Zimo He | 何子默

Welcome to my page!

I'm a Ph.D. student at School of Computer Science, Peking University, advised by Prof. Yixin Zhu and Prof. Yizhou Wang. I'm also a research intern at General Vision Lab, BIGAI, working on general 3D vision and humanoid robot learning. I earned my Bachelor's degree from Yuanpei College, Peking University, majoring in Artificial Intelligence.

Email  /  GitHub

Research

My research interests lie in the intersection of 3D vision, robotics and computer graphics, including human-scene interaction, humanoid robot and human motion synthesis. My long-term goal is to build a general embodied AI that can see, understand and interact with environments, both in virtual and real world.

ViT-H generates humanoid motions directly from monocular videos and uses motion-tracking feedback to improve their executability, enabling real-time video teleoperation with sub-400 ms latency.

We introduce a unified framework that enables humanoid robots to execute multimodal instructions by generating motion tokens through a shared representation and achieving real-time whole-body control.

By finetuning a pretrained multimodal LLM on large-scale motion data, MotionMaster unifies text-driven motion generation and editing in an end-to-end framework, achieving strong zero-shot generalization across multi-action composition and fine-grained body-part editing.

Our text-guided motion editor combines dynamic body-part blending augmentation with an auto-regressive diffusion model to enable diverse motion editing without extensive pre-collected training data.

We propose a framework synthesizing multi-stage scene-aware human motions autonomously and directly from the text instruction and goal location. We also introduce LINGO, a comprehensive language-annotated MoCap dataset, featuring Human-Scene Interaction (HSI) motions.

Teaching

Experience

This website is a modification of Jon Barron's website.