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

Thanh Nguyen Canh

Anh Tuan Pham

Xiem HoangVan

VNU University of Engineering and Technology, Vietnam
Public Transport, 2025.

[Paper]

[Code]

Traffic signal control (TSC) is an important and challenging real-world problem aimed at reducing travel time as well as conserving energy. Recent studies have made significant strides in applying intelligent meth- ods to TSC at four-way crossroads to address traffic signal scheduling problems. However, research on efficient TSC at three-way crossroads remains limited. Therefore, this paper introduces an efficient TSC solu- tion for a three-way crossroad environment (TW-TSC), utilizing a deep reinforcement learning approach called Soft Actor-Critic (TWSAC). Ini- tially, we develop a simulation environment for three-way intersections using the Unity framework, which includes various types of transporta- tion and two parallel lanes. To accurately model transportation dynamics at three-way crossroads, we carefully design agents that significantly influence transportation movement, including waiting times at traffic sig- nal, vehicle speeds, and the number of vehicles successfully passing. To optimize TW-TSC efficiency, we propose a novel reward function together with a design of the TWSAC algorithm. Experimental results indicate that our TWSAC method outperforms both fixed-time TSC methods and relevant reinforcement learning algorithms in terms of performance.

Paper

Thanh Nguyen Canh, Anh Tuan Pham, Xiem HoangVan

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

Public Transport 2025.

[pdf]

Overview

Overall TW-TSC architecture.

Three-way crossroads and traffic flow scenario.

The structured architecture of the TWSAC model.

Experiments

Cumulative reward over 7,000 episodes when training: (a) DDPG, (b) PPO, (c) TD3, (d) TWSAC.

Comparing the number of crossing vehicles.

Code

[github]

Citation

1. Canh T. N., Pham A-T, HoangVan X. Design of Deep Reinforcement Learning Approach for Traffic Signal Control at Three-way Crossroads. Public Transort, 2025.

@inproceedings{canh2025design, 
                                  

                                    author    = {Canh, Thanh Nguyen and Pham, Anh Tuan and HoangVan, Xiem}, 
                                  

                                    title     = {{Design of Deep Reinforcement Learning Approach for Traffic Signal Control at Three-way Crossroads}}, 
                                  

                                    journal   = {Public Transort}, 
                                  

                                    year      = {2025}, 
                                  

                                }

Acknowledgements

Thanh Nguyen Canh was funded by the Master, PhD Scholarship Programmer of Vingroup Innovation Foundation (VINIF), code VINIF.2023.ThS.120
This webpage template was borrowed from https://akanazawa.github.io/cmr/.