Dexterous World Models
We present DWM, a scene-action-conditioned video diffusion model that simulates dexterous human interactions in static 3D scenes.
Taeksoo Kim
Hi, I'm a 4th year MS/PhD student at Seoul National University, where I am advised by Prof. Hanbyul Joo. My current research interests lie in simulating the physical world with agents using video generative models. I received a B.S in Industrial Engineering and a B.S. in Computer Science and Engineering from Seoul National University.
I'm currently seeking research internship opportunities starting anytime.
Feel free to reach out if you're interested in collaboration!
News
- Feb 2026: Our work, Dexterous World Models, got accepted in CVPR 2026. See you in Denver!
- Jan 2026: Our work, Target-Aware Video Diffusion Models, got accepted in ICLR 2026. See you in Rio, Brazil!
- Mar 2023: Joined SNU Visual Computing Lab.
Research
Target-Aware Video Diffusion Models
project page code & data arXiv openreview
Our target-aware video diffusion model generates a video in which an actor accurately interacts with the target, specified with its segmentation mask.
Robot-Factored World Models via Robot Rendering
We factor action realization and robot appearance out of the world model, presenting actions as explicitly rendered robot geometry.
AutoDex: An Automated Real-World System for Dexterous Grasping Data Collection
We present an automated real-world system that collects dexterous grasp data and build a retrievable database for execution in novel scenes.
ZeroDex: Zero-Shot Long-Horizon Dexterous Manipulation via Multi-View 3D-Grounded VLM Reasoning
We leverage VLMs to generate executable 3D manipulation plans from calibrated multi-view imagery without task-specific training.
GALA: Generating Animatable Layered Assets from a Single Scan
We present a framework that takes as input a single-layer clothed 3D human mesh and decomposes it into complete multi-layered 3D assets.
NCHO: Unsupervised Learning for Neural 3D Composition of Humans and Objects
project page code & data arXiv
We present a framework for learning a compositional generative model of humans and objects (backpacks, coats, and more) from real-world 3D scans.