Demo: Embedded Foundation Models as Edge Services: A Programmable Vibrometry Sensing System
Year: 2026
Type: conference-paper
Abstract: Foundation models enable reusable edge intelligence by separating representation learning from downstream sensing tasks. This demonstration presents a programmable vibrometry sensing system that integrates VibroFM with Acies-OS edge middleware on low-cost devices. The demonstration allows participants to select and observe real-time edge inference on acoustic and seismic traces of moving targets through frontend UIs, including ground-vehicle classification and UAS detection, and interactively retarget the system by hot-swapping lightweight task heads on a shared backbone. Beyond runtime reuse, we demonstrate cross-application transfer: a backbone pretrained exclusively on ground-vehicle data is adapted to UAS detection using limited labeled drone data while remaining competitive with models trained directly on the drone corpus.
Institution: University of Illinois Urbana-Champaign
Topic: IoT and Edge/Fog Computing
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funders: Boeing, Army Research Laboratory
OpenAlex ID: W7223135117