Agentic Federated LLMs at the Edge for Promptable Guardrail Inspection with YOLO-E

Le résumé fourni par la source

This work presents an agentic AI framework that integrates Federated Learning (FL) for Large Language Models (LLMs) and open-vocabulary object detection (OVD) to automate roadside infrastructure monitoring. We propose a distributed pipeline operating throughout the Edge–Cloud Continuum, where a YOLO-E detector running on edge devices identifies guardrails and potential damages in real time, while a federated LLM summarizes the results into structured inspection and maintenance reports. The system leverages Parameter-Efficient Fine-Tuning (PEFT) for LLMs, secure aggregation, and energy- and QoS-aware orchestration among heterogeneous clients. Experimental results in preliminary roadside scenarios demonstrate the feasibility of performing open-vocabulary defect detection and natural language reporting under privacy-preserving and bandwidth-constrained conditions.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.