Zhejiang University · Computer Science

Junhan Sun 孙俊涵

I am a first-year Ph.D. student at Zhejiang University, working with Prof. Guofeng Zhang at Zhejiang University and Prof. Hao Zhao at Tsinghua University.

My research connects predictive representations, world models, and embodied control. I am interested in how learned models turn visual goals into efficient, purposeful action.

Previously: Chu Kochen Honors College, Zhejiang University.

Email Google Scholar GitHub X

Portrait of Junhan Sun — CAD&CG · Hangzhou, China

Research

RepresentationIntentAction · Learning the world to act in it.

Explore INTACT

Publications

Public preprints · * Equal contribution

I-BFM humanoid carrying a box in the real-world robustness demonstration. — Real-world interaction

arXiv preprint · 2026

I-BFM: Reward-Conditioned Robust Humanoid Interaction via Unsupervised Reinforcement Learning

Ziqi Han*, Yitang Li*, Junhan Sun*, Fanrong Dong, Yaojie Shen, Lei Ye, Zetong Jing, Yongqi Zhang, Yiming Zhang, Xue Wang, Hao Zhao

A reward-conditioned humanoid policy learns reusable object interaction and recovery skills through unsupervised reinforcement learning.

Paper Project Co-first author

Simulation · 8 s

arXiv preprint · 2026

FlexiWorld: Learning and Planning via Flexible Action Chunks Across Multiple Time Scales

Shidu Ren*, Qilin Gu*, Zhenghao Ni*, Junhan Sun, Jiaqi Wang, Damien Scieur, Yunze Liu

A JEPA world model combines variable-length action chunks and goal-conditioned action generation for planning across time scales.

Paper Project Code Models

PredActor humanoid in the project’s real-world hardware demonstration. — Onboard humanoid control

arXiv preprint · 2026

PredActor: Predictive Action Diffusion for Steerable Onboard Humanoid Control

Lei Ye, Haibo Gao, Yitang Li, Peng Xu, Zetong Jing, Junhan Sun, Fanrong Dong, Ziqi Han, Xue Wang, Jianhua Sun, Cewu Lu, Hao Zhao, Liang Ding

Joint state–action diffusion predicts future motion for guidance and directly executes humanoid actions from proprioceptive observations.

Paper Project

INTACT project graphic connecting latent intent to action. — Intent → action

arXiv preprint · 2026

INTACT: Isomorphic Intent-to-Action Learning for Search-Free World Models

Junhan Sun, Hao Zhao, Guofeng Zhang

Learning action-aligned latent intent turns visual goals into direct, search-free control within a world model.

Paper Project Code

Real-world demo · 3× playback

arXiv preprint · 2026

LAMP: Lift Image-Editing as General 3D Priors for Open-world Manipulation

Jingjing Wang, Zhengdong Hong, Chong Bao, Yuke Zhu, Junhan Sun, Guofeng Zhang

Image-editing priors are lifted into inter-object 3D transformations to guide open-world robotic manipulation.

Paper Project

Software

CLEAR-LeWMCommunity evaluation software

An auditable evaluation suite for LeWM-compatible latent world models, with official-compatible and difficulty-controlled task protocols.

Project GitHub

Updates

Oct. 2026
I-BFM is now a public preprint.
Sep. 2026
New preprints: FlexiWorld and PredActor.
Aug. 2026
Released INTACT code, evaluation stack, and model artifacts.

Earlier updates

Jul. 2026
Released the INTACT preprint and project page.
Jul. 2026
Open-sourced CLEAR-LeWM for world-model evaluation.

Education

2026 — Present

Ph.D. in Computer Science

Zhejiang University · College of Computer Science and Technology
State Key Laboratory of CAD&CG

2022 — 2026

Undergraduate study

Chu Kochen Honors College, Zhejiang University

Contact

I welcome conversations about world models, representation learning, and robot learning.

12621185@zju.edu.cn luoliibaqi4747@gmail.com