EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Resilient Edge-Cloud Orchestration Framework for Sustaining Tactical Intelligence Continuity in LLM-Enabled V2X Systems

Qingqing Liu, Peilong Zhang, Lingbing Meng, Huihui Han

IEEE Access · 2026 · 14 · 79673-79684

DOI: 10.1109/access.2026.3696386 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Qingqing Liu und 3 weiteren Autor:innen. Die Arbeit nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Software and System Engineering Research Center of Smart Car, Anhui Institute of Information Technology, Youth Research Fund Project of Anhui Institute of Information Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Integrating Multimodal Large Language Models (MLLMs) into Vehicle-to-Everything (V2X) systems represents a fundamental prerequisite for achieving cognitive-level autonomous driving. However, the substantial computational requirements of MLLMs present a significant challenge to the inherent volatility of edge-cloud communication channels, often leading to severe reasoning interruptions during network jitters or stochastic link transitions. To bridge this chasm, we introduce the Resilient Edge-Cloud (REC) framework, a mission-critical orchestration system leveraging multimodal sensor streams (Camera and LiDAR) specifically engineered to perpetuate tactical intelligence continuity. The REC framework operates via three symbiotic components: 1) RepDistiller (Module A) that executes cross-scale tactical semantic alignment via Knowledge Distillation (KD) to compress the cloud model’s reasoning heuristics into the edge surrogate; 2) Robust Collaborative Fusion layer (Module B) that provides link-aware feature synchronization by internalizing transmission latency into Delay-aware Positional Encodings (DPE); and (3) RL-CBF Orchestrator (Module C) that enforces task-offloading within a mathematically rigorous cognitive safety envelope by integrating Reinforcement Learning with Control Barrier Functions. Rigorous evaluations across LLM-CARLA and V2X-Sim 2.0 benchmarks demonstrate that REC achieves a superior trade-off between reasoning accuracy and inference latency. Specifically, REC maintains 86.5% intelligence continuity and achieves a 2.7x speedup relative to standard cloud-centric offloading by eliminating network-induced queuing delays during communication collapse, functioning as a robust contingency mechanism in scenarios where conventional static paradigms exhibit significant performance degradation. The official implementation of the REC framework is available at: https://github.com/peilongzhang88/REC-Framework

Förderung

  • Software and System Engineering Research Center of Smart Car, Anhui Institute of Information Technology
  • Youth Research Fund Project of Anhui Institute of Information Technology

Zitieren

APA

Liu, Q., Zhang, P., Meng, L., & Han, H. (2026). A Resilient Edge-Cloud Orchestration Framework for Sustaining Tactical Intelligence Continuity in LLM-Enabled V2X Systems. IEEE Access, 14, 79673-79684. https://doi.org/10.1109/access.2026.3696386

Vancouver

Liu Q, Zhang P, Meng L, Han H. A Resilient Edge-Cloud Orchestration Framework for Sustaining Tactical Intelligence Continuity in LLM-Enabled V2X Systems. IEEE Access. 2026;14:79673-79684. doi:10.1109/access.2026.3696386

BibTeX
@article{Liu2026,
  author = {Liu, Qingqing and Zhang, Peilong and Meng, Lingbing and Han, Huihui},
  title = {A Resilient Edge-Cloud Orchestration Framework for Sustaining Tactical Intelligence Continuity in LLM-Enabled V2X Systems},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {79673--79684},
  doi = {10.1109/access.2026.3696386}
}
Kostenlos starten →