EUVIMEDEUVIMED entdecken
Crossref

Agentic AI for Low-Altitude Economic Networks via LoRA-Assisted Cloud–Edge Collaboration

Ziqiang Ye, Yulan Gao, Yue Xiao, Zhonghao Lyu, Ping Yang, Agata Manolova

IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 9474-9488

DOI: 10.1109/tccn.2026.3705845 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Ziqiang Ye und 5 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Gefördert von: National Key Research and Development Program of China, Natural Science Foundation of Sichuan Province, Bulgarian National Science Fund project.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Low-altitude economic networks (LAENets) underpin the emerging low-altitude economy, where uncrewed aerial vehicles (UAVs) support reliable communications under highly dynamic 3D propagation conditions. However, severe Doppler shifts, channel aging, beam misalignment, and interference fundamentally challenge conventional processing, especially under stringent onboard constraints. In this work, we investigate multi-task intelligence in LAENets from an agentic and embodied intelligence perspective, where a UAV operates as an autonomous agent that dynamically executes heterogeneous tasks on demand. To address this challenge, we propose a multi-task low-rank adaptation (LoRA)-assisted cloud-edge collaborative framework that decomposes agent decision-making into task-specific skill components over a shared foundation-model backbone. Specifically, a frozen and quantized large language model (LLM) is deployed at the ground base station as a universal cognitive reasoning module, while lightweight task-specific LoRA adapters, together with dedicated encoders and decoders, are hosted on UAV platforms and dynamically activated according to mission objectives, for avoiding negative transfer among heterogeneous tasks as well as eliminating unnecessary computation during single-task execution. Extensive simulations demonstrate that the proposed framework consistently outperforms conventional model-based methods, independently trained deep neural networks, and fully fine-tuned foundation models in channel prediction, beamforming, and signal detection, while achieving orders-of-magnitude reductions in UAV-side memory footprint and inference latency.

Förderung

  • National Key Research and Development Program of China
  • Natural Science Foundation of Sichuan Province
  • Bulgarian National Science Fund project
  • European Union’s Horizon Europe research and innovation programme under the Marie Skłodowska-Curie Actions Postdoctoral Fellowship
  • project Multi-RM-6GIoT-GAI

Zitieren

APA

Ye, Z., Gao, Y., Xiao, Y., Lyu, Z., Yang, P., & Manolova, A. (2026). Agentic AI for Low-Altitude Economic Networks via LoRA-Assisted Cloud–Edge Collaboration. IEEE Transactions on Cognitive Communications and Networking, 12, 9474-9488. https://doi.org/10.1109/tccn.2026.3705845

Vancouver

Ye Z, Gao Y, Xiao Y, Lyu Z, Yang P, Manolova A. Agentic AI for Low-Altitude Economic Networks via LoRA-Assisted Cloud–Edge Collaboration. IEEE Transactions on Cognitive Communications and Networking. 2026;12:9474-9488. doi:10.1109/tccn.2026.3705845

BibTeX
@article{Ye2026,
  author = {Ye, Ziqiang and Gao, Yulan and Xiao, Yue and Lyu, Zhonghao and Yang, Ping and Manolova, Agata},
  title = {Agentic AI for Low-Altitude Economic Networks via LoRA-Assisted Cloud–Edge Collaboration},
  journal = {IEEE Transactions on Cognitive Communications and Networking},
  year = {2026},
  volume = {12},
  pages = {9474--9488},
  doi = {10.1109/tccn.2026.3705845}
}
Kostenlos starten →