First-year CIS PhD Student, University of Pennsylvania
Advised by Prof. Kostas Daniilidis at the GRASP Lab
I am working toward robots with the agility and dexterity of humans and animals. My research uses 3D computer vision to recover how they move and interact with the world from everyday video, and turns that understanding into robot skills. Broadly, I am excited about the intersection of 3D vision, robot learning, and agentic perception. I am currently seeking research internship opportunities. If you think there is a good match, feel free to reach out.
Previously, I was a research intern at Stanford University, working with Prof. Leonidas Guibas, Dr. Adam W. Harley, and Prof. Shangzhe Wu. I also spent a year at UC Berkeley with the Nerfstudio team at BAIR, under Prof. Angjoo Kanazawa. I received my undergraduate degree from ShanghaiTech University.
Learning dexterous robot manipulation from monocular human videos: reconstructed hand-object interactions serve as imperfect references, refined with reinforcement learning across object variants in simulation before real-world deployment.
A policy that maintains a belief over an articulated object's joint type and parameters, acting to maximize the information each interaction reveals about its hidden kinematics.
A comprehensive framework for learning and generating 3D quadruped motion directly from large-scale video data.
A pipeline for dense animal pose annotation by optimizing general-purpose point trackers — fine-tuning a lightweight appearance embedding from a few annotated frames to label an entire video with minimal supervision.
Recovering human avatars from self-occluded internet videos where people show only parts or sides of their body.
An open-source library for CUDA-accelerated differentiable rasterization of 3D Gaussians with Python bindings. I worked on the Python bindings, CUDA testing, and alpha-channel rendering.
A modular framework for neural radiance field development. I worked mainly on the integration of Gaussian splatting.
