We are witnessing a rapid paradigm shift in how we interact with the physical world. To achieve true physical intelligence, computing systems must possess the ability to perceive, interpret, reason, and act within physical spaces. My research envisions democratizing these Physical-AI services, where ubiquitous everyday devices serve as pervasive sensing interfaces, seamlessly powering human-centric applications.
To realize this vision, my work operates at the dynamic intersection of Embodied AI, HCI, and Cyber-Physical Systems (CPS). During my Ph.D., I have primarily leveraged acoustics as a powerful medium to communicate with and sense the physical world. Specifically, I interpret acoustic signals through various dimensions to demonstrate their transformative power:
Perceive the Ambient Space: For AI to act natively in the physical world, it must first understand spatial geometry. By modeling the acoustic field, we develop practical solutions capable of recovering the intricate geometry of indoor spaces. This enables advanced applications including spatial reasoning, fine-grained mapping, and high-fidelity field reconstruction.
Interpret Human Activities: Physical AI must interact seamlessly with users. By fusing multi-modal signals (e.g., acoustics, video, RF), we decode complex human motions and contexts. This empowers critical human-centric applications, ranging from continuous health monitoring and fall detection to robust human activity recognition.
Enable Scalable Physical Interfaces: For Physical AI to be truly ubiquitous, it must be scalable and energy-efficient. We develop systematic approaches to let the sensing algorithms seamlessly apply to the existing infrastructure and devices, without relying on specialized hardware. Concurrently, we explore novel low-power and self-sustaining sensing modalities to sustainably power next-generation Physical-AI services.