Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Dual Large Language Models Architecture with Herald Guided Prompts for Parallel Fine Grained Traffic Signal Control

Qing Guo, Xinhang Li, Junyu Chen, Zheng Guo, Xiaocong Li, Lin Zhang, Lei Li

TL;DR

HeraldLight is proposed, a dual LLMs architecture enhanced by Herald guided prompts that outperforms state of the art baselines, achieving a 20.03% reduction in average travel time across all scenarios and a 10.74% reduction in average queue length on the Jinan and Hangzhou scenarios.

Abstract

Leveraging large language models (LLMs) in traffic signal control (TSC) improves optimization efficiency and interpretability compared to traditional reinforcement learning (RL) methods. However, existing LLM-based approaches are limited by fixed time signal durations and are prone to hallucination errors, while RL methods lack robustness in signal timing decisions and suffer from poor generalization. To address these challenges, this paper proposes HeraldLight, a dual LLMs architecture enhanced by Herald guided prompts. The Herald Module extracts contextual information and forecasts queue lengths for each traffic phase based on real-time conditions. The first LLM, LLM-Agent, uses these forecasts to make fine grained traffic signal control, while the second LLM, LLM-Critic, refines LLM-Agent's outputs, correcting errors and hallucinations. These refined outputs are used for score-based fine-tuning to improve accuracy and robustness. Simulation experiments using CityFlow on real world datasets covering 224 intersections in Jinan (12), Hangzhou (16), and New York (196) demonstrate that HeraldLight outperforms state of the art baselines, achieving a 20.03% reduction in average travel time across all scenarios and a 10.74% reduction in average queue length on the Jinan and Hangzhou scenarios. The source code is available on GitHub: https://github.com/BUPT-ANTlab/HeraldLight.

A Dual Large Language Models Architecture with Herald Guided Prompts for Parallel Fine Grained Traffic Signal Control

TL;DR

HeraldLight is proposed, a dual LLMs architecture enhanced by Herald guided prompts that outperforms state of the art baselines, achieving a 20.03% reduction in average travel time across all scenarios and a 10.74% reduction in average queue length on the Jinan and Hangzhou scenarios.

Abstract

Leveraging large language models (LLMs) in traffic signal control (TSC) improves optimization efficiency and interpretability compared to traditional reinforcement learning (RL) methods. However, existing LLM-based approaches are limited by fixed time signal durations and are prone to hallucination errors, while RL methods lack robustness in signal timing decisions and suffer from poor generalization. To address these challenges, this paper proposes HeraldLight, a dual LLMs architecture enhanced by Herald guided prompts. The Herald Module extracts contextual information and forecasts queue lengths for each traffic phase based on real-time conditions. The first LLM, LLM-Agent, uses these forecasts to make fine grained traffic signal control, while the second LLM, LLM-Critic, refines LLM-Agent's outputs, correcting errors and hallucinations. These refined outputs are used for score-based fine-tuning to improve accuracy and robustness. Simulation experiments using CityFlow on real world datasets covering 224 intersections in Jinan (12), Hangzhou (16), and New York (196) demonstrate that HeraldLight outperforms state of the art baselines, achieving a 20.03% reduction in average travel time across all scenarios and a 10.74% reduction in average queue length on the Jinan and Hangzhou scenarios. The source code is available on GitHub: https://github.com/BUPT-ANTlab/HeraldLight.

Paper Structure

This paper contains 29 sections, 8 equations, 8 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Intersection Modeling
  3. Lanes, Phases, and Duration
  4. Herald Module
  5. Herald Learns Scenario-Specific Knowledge
  6. Herald Module Assist Process for Duration Selection
  7. HeraldLight Framework
  8. Task Description ($d_{\text{task}}$)
  9. Signal Phase Selection
  10. Duration Determination
  11. Distributed Dual LLMs Agents
  12. Imitation Fine-tuning with LoRA
  13. Dual LLMs architecture
  14. Experiments and Results
  15. Datasets and Scenario Configuration
  16. ...and 14 more sections

Figures (8)

  • Figure 1: Overview of HeraldLight. Herald Module learns and updates; ChatGPT assisted with Herald Module curates imitation data; Llama-3.1-8B is LoRA-finetuned to instantiate the LLM-Agent; an LLM-Critic (ChatGPT) evaluates and revises decisions; scored corrections drive iterative, score-based fine-tuning.
  • Figure 2: Herald Module Assist for Duration Calculation
  • Figure 3: Score-Based Dual LLMs Correction
  • Figure 4: Visualizations of road networks in three study areas.
  • Figure 5: The performance in the large-scale road network
  • ...and 3 more figures