing the one that best matches each intersection at the moment. Instead of grouping intersections only because their current traffic looks similar, SLight also considers how recent signal decisions affected the traffic. It then matches each intersection to a learned “control principle,” such as a strategy for clearing queues or improving flow. This makes sharing safer: similar-looking intersections are not forced to use the same behavior when they need different actions. In experiments on road networks from JiNan, HangZhou, and Manhattan, SLight reduced travel time and improved large-scale coordination. This could help future traffic-control systems become more adaptive, efficient, and scalable.