Abstract
Traffic congestion remains a significant source of economic inefficiency and environmental degradation. While modern schedule-driven traffic control systems have improved upon traditional cycle-based methods, they often rely on simplified models that group lanes together, sacrificing "expressiveness" for computational feasibility. This paper introduces Expressive Real-Time Intersection Scheduling (ERIS), a decentralized framework that leverages high-fidelity, lane-by-lane modeling. By utilizing an A* search algorithm with a novel additive heuristic, ERIS optimizes signal timing every second. Simulation results demonstrate that ERIS reduces vehicle delay by up to 20% compared to current state-of-the-art systems like SURTRAC.
Introduction
The primary cause of urban traffic congestion is often inefficient signal timing. Traditional systems typically use fixed cycles or update based on historical averages, failing to adapt to the high variability of real-time traffic. Newer "schedule-driven" approaches, such as SURTRAC, move toward real-time adaptation but often simplify the intersection model to maintain computational speed. This paper argues that such simplifications lead to sub-optimal decisions. ERIS is designed to maintain a highly accurate model of individual lanes while remaining fast enough for second-by-second decision-making.
Methodology
ERIS operates as a decentralized system where each intersection optimizes its own schedule while coordinating with neighbors through shared data. The system consists of two primary components:
- The Executor (Prediction): Utilizing vehicle-to-infrastructure (V2I) data—including position, speed, and heading—the system clusters vehicles for every individual lane. It also incorporates predicted arrival data from adjacent intersections.
- The Scheduling Module (Optimization): ERIS treats intersection control as a scheduling problem solved via A* search. Unlike previous models that group lanes, ERIS maintains separate clusters for every lane, providing the "expressiveness" necessary for precise timing.
To ensure real-time performance despite this increased complexity, the system employs an admissible heuristic that decomposes the scheduling problem into additive sub-problems. This allows for optimal schedule discovery within the one-second operational window.
Evaluation and Results
The performance of ERIS was evaluated using an open-source traffic simulator. The system was compared against SURTRAC and other real-time strategies. Key findings include:
- Delay Reduction: ERIS achieved up to a 20% reduction in vehicle delay compared to SURTRAC.
- Computational Efficiency: The A* search approach successfully met real-time deployment requirements, running every second despite the high-fidelity model.
Discussion and Future Work
The core insight of ERIS is that "expressiveness is efficiency." Modeling traffic at the lane level provides critical information that intersection-level models miss. Furthermore, decentralized coordination allows intersections to proactively schedule for incoming traffic waves by sharing predicted outflows.
However, several challenges remain. Current limitations include the assumption of protected movements and high communication overhead. Future work will focus on incorporating unprotected turns, reducing data requirements, and validating the system through a 50-light field study in Pittsburgh, Pennsylvania. Additionally, the framework will be expanded to include multi-modal traffic, such as pedestrians and transit.
Conclusion
ERIS demonstrates that high-fidelity modeling and efficient search algorithms can significantly improve urban traffic flow. By moving beyond simplified intersection models, ERIS sets a new benchmark for real-time, decentralized traffic control.
Summary updated