Design of Deep Reinforcement Learning Approach for Traffic Signal Control at Three-way Crossroads

This repository presents the following article in Python:

Thanh Nguyen Canh, Anh Pham Tuan, Xiem HoangVan, "Design of Deep Reinforcement Learning Approach for Traffic Signal Control at Three-way Crossroads," Under Review. [Research Square] [Citation]

Citation

@article{canh2023design,
  title={Design of Deep Reinforcement Learning Approach for Traffic Signal Control at Three-way Crossroads},
  author={Canh, Thanh Nguyen and Tuan, Anh Pham and HoangVan, Xiem},
  year={2023}
}

• Overview
• Overview
• 2. Installation
• 3. Design
• 4. Results
• 5. Branch
• 6. Reference

Oveview

1. Requirements

• Unity: 2020.3.33f1
• ML-Agents: 0.28.0
• C#: 8.0
• Python: 3.9.6
• PyTorch: 1.12.1

2. Installation

git clone https://github.com/thanhnguyencanh/DRL4Three-wayTSC
cd DRL4Three-wayTSC
pip3 install -r requirement.txt

3. Design

• Environment for testing Reinforcement Learning algorithms.

• Properties:

  Property	Detail
  Action Space	Discrete(8)
  Observation Shape	(12,)

  Observation space includes:

  • Waiting time of vehicles.
  • Speed of first car on each lane.
  • Timer for green lights.
  • Current state of traffic lights.

• Can be used for testing external algorithms or ML-Agents built-in algorithms.

• Includes built-in debug logs when using executable.

4. Results

Scenario 1	Scenario 2

5. Branch

There are 2 branches for 2 versions of the environment:

• Reinforcement Learning: https://github.com/Spencer266/T-Juntion/tree/multi
• Fixed-time: https://github.com/Spencer266/T-Juntion/tree/linear

Note: The second one is used for comparing Reinforcement Learning and Fixed-time method. However, it can acts as a standalone own project.

6. Reference

• ML-Agents: https://github.com/Unity-Technologies/ml-agents