Uni-Fi: Integrated Multi-Task Wi-Fi Sensing
Mengning Li, Wenye Wang
TL;DR
This work tackles the fragmentation of Wi-Fi sensing tasks by proposing Uni-Fi, a unified framework that captures the common forward-model and inverse-model structure across sensing tasks and couples it with a scalable, data-driven pipeline to assemble aggregated inverse models. By formalizing single-task sensing as feature-to-task mappings and extending to multi-task sensing via a forward-model network, Uni-Fi enables plug-and-play integration of new tasks and features without redesigning the entire system. The approach combines a forward-model-based theoretical framework, a simulator-driven modeling pipeline, and a Transformer-enabled inference engine to achieve robust device-free tracking and state recognition across environments, reporting localization errors around $0.54\,\mathrm{m}$ and multi-task accuracies near the high-90s. The results demonstrate that feature-level representations and modular architecture yield stable, scalable, and high-performance Wi-Fi sensing suitable for long-term smart-home deployments, with practical real-time capabilities on commodity hardware.
Abstract
Wi-Fi sensing technology enables non-intrusive, continuous monitoring of user locations and activities, which supports diverse smart home applications. Since different sensing tasks exhibit contextual relationships, their integration can enhance individual module performance. However, integrating sensing tasks across different research efforts faces challenges due to the absence of two key elements. The first is a unified architecture that captures the fundamental nature shared across diverse sensing tasks. The second is an extensible pipeline that can integrate sensing methodologies proposed in potential future research. This paper presents Uni-Fi, an extensible framework for multi-task Wi-Fi sensing integration. This paper makes the following contributions. First, we propose a unified theoretical framework that reveals the fundamental differences between single-task and multi-task sensing. Second, we develop a scalable sensing pipeline that automatically generates multi-task sensing solvers, enabling seamless integration of multiple sensing models. Experimental results show that Uni-Fi achieves robust performance across tasks, with a localization error of approximately 0.54 meters, 98.34 percent accuracy for activity classification, and 98.57 percent accuracy for presence detection.