A Fully Data-Driven Approach for Realistic Traffic Signal Control Using Offline Reinforcement Learning

The optimization of traffic signal control (TSC) is critical for an efficient transportation system. In recent years, reinforcement learning (RL) techniques have emerged as a popular approach for TSC and show promising results for highly adaptive control. However, existing RL-based methods suffer fr...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Data science for Transportation Jg. 7; H. 3; S. 25
Hauptverfasser: Li, Jianxiong, Lin, Shichao, Shi, Tianyu, Tian, Chujie, Mei, Yu, Song, Jian, Zhan, Xianyuan, Li, Ruimin
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Singapore Springer Nature Singapore 01.12.2025
Springer Nature B.V
Schlagworte:
Adaptive control
Computational Intelligence
Data Mining and Knowledge Discovery
Datasets
Engineering
Flow theory
Machine learning
Neural networks
Optimization
Regularization methods
Simulation
Simulators
Traffic control
Traffic flow
Traffic information
Traffic intersections
Traffic signals
Transportation systems
Transportation Technology and Traffic Engineering
Real-world traffic data
Shockwave theory
Offline reinforcement learning
Traffic signal control
ISSN:2948-135X, 2948-1368
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The optimization of traffic signal control (TSC) is critical for an efficient transportation system. In recent years, reinforcement learning (RL) techniques have emerged as a popular approach for TSC and show promising results for highly adaptive control. However, existing RL-based methods suffer from serious real-world transferability issues and hardly have any successful deployments. The reasons for such failures are mostly due to the reliance on over-idealized traffic simulators for policy optimization, as well as using unrealistic fine-grained state observations and reward signals that are not directly obtainable from real-world sensors. In this paper, we propose a fully data-driven and simulator-free framework for realistic Traffic Signal Control. Specifically, we combine well-established traffic flow theory with machine learning to construct a reward inference model to infer the reward signals from coarse-grained traffic data. With the inferred rewards, we further propose a sample-efficient offline RL method to enable direct signal control policy learning from historical offline data sets of real-world intersections. To evaluate our approach, we collect historical traffic data from a real-world intersection, and develop a highly customized simulation environment that strictly follows real data characteristics. We demonstrate through extensive experiments that our approach achieves superior performance over conventional and offline RL baselines, and also enjoys much better real-world applicability.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2948-135X
2948-1368
DOI:10.1007/s42421-025-00139-z