A Foundation Model for Daily Activity Sensing Using Ambient Wireless Signals

Professor: Xinyu Zhang

Description: Knowledge about what a person does across the day is a critical input for many ubiquitous computing applications, such as life logging, elderly care, in-home patient care, etc. To obtain such information, existing approaches use either specialized on-body sensors which are intrusive and cumbersome to maintain, or cameras which do not work in low-light condition and often impinge on people’s privacy. In this project, we propose to reuse the ambient wireless signals such as WiFi and radar to track people’s activities. We will develop a foundation model, trained on synthetic datasets, to ensure reliable sensing across diverse environments. The goal is to achieve near-vision sensing resolution using the non-visual sensors. This project will involve substantial amount of data collection, machine-learning model design and testbed implementation.

Preferred Qualifications: Machine learning; Python programming; Signals and Systems