I am most drawn to intelligence that can perceive, reason, create, use tools, and act in the physical world. My work spans human-motion recovery, motion retargeting, robot data collection, generative policies, simulation, and deployment.
CS & Robotics @ HKUST · Hong Kong
- Embodied intelligence — recovering and retargeting human motion for scalable robot data.
- Generative control — policies that learn actions from multimodal demonstrations.
- Systems — simulation, data pipelines, evaluation, and real-world deployment.
- Interfaces — agents and spatial tools for directing increasingly capable models.
| Project | What it explores |
|---|---|
| OpenMarble | A spatial-computing experiment that turns a single image into an explorable 3D Gaussian-splat world. |
- Worked across machine-learning research, startup building, robotics, and product engineering.
- Researched steer-decoding algorithms for language models at ETH Zürich's SRI Lab.
- Co-founded Morpha to build computer-vision systems for manufacturers.
- Built with the HKUST Robotics Smart Car team and competed with HKUST's championship-winning CSE programming team.
I am happiest close to the whole loop: new capabilities → useful systems → deployment → feedback from the real world.