Zhikai Zhang

Zhikai Zhang | 张智楷

I am a second-year Ph.D. student in the IIIS at Tsinghua University, advised by Prof. Li Yi.

Currently, I do research about humanoid robot learning at Galbot. We are actively looking for interns and full-time employees here about humanoid research.

Email / Scholar / Github / X

News

[2026/01] Any2Track is accepted by ICRA 2026.

[2025/12] R²S² is accepted by RA-L.

[2024/07] FreeMotion is accepted by ECCV 2024.

[2024/02] FreePoint is accepted by CVPR 2024.

Open-source projects

I maintain several open-source projects based on our research works. Feel free to use them and we welcome any feedback from the community.

LATENT . A full-stack pipeline , from motion tracking to latent action space construction to high-level policy learning. Enable the humanoid to learn tennis and potentially many other athletic skills .

Click and Traverse . A humanoid teleoperation system with local spatial intelligence. Users can guide the humanoid through cluttered indoor scenes with just a single click, while avoiding collisions.

OpenTrack . A humanoid motion tracking training framework, based on our work Any2Track. Key features: directly train in MuJoCo, support multi-GPU parallel training.

OpenWBT . A cross-embodiment, easy-to-deploy VR-based humanoid whole-body-teleoperation system, based on our work R²S². Try it!

Research

	Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data Zhikai Zhang, Haofei Lu, Yunrui Lian, Ziqing Chen, Yun Liu, Chenghuai Lin, Han Xue, Zicheng Zeng, Zekun Qi, Shaolin Zheng, Qing Luan, Jingbo Wang, Junliang Xing, He Wang, Li Yi arxiv, 2026 project page / arXiv / code We present LATENT*. Our method constructs a latent action space from imperfect human motion data to learn athletic humanoid tennis skills.
	Collision-Free Humanoid Traversal in Cluttered Indoor Scenes Han Xue, Sikai Liang, Zhikai Zhang, Zicheng Zeng, Yun Liu, Yunrui Lian, Jilong Wang, Qingtao Liu, Xuesong Shi, Li Yi arxiv, 2026 project page / arXiv / code We present Click and Traverse*. It enables collision-free humanoid traversal in cluttered indoor scenes with Humanoid Potential Field (HumanoidPF), a general guidance for learning obstacle avoidance skills.
	Track Any Motions under Any Disturbances Zhikai Zhang, Jun Guo, Chao Chen, Jilong Wang, Chenghuai Lin, Yunrui Lian, Han Xue, Zhenrong Wang, Maoqi Liu, Jiangran Lyu, Huaping Liu, He Wang, Li Yi ICRA, 2026 project page / arXiv / code (OpenTrack) We present Any2Track. It learns to track any motions (over 40 hours, highly-dynamic, contact-rich) under any disturbances (terrain, external forces, payloads) by introducing a dynamics world model for policy fine-tuning.
	Unleashing Humanoid Reaching Potential via Real-world-Ready Skill Space Zhikai Zhang, Chao Chen, Han Xue, Jilong Wang, Sikai Liang, Zongzhang Zhang, He Wang, Li Yi RA-L, 2025 LEAP Workshop @ CoRL, 2025 (Spotlight) project page / arXiv / code (OpenWBT) We present Real-world-Ready Skill Space (R²S²)*. It constructs a latent action space from a set of pre-trained motor skills to solve loco-manipulation tasks with a large reachable space.
	FreeMotion: MoCap-Free Human Motion Synthesis with Multimodal Large Language Models Zhikai Zhang, Yitang Li, Haofeng Huang, Mingxian Lin, Li Yi ECCV, 2024 project page / arXiv Our method explores open-set human motion synthesis using natural language instructions without any motion data.
	FreePoint: Unsupervised Point Cloud Instance Segmentation Zhikai Zhang, Jian Ding, Li Jiang, Dengxin Dai, Guisong Xia CVPR, 2024 paper / arXiv / code Our method explores unsupervised point cloud instance segmentation.

Thanks for watching — whether you're human or robot :)
Template stolen from Jon Barron.
Last updated: March, 2026