Ph.D. Student, Columbia University
I am a Ph.D. student at Columbia, as part of the Columbia Core AI Lab (CAIL), advised by Yunzhu Li.
My main research interest is robot learning, with a focus on understanding the role of physics simulators and digital twins in learning-driven robotic pipelines, and leveraging real-to-sim-to-real methods to achieve scalable data collection, training, and evaluation of robot policies. I am currently a research intern at World Labs. Previously, I was a research intern at SceniX during the summer of 2025, working with Yunzhu Li and Changxi Zheng. My research is partially supported by the Qualcomm Innovation Fellowship.
Before Columbia, I received my Bachelor's degree from Tsinghua University (Yao Class). I am fortunate to receive mentorship from Kris Hauser during my Ph.D. study, and Xiaolong Wang, Yang Gao, Li Yi during my undergrad.
* indicates equal contribution. Representative papers are highlighted.
ICRA 2026
CVPR 2026 4DV Workshop (Oral Presentation)
A framework for robot policy evaluation in simulation environments, using Gaussian Splatting for rendering and a soft-body digital twin for dynamics.
IROS 2025 RoDGE Workshop
An interactive digital twin construction (real-to-sim) framework that learns the full dynamics of elastoplastic articulated objects from videos.
ICCV 2025
We optimize a spring-mass physics model of deformable objects and integrate it with 3D Gaussian Splatting for real-time re-simulation with rendering.
RSS 2024
ICRA 2024 RMDO Workshop (Best Abstract Award)
We learn a material-conditioned neural dynamics model using a graph neural network to enable predictive modeling of diverse real-world objects and efficient manipulation via model-based planning.
CVM 2026
We achieve 4D neural implicit reconstruction from only a single-view scan using deformation and topology regularizations.
ICLR 2023
A fully self-supervised method for category-level 6D object pose estimation by learning dense 2D-3D geometric correspondences, trainable on image collections without any 3D annotations.
ECCV 2022 (Oral Presentation)
We show that fusing fine-grained features learned with low-level contrastive objectives and semantic features from image-level objectives can improve SSL pretraining.