BesiegeField
Language-model agents build machines as programs and refine them through physical simulation.
Understand.
Create.
We study generative models, 3D reasoning, and how learning systems interact with the physical world.
Selected research
Scroll horizontally or drag to explore projects. With this list focused, use the left and right arrow keys, or Home and End.
EVA
Aligning video world models with actions a robot can execute.
DrPO
Aligning one-step image generators with preferences without differentiating through rewards.
PAct
From a single image to a 3D object with parts that move.
ASAP
Reducing anatomical errors in generated human images through synthetic, localized preferences.
VGG-Flow
Aligning generative flow models with rewards while preserving their learned priors.
Nabla-GFlowNet
Using reward gradients to align diffusion models while preserving diversity and pretrained knowledge.
G-Shell
Reconstructing and generating 3D shapes, from solid objects to thin, open surfaces.
MeshDiffusion
Generating detailed 3D meshes directly with diffusion models.
Team
Faculty
Principal Investigator
Core members
PhD student
Zhou Jiang
PhD student
Jinming Ren
MPhil student
Research assistants & visiting students
Yaocheng Guo
Master’s student
Xiangwei Shen
Master’s student
Zihao Zhou
Undergraduate
Zhuoran Xia
Undergraduate
Jinlong Yang
Undergraduate
Zigeng Xu
Undergraduate
Zixuan Wang
Undergraduate
Alumni
From MPhil, CUHK-SZ to PhD, HKUST
Zhou Jiang
From Undergraduate, SCUT to PhD, CUHK-SZ
Yuping Zheng
From Undergraduate, CUHK-SZ to PhD, University of Virginia
Bao Li
From Master’s, Chinese Academy of Sciences (with Yuliang Xiu) to PhD, Chinese Academy of Sciences
External collaborators
Weiyang Liu
CUHK
Yandong Wen
Westlake University
Yuliang Xiu
Westlake University
Jie Fu
iQuestLab