Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CoLLMLight: Cooperative Large Language Model Agents for Network-Wide Traffic Signal Control

Zirui Yuan, Siqi Lai, Hao Liu

TL;DR

CoLLMLight addresses the challenge of network-wide traffic signal control by enabling cooperative LLM agents that reason over a structured spatiotemporal graph of intersections. It introduces a complexity-aware reasoning mechanism to adapt inference depth to current congestion and a simulation-driven fine-tuning pipeline that builds task-specific reasoning chains and refines policies using environmental feedback. The framework demonstrates strong generalization across synthetic and seven real-world datasets, outperforming traditional methods, RL-based controllers, and generic LLM baselines, particularly in highly coordinated networks like New York. Overall, CoLLMLight offers a scalable, efficient approach to cross-intersection coordination with concrete mechanisms for spatiotemporal reasoning, adaptive computation, and continual improvement through simulation and feedback.

Abstract

Traffic Signal Control (TSC) plays a critical role in urban traffic management by optimizing traffic flow and mitigating congestion. While Large Language Models (LLMs) have recently emerged as promising tools for TSC due to their exceptional problem-solving and generalization capabilities, existing approaches fail to address the essential need for inter-agent coordination, limiting their effectiveness in achieving network-wide optimization. To bridge this gap, we propose CoLLMLight, a cooperative LLM agent framework for TSC. Specifically, we first construct a structured spatiotemporal graph to capture real-time traffic dynamics and spatial relationships among neighboring intersections, enabling the LLM to reason about complex traffic interactions. Moreover, we introduce a complexity-aware reasoning mechanism that dynamically adapts reasoning depth based on real-time traffic conditions, ensuring optimal computational efficiency without sacrificing decision quality. Besides, we propose a fine-tuning strategy that leverages iterative simulation-driven data collection and environmental feedback to build a lightweight LLM tailored for cooperative TSC. Extensive experiments on both synthetic and real-world datasets demonstrate that CoLLMLight outperforms state-of-the-art methods in diverse traffic scenarios, showcasing its effectiveness, scalability, and robustness.

CoLLMLight: Cooperative Large Language Model Agents for Network-Wide Traffic Signal Control

TL;DR

CoLLMLight addresses the challenge of network-wide traffic signal control by enabling cooperative LLM agents that reason over a structured spatiotemporal graph of intersections. It introduces a complexity-aware reasoning mechanism to adapt inference depth to current congestion and a simulation-driven fine-tuning pipeline that builds task-specific reasoning chains and refines policies using environmental feedback. The framework demonstrates strong generalization across synthetic and seven real-world datasets, outperforming traditional methods, RL-based controllers, and generic LLM baselines, particularly in highly coordinated networks like New York. Overall, CoLLMLight offers a scalable, efficient approach to cross-intersection coordination with concrete mechanisms for spatiotemporal reasoning, adaptive computation, and continual improvement through simulation and feedback.

Abstract

Traffic Signal Control (TSC) plays a critical role in urban traffic management by optimizing traffic flow and mitigating congestion. While Large Language Models (LLMs) have recently emerged as promising tools for TSC due to their exceptional problem-solving and generalization capabilities, existing approaches fail to address the essential need for inter-agent coordination, limiting their effectiveness in achieving network-wide optimization. To bridge this gap, we propose CoLLMLight, a cooperative LLM agent framework for TSC. Specifically, we first construct a structured spatiotemporal graph to capture real-time traffic dynamics and spatial relationships among neighboring intersections, enabling the LLM to reason about complex traffic interactions. Moreover, we introduce a complexity-aware reasoning mechanism that dynamically adapts reasoning depth based on real-time traffic conditions, ensuring optimal computational efficiency without sacrificing decision quality. Besides, we propose a fine-tuning strategy that leverages iterative simulation-driven data collection and environmental feedback to build a lightweight LLM tailored for cooperative TSC. Extensive experiments on both synthetic and real-world datasets demonstrate that CoLLMLight outperforms state-of-the-art methods in diverse traffic scenarios, showcasing its effectiveness, scalability, and robustness.

Paper Structure

This paper contains 34 sections, 13 equations, 12 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Cooperative LLM Agent Framework
  4. Spatiotemporal-Aware Cooperative Decision-Making
  5. Observation Collection
  6. Spatiotemporal Modeling
  7. Cooperative Decision-Making
  8. Complexity-Aware Reasoning
  9. Simulation-Driven Fine-Tuning
  10. Reasoning Chain Optimization
  11. Refinement with Environment Feedback
  12. Experiments
  13. Experimental Setup
  14. Datasets
  15. Environment Settings
  16. ...and 19 more sections

Figures (12)

  • Figure 1: (a): Both RL-based and LLM-based methods can benefit from cross-intersection collaboration. (b): Congestion levels vary significantly across different locations.
  • Figure 2: The overview of CoLLMLight framework.
  • Figure 3: An illustration of traffic scenarios with varying coordination complexity. Critical lanes are circled in red. (a) $n_c=0$ (No cooperation): Signals can determined locally. (b) $n_c=1$ (Simple cooperation): The agent needs to consider the interaction with one critical lane. (c) $n_c>1$ (Complex Cooperation): The agent needs to consider interactions with multiple critical lanes.
  • Figure 4: Comparative Performance of Learning-based Methods at Syn-Train
  • Figure 5: Performance and efficiency of various variants across four datasets. Methods positioned towards the lower left ($\swarrow$) exhibit superior performance ($\downarrow$) while incurring lower time costs ($\leftarrow$).
  • ...and 7 more figures

Theorems & Definitions (3)

  • Definition 1: Road Network
  • Definition 2: Traffic Signals
  • Definition 3: Intersection Connectivity Index (ICI)