Research

Research Directions
  • Physical Human Intelligence — Modeling, understanding, and generating the human body, motion, and its interaction with the physical world, from images and video.
    3D human biomechanics motion & skill embodied AI
  • Understanding People in Long, Real-World Video — Persistently understanding who a person is, what they do, and how they change across long, multi-camera, real-world video.
    video Re-ID temporal reasoning aerial-ground anomaly discovery
  • Trustworthy Biometrics and Digital Identity — Creating, recognizing, and protecting human identity in a world increasingly populated by synthetic people, avatars, and AI-generated media.
    face / body / gait deepfake detection synthetic identity open-set
  • Generative and Multimodal AI for the Physical World — Geometry-consistent, physically plausible, and controllable generation and multimodal reasoning over 3D, video, and language.
    3D generation VLMs simulation domain generalization
Impact Areas
  • AI for Health and Human Performance — Interpretable, video-based digital biomarkers that quantify how a person moves, changes over time, and responds to intervention.
    clinical gait rehabilitation coaching remote assessment
  • AI for Education and Scientific Learning — Multimodal AI that reasons about student thinking, instructional practice, and the evidence educators use to make decisions.
    classroom video student drawings teacher feedback pedagogical reasoning