Academic homepage
Ph.D. Candidate, Computer Science & Engineering · Michigan State University
I study Spatial Intelligence: how AI can move from incomplete multimodal observations to a persistent understanding of dynamic physical spaces, and ultimately support grounded action.
A little context
I am a Ph.D. candidate in Computer Science & Engineering at Michigan State University, advised by Dr. Zhichao Cao. My background in mathematics and mobile computing led me to a practical question: how can AI remain useful when the physical world is incomplete, noisy, and changing?
I build sensing and learning systems that connect multimodal observations to physical state. My long-term interests are in Spatial Intelligence, dynamic world understanding, and human-centered systems that can reason and act with physical context.
Research Vision
The physical world is only partially observed. I study how AI can maintain a useful account of a changing space: where entities are, how they relate, and what may happen next.
Spatial Intelligence, operationally
Spatial Intelligence moves from incomplete multimodal observations to persistent representations of dynamic physical spaces, then toward reasoning, prediction, and grounded action.
One visual, six questions: the progression remains readable without motion or JavaScript.
Featured Research
These projects are evidence for different parts of the Spatial Intelligence agenda: three published foundations, and one ongoing direction toward grounded reasoning and action. Each project is linked to its full research context.
Multimodal Perception
Complementary mmWave–vision fusion for sensing under adverse conditions
Research question. Can complementary sensing modalities preserve physical perception when one channel degrades?
Core idea. Fuse mmWave depth images and RGB features with a learned multimodal encoder.
Key result. Up to 96% accuracy across evaluated scenarios, with around 90% accuracy in rainy, dawn, or poorly lit farm conditions.
Cross-modal Learning
Knowledge transfer across sensing modalities when one is unavailable
Research question. Can knowledge transfer preserve mmWave sensing when RGB is unavailable or unreliable?
Core idea. Use an RGB teacher to guide a mmWave SAR student through cross-modality knowledge transfer.
Key result. Up to 96.3% accuracy across varied environmental scenarios.
Physical Representation
Recovering fine-grained physical state from sparse RF observations
Research question. Can sparse mmWave observations recover fine-grained physical state?
Core idea. Combine signal processing, contrastive feature learning, and calibration for efficient mmWave imaging.
Key result. Leaf-wetness MAE of 4.43 in controlled settings and 6.49 in farm settings, versus 11.84 and 14.32 for traditional sensors.
Dynamic Spatial Understanding · Ongoing
Human-centered sensing and safety-aware support for older drivers
Research question. Can in-car observations support safer, earlier intervention in a dynamic driving space?
Core idea. Combine simulator assessment, continuous in-car monitoring, realistic testing, and community education.
Key result. Ongoing research; this page does not claim a published spatial-reasoning model or closed-loop action policy.
Publications
Selected peer-reviewed papers and research artifacts spanning multimodal sensing, cross-modal learning, physical representation, and agricultural intelligence. Full profile on Google Scholar →
Highlighted papers & research artifacts
All publications
Updates
The Hydra-Bench multimodal leaf-wetness dataset (mmWave raw + SAR + RGB) is released on arXiv.
My journey from math to computing was highlighted by MSUTODAY.
Proteus was accepted at ACM SenSys 2025.
Adonis and AeroEcho were accepted at IEEE INFOCOM 2025.
Invited talk at the University of Hawaiʻi at Mānoa.
Hydra was accepted at ACM MobiCom 2024 and presented in Washington, D.C.
About / Activities
For advisor, education, service, talks, collaboration interests, and contact, here is the fuller academic profile.