Research
AI, computer vision, and extended reality for accessible health and human motion.
My group builds AI systems that capture, understand, and re-render human motion, with a focus on making rehabilitation and health technology accessible: a single camera instead of a motion-capture studio, a treadmill anyone can buy instead of a clinical split-belt system, and reasoning models that can explain what they see.
Research Areas
Neural rendering and accessible XR
Free-viewpoint holographic rendering of patients from a single camera, using 3D Gaussian Splatting adapted to real-world clinical settings, so that clinicians and patients can meet in shared XR spaces for telerehabilitation. This line of work continues in my NSF-funded project on an accessible XR system for human motion monitoring, reasoning, and rendering with 3D vision-language models.
Video understanding with vision-language models
Teaching vision-language models to reason about motion and events rather than just label them: reflection-aware learning for video anomaly understanding, tokenizing 2D motion so language models can read fine-grained rehabilitation movements, and faithfulness audits of reinforcement-trained VLMs.
Intelligent rehabilitation devices
Real-time, fine-resolution gait phase recognition with deep learning, and intelligent treadmill control that simulates a split-belt treadmill on a single belt for post-stroke rehabilitation. Two U.S. patents have been granted on this work.
AI for health and society
Deep learning on satellite imagery and spatio-temporal data for opioid overdose monitoring in rural Alabama, structured knowledge graphs for retrieval-augmented generation, and earlier work on deep-learning reconstruction and super-resolution for optical coherence tomography (OCT).
Funding
- National Science Foundation, Award #2553446 (Sole PI). ERI: Accessible XR System for Human Motion Monitoring, Reasoning, and Rendering with 3D Vision-Language Models. Division of Electrical, Communications and Cyber Systems. 2026–2028.
- National Science Foundation, Award #2612358 (Co-PI). Collaborative Research: CyberTraining: Implementation: Medium: ACAI-Train: Scalable Instructor Training for Infusing AI and Advanced CI Concepts into Early Core Computing Courses. Office of Advanced Cyberinfrastructure. PI: David P. Bunde. 2026–2030.
Students
Undergraduates are full members of the group. Recent projects include a College Honors thesis on fine-tuning single-image-to-3D models for 4D XR human visualization (Richie Dix, 2026) and a summer 2026 cohort of four Richter Grant and Knox Summer Scholar researchers, one of whose projects led to a first-authored conference submission.
Prospective students: if you are a Knox student who wants to work on AI, computer vision, or XR, email me with a short note about what you would like to build and which courses you have taken. No prior research experience is required.