Traffic Signal Control Based on Reinforcement Learning

This repository provides the source code, experiments, and publications for a series of research projects focused on Reinforcement Learning (RL)-based Traffic Signal Control. The aim is to optimize signalized intersections in urban road networks using advanced deep reinforcement learning (DRL) algorithms to improve traffic efficiency, reduce vehicle delay, and minimize congestion.

It also includes the research papers published by our group, showcasing our contributions to this domain, from theoretical models to real-world deployment strategies.

---

📄 Overview

• 💡 Problem: How to efficiently control traffic signals in complex urban settings using RL
• 🤖 Approach: State-of-the-art DRL algorithms including DQN, Diffusion RL, and fuzzy decision-making
• 🏙️ Application: Simulated and real-world intersections across multiple cities
• 📊 Metrics: Delay, queue length, average travel time, robustness

---

📝 Publications

📚 A list of peer-reviewed and preprint papers by our team on RL-based traffic signal control.

1. Efficient Pressure: Improving efficiency for signalized intersections
   arXiv, 2021
   📄 Paper Link
   Keywords: pressure-based control, delay minimization, lightweight RL models

2. Expression might be enough: Representing pressure and demand for reinforcement learning based traffic signal control
   ICML 2022
   📄 Paper Link
   Keywords: representation learning, traffic demand modeling, GNN-RL

3. A General Hyperbolic Reinforcement Learning Paradigm and Method for Traffic Signal Control (Under Review)
   A novel RL formulation using hyperbolic embeddings for scalable and generalizable intersection control.
   Keywords: hyperbolic geometry, generalization, MARL

4. FuzzyLight: A Robust Two-Stage Fuzzy Approach for Traffic Signal Control That Works in Real Cities
   KDD 2025
   📄 Paper Link 📄Code Link

   Keywords: fuzzy logic, hierarchical control, real-world deployment

5. RobustLight: Improving Robustness via Diffusion Reinforcement Learning for Traffic Signal Control
   ICML 2025
   📄 Paper Link 📄Code Link

   Keywords: diffusion models, robustness, noisy observation handling

6. CFLight: Enhancing Safety with Traffic Signal Control through Counterfactual Learning
   KDD 2026
   📄 Paper Link 📄Code Link

   Keywords: traffic signal control, reinforcement learning, counterfactual learning