2025 Theses Doctoral

Embedded AI and Sensing for Wellness, Health, and Sustainability: Intelligent, Privacy-Aware Wearable Devices and AIoT Systems

Nie, Jingping

With the rise of connected devices and the Artificial Intelligence of Things (AIoT), there is immense potential to transform personal wellness, health care, and environmental responsibility through embedded AI, sensing, and machine learning technologies. However, existing solutions often fall short in terms of accessibility, privacy, and sustainability, motivating the need for innovative, human-centric systems that can seamlessly blend into daily life while maximizing social welfare and sustainability. In this thesis, we explore the design and deployment of intelligent, privacy-aware, and pervasive wearable devices and AIoT systems to enhance personal health, well-being, and sustainability.

Motivated by these challenges, my broader research spans four interconnected thrusts: (1) smart wearables and platforms for emotion, fitness, and wellness; (2) model-driven biosignal analysis and emotion detection; (3) AI-based mental and physical health care in smart home environments; and (4) city-scale human-in-the-loop EV-interfaced grid optimization. This thesis presents the major topics and representative systems developed across these four thrusts. Unifying these efforts is a shared human-centric design philosophy that begins with sensing the individual and extends to personalized support from the individual level to city scale.

The first topic introduces low-cost wearable systems for real-time monitoring of emotion and fitness, including wireless glasses for facial expression classification and real dimensional emotion monitoring. The second topic explores contactless audio-based platforms for treadmill running metrics estimation and model-driven analysis of physiological signals using machine learning and acoustic foundation models on post-exercise speech and biosignals.

In the third topic, we design AI-enhanced smart home solutions for daily functioning and mental health support. Through collaborations with psychotherapists, we develop an intelligent home-based screening and intervention framework that integrates ambient sensing, conversational agents, and robotic feedback for scalable, in-situ mental health and day-to-day functioning support.

Scaling to city scale, this thesis, in the fourth topic, advances human-centric optimization of electric vehicle (EV)-interfaced microgrids to support sustainable urban energy systems. We first develop a constrained optimization solver and a reinforcement learning–based control framework to estimate and regulate power flows under partial observability and IEEE-standardized grid services. Building on this, we design a practical, data-driven EV charging recommendation system that accounts for user preferences, charger availability, pricing, and grid capacity. Together, these tools enhance the integration of EVs into smart grids while co-optimizing outcomes for drivers, charger owners, and utility operators.

Across all topics, this dissertation emphasizes hardware-software co-design, real-world deployment, and multidisciplinary collaboration. It proposes a unified framework that senses, interprets, and learns from human behavior through multimodal sensing data, enabling intelligent systems to deliver personalized recommendations and interventions for fitness, mental and physical health, sustainability, and overall well-being. Through rigorous evaluation, public datasets, and open-source tools, this work demonstrates how embedded AI and AIoT systems can meaningfully enhance both individual wellness and community-scale services. The findings point toward a future where pervasive, privacy-aware technologies are seamlessly integrated into daily life, supporting proactive care, sustainable living, and human-centered design at scale.