OverFlowLight Prevents Gridlock, Optimizes Traffic Signals in Real-Time

Urban traffic congestion frequently leads to queue overflow, where vehicle lines exceed intersection capacity, blocking upstream traffic and causing cascading gridlocks. Current traffic signal control (TSC) algorithms, often focused on maximizing throughput, struggle with overflow during peak hours, worsening congestion and safety. Researchers introduce OverFlowLight, a real-time framework designed to proactively resolve these overflow issues and enhance overall TSC performance. OverFlowLight employs multi-modal sensors, including cameras and radars, to accurately detect overflow in real-time. Upon detection, it dynamically generates and inserts specific "overflow phases" into the signal cycle to clear the blocking queues. This system uses a hybrid control design, combining rapid rule-based overflow intervention with backend controllers like reinforcement learning for long-term efficiency. Extensive real-world deployments across 43 intersections in three major cities demonstrated that OverFlowLight seamlessly integrates with existing RL-based TSC agents. Empirical results show a 60.4% reduction in overflow incidents and an 18.2% increase in network throughput compared to deployed baselines, significantly reducing the need for manual intervention.