Scalable Adaptive Traffic Light Control Over a Traffic Network Including Turns, Transit Delays, and Blocking

Yingqing Chen, Christos G. Cassandras

Abstract

Coordinating traffic signals across large urban networks is a significant challenge due to the high dimensionality of state spaces and the stochastic nature of traffic flow. This paper introduces a scalable, data-driven framework using Infinitesimal Perturbation Analysis (IPA) to adaptively control traffic lights. By modeling the network as a Stochastic Hybrid System (SHS), the approach accounts for complex real-world factors such as turning movements, transit delays between intersections, and downstream blocking. Experimental results using the SUMO simulator indicate that the proposed controller significantly reduces average waiting times (up to 46%) and scales linearly with network size, offering a robust alternative to traditional static and reinforcement learning-based methods.

Introduction

Managing urban traffic congestion requires sophisticated coordination of signalized intersections. Traditional model-based optimization often fails to capture non-linear stochastic behaviors, while reinforcement learning approaches frequently suffer from poor scalability and high data requirements. This research addresses four critical hurdles in traffic network control: flow propagation between intersections, computational scalability, transit delays, and the "blocking" effect where downstream congestion halts upstream flow.

Flow propagation and turning movements between adjacent intersections.

Methodology

The proposed framework models the traffic network as a Stochastic Hybrid System (SHS), which integrates discrete signal switching with continuous vehicle flow dynamics. The core of the control logic is Infinitesimal Perturbation Analysis (IPA), a gradient estimation technique that allows the controller to optimize parameters like green light duration based on observed real-time events.

Event-Driven Control

Unlike time-driven models that update the state at fixed intervals, this IPA-based approach is event-driven. It reacts to specific triggers such as a queue becoming empty or a "flow burst" arriving from an upstream intersection. This significantly reduces computational overhead, as the system only performs calculations when relevant changes occur.

Queue length dynamics and event-driven transitions.

Handling Blocking and Transit Delays

A key innovation is the explicit modeling of "flow bursts" — the surge of vehicles released when a light turns green — and the associated travel time to the next intersection. Furthermore, the controller incorporates blocking constraints into its gradient estimations, prioritizing the clearance of downstream bottlenecks to prevent network-wide gridlock.

Experimental Results

The framework was evaluated using the SUMO (Simulation of Urban MObility) package on a $2 \times 3$ grid of intersections.

  • Performance: Compared to the standard Webster’s Method, the adaptive IPA controller reduced average waiting times by 0.75% to 46% across various traffic demand scenarios.
  • Scalability: Computational tests confirmed that the processing time increases linearly with the number of intersections. This linear scaling makes the method feasible for large-scale city deployments where traditional algorithms often experience exponential complexity growth.
  • Robustness: The system demonstrated an ability to naturally induce "green waves" (synchronized signal timing) and adapt quickly to sudden spikes in traffic demand without requiring a centralized master plan or prior knowledge of arrival distributions.

Conclusion and Future Work

The IPA-based adaptive control framework provides a scalable and model-agnostic solution for urban traffic management. It effectively mitigates congestion by responding to real-time events and accounting for network-level dependencies like blocking. Future research will focus on extending the model to include heterogeneous traffic (pedestrians and cyclists) and automating the optimization of algorithm step-sizes for even faster convergence in highly dynamic environments.

Summary updated

BibTeX
@article{chen2025scalable,
  author = {Yingqing Chen and Christos G. Cassandras},
  title = {{Scalable Adaptive Traffic Light Control Over a Traffic Network Including Turns, Transit Delays, and Blocking}},
  journal = {Discret. Event Dyn. Syst.},
  year = {2025},
  doi = {10.1007/s10626-025-00416-7},
}
Figure