Two-Stage Reinforcement Learning for Proactive Cooperative Perception in CAVs via Mutual Partner Selection and Joint Resource Optimization
IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 23017-23034
DOI: 10.1109/tmc.2026.3725763 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Yixin Fan und 3 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Cooperative perception demonstrates substantial potential for enhancing the environmental awareness of connected and autonomous vehicles. However, its practical deployment is significantly constrained by the highly dynamic nature of urban traffic and inherent onboard resource heterogeneity. These factors induce real-time fluctuations in network conditions, computational capacity, and perception demands, resulting in data redundancy, imbalanced resource allocation, and degraded perception performance. To address these challenges, this paper proposes a proactive cooperative perception framework capable of adaptively responding to dynamic environments and heterogeneous resource conditions. The proposed framework introduces a mutual partner selection mechanism to facilitate targeted and resource-efficient data sharing. This mechanism enables a two-stage optimization paradigm that first maximizes collective perception accuracy through optimal partner selection and subsequently minimizes end-to-end latency through joint resource optimization, including dynamic deep neural network (DNN) inference partitioning and communication subchannel allocation over the selected cooperative links. To solve this problem, we develop a novel two-stage reinforcement learning (TSRL) algorithm. The TSRL algorithm consists of a global state proximal policy optimization module for intelligent partner matching and a multi-agent proximal policy optimization module with a transformer mechanism for adaptive DNN partitioning points selection and communication resource allocation. Experimental results demonstrate that the proposed approach effectively enhances perception accuracy and reduces latency in dynamic vehicular network scenarios.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Fan, Y., Peng, H., Su, Z., & Luan, T. H. (2026). Two-Stage Reinforcement Learning for Proactive Cooperative Perception in CAVs via Mutual Partner Selection and Joint Resource Optimization. IEEE Transactions on Mobile Computing, 25(12), 23017-23034. https://doi.org/10.1109/tmc.2026.3725763
Vancouver
Fan Y, Peng H, Su Z, Luan TH. Two-Stage Reinforcement Learning for Proactive Cooperative Perception in CAVs via Mutual Partner Selection and Joint Resource Optimization. IEEE Transactions on Mobile Computing. 2026;25(12):23017-23034. doi:10.1109/tmc.2026.3725763
BibTeX
@article{Fan2026,
author = {Fan, Yixin and Peng, Haixia and Su, Zhou and Luan, Tom H.},
title = {Two-Stage Reinforcement Learning for Proactive Cooperative Perception in CAVs via Mutual Partner Selection and Joint Resource Optimization},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {12},
pages = {23017--23034},
doi = {10.1109/tmc.2026.3725763}
}
EUVIMED