Featured Projects
A selection of my recent and ongoing projects in AI, computer vision, and robotics.
Mixture of Experts Imitation Learning Policy
A novel imitation-learning policy that uses a mixture-of-experts with unsupervised task decomposition to reuse skills efficiently.
- Demonstrates strong scalability and data efficiency through expert skill reuse
- Applied to automate surgical subtasks and provide surgeon-assist behaviors on real robotic platforms
C2PNet: Non-Rigid Point Cloud Registration
The first AI-based non-rigid point-cloud registration method, surpassing classical approaches in accuracy and speed.
- Enables robust registration for applications spanning medical imaging, 3D scene reconstruction, and robotics
- Designed to handle noisy, partial, and real-world scans with fast inference
Non-Rigid Registration for Middle Ear Diagnostics with OCT
An OCT-based pipeline that generates 3D middle-ear models and uses C2P-Net registration to improve diagnostic reconstruction quality.
- Built a synthetic training-data pipeline to model noisy and partial anatomy
- Improved anatomical model fidelity through non-rigid point-cloud alignment
- Published at IPCAI 2023; received the Siemens Best Paper Award
