FutureLight: An Efficient Future Traffic Data-Driven Reinforcement Learning Framework for Traffic Signal Controls

Reinforcement learning (RL) has emerged as a promising approach for Traffic Signal Control (TSC), enabling intersections to learn control policies under dynamic traffic conditions. However, existing RL-based methods only rely on current traffic states and expected rewards, without leveraging future traffic conditions for better decision making. To address this limitation, we propose FutureLight , the first RL-based TSC framework that leverages future route data. Specifically, we design FutureLight-RouteSys , a macroscopic, signal-aware, and lane-level simulator for efficient and accurate estimation of future traffic conditions. The predicted future information is then encoded by FutureLight-Encoder and integrated into FutureLight-DQN through state augmentation, reward shaping, and hybrid value estimation, which combines simulated near-future rewards with bootstrapped returns. Finally, we introduce pruning techniques to avoid redundant computation and improve training efficiency. Experimental results show that FutureLight consistently improves TSC performance while achieving more than 30× faster training.