LMI-Screened Attention-DRL for Heterogeneous Platooning via RSU Coordination
IEEE Internet of Things Journal · 2026-11 · 13 (22) · 54022-54040
DOI: 10.1109/jiot.2026.3729792 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Internet of Things Journal, 2026-11. Verfasst von Zekai Lv und 1 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Gefördert von: Guangdong Basic and Applied Basic Research Foundation.
Abstract (Originalsprache)
In mixed traffic environments, heterogeneous vehicle platoons composed of connected and automated vehicles (CAVs) and human-driven vehicles (HDVs) face major challenges in handling HDV behavioral uncertainty and scaling control to large platoons under diverse traffic conditions. This article proposes a reference-speed-informed global-and-local attention-based deep reinforcement learning (GLA-DRL) framework for large-scale heterogeneous platoon control within an infrastructure-centralized cooperative architecture. First, a hierarchical control architecture is developed to integrate data-driven learning with linear matrix inequality (LMI)-based gain admissibility screening, where an upper level GLA-DRL agent generates adaptive consensus gains and a lower level robust consensus controller executes the validated gains while accounting for actuator saturation and time-varying communication delays. The LMI-based admissibility verifier screens candidate gains for fixed-gain closed-loop stability before execution. Second, a global-and-local attention (GLA) mechanism is designed to process heterogeneous platoon observations, in which local attention adaptively fuses absolute vehicle states and relative tracking errors, while global attention captures platoon-wide intervehicle dependencies. The attention and LMI components are integrated within the gain-level learning, gain-hold admissibility screening, and roadside unit (RSU) traffic-state guidance pipeline. Third, a traffic-state-informed reference speed strategy incorporates an aggregate traffic-state cue, enabling speed adaptation under restrictive traffic conditions through a continuous saturated speed mapping. Furthermore, a modular decomposition strategy evaluates larger heterogeneous platoons by partitioning the traffic stream into repeated RSU-managed four-follower subplatoons with standardized interfaces. Comprehensive simulations using real-world naturalistic driving datasets indicate that the proposed framework improves tracking accuracy while maintaining comfort and collision-free operation in the evaluated scenarios.
Förderung
- Guangdong Basic and Applied Basic Research Foundation
Zitieren
APA
Lv, Z., & Chen, J. (2026). LMI-Screened Attention-DRL for Heterogeneous Platooning via RSU Coordination. IEEE Internet of Things Journal, 13(22), 54022-54040. https://doi.org/10.1109/jiot.2026.3729792
Vancouver
Lv Z, Chen J. LMI-Screened Attention-DRL for Heterogeneous Platooning via RSU Coordination. IEEE Internet of Things Journal. 2026;13(22):54022-54040. doi:10.1109/jiot.2026.3729792
BibTeX
@article{Lv2026,
author = {Lv, Zekai and Chen, Jianzhong},
title = {LMI-Screened Attention-DRL for Heterogeneous Platooning via RSU Coordination},
journal = {IEEE Internet of Things Journal},
year = {2026},
volume = {13},
number = {22},
pages = {54022--54040},
doi = {10.1109/jiot.2026.3729792}
}
EUVIMED