Zichen Zhou 周子琛

First-Year Master's Student 硕士一年级研究生

Beijing Institute of Technology 北京理工大学

Advisor: Yufeng Yue 导师：岳裕丰教授

Hi! I am a first-year master's student at Beijing Institute of Technology. My research interests mainly focus on Embodied AI and Robotics, particularly developing Vision-Language-Action Models and Visuomotor Policies for dexterous manipulation in real-world scenarios. 您好！我是北京理工大学的一名硕士一年级研究生。我的主要研究兴趣集中在具身智能 (Embodied AI)和机器人学 (Robotics)。我致力于开发针对真实世界灵巧操作的视觉-语言-动作模型和视觉运动策略。

Gmail Google Scholar GitHub

Latest News最新动态

2026.07

🚀 New preprint released: "HOST: Robots Acquire Manipulation Skills in Seconds from a Single Human Video". The project page and code are now available! 🚀 新预印本发布："HOST：机器人从单段人类视频中快速学习操作技能"。项目主页与代码现已公开！
2026.06

🎉 Two papers accepted by IROS 2026: "GLUE" and "FCAR: Force-Conditioned Action Residual Policies for Contact-Rich Manipulation"! 🎉 两篇论文被 IROS 2026 接收："GLUE" 与 "FCAR: Force-Conditioned Action Residual Policies for Contact-Rich Manipulation"！
2026.03

🎉 Our paper "Learning a Unified Latent Action Space from Videos with Action-centric Cycle Consistency" was accepted by CVPR 2026! 🎉 我们的论文 "Learning a Unified Latent Action Space from Videos with Action-centric Cycle Consistency" 被 CVPR 2026 接收！
2026.06

🏆 Won 1st Place at the ManiSkill-ViTac Challenge 2026 at the CVPR 2026 Embodied AI Workshop! 🏆 在 CVPR 2026 Embodied AI Workshop 的 ManiSkill-ViTac Challenge 2026 中获得冠军！
2025

🚀 New preprint released: "See Once, Then Act" on Vision-Language-Action Models. 🚀 发布了关于视觉-语言-动作模型的新预印本论文："See Once, Then Act"。
2025

🎉 Our paper "CordViP" was accepted by RSS 2025! 🎉 我们的论文 "CordViP" 被机器人领域顶级会议 RSS 2025 接收！

Publications 论文发表

Robots Acquire Manipulation Skills in Seconds from a Single Human Video

G Chen, M Wang, T Cui, Zichen Zhou, Q Shao, X Li, H Su, R Gan, H Wang*, M Fu*, Y Yang*, Y Yue*

arXiv preprint arXiv:2607.20033, 2026. arXiv preprint arXiv:2607.20033, 2026.

We introduce HOST, a framework that enables robots to acquire previously unseen manipulation skills at inference time from a single human video, without task-specific fine-tuning, while retaining previously mastered skills. 我们提出 HOST，一种让机器人在推理阶段仅通过一段人类视频学习全新操作技能的框架。该方法无需针对特定任务进行微调，同时能够保留机器人已经掌握的技能。

arXiv Project GitHub

Learning a Unified Latent Action Space from Videos with Action-centric Cycle Consistency

G Chen, Q Shao, T Cui, Zichen Zhou, W Mao, L Yang, M Wang, Y Yang, H Chen, Y Yue*

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026.

arXiv

GLUE: Global-Local Unified Encoding for Imitation Learning via Key-Patch Tracking

Y Chen, Zichen Zhou, J Dou, T Cui, Y Yang, Y Yue

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026.

We propose GLUE, a global-local unified encoding framework for imitation learning based on key-patch tracking. By fusing global and local visual representations, GLUE steers attention toward task-relevant objects and improves policy robustness in cluttered and occluded settings. 我们提出了 GLUE，一种基于关键点 patch 跟踪的全局-局部统一编码模仿学习框架。通过融合全局与局部视觉表征，GLUE 将注意力引导至任务相关物体，提升在杂乱与遮挡场景下的策略鲁棒性。

arXiv

CordViP: Correspondence-based Visuomotor Policy for Dexterous Manipulation in Real-World

Y Fu, Q Feng, N Chen, Zichen Zhou, M Liu, M Wu, T Chen, S Rong, J Liu

Robotics: Science and Systems (RSS), 2025. Robotics: Science and Systems (RSS), 2025.

We propose CordViP, a novel correspondence-based visuomotor policy that significantly improves the robustness and generalization of dexterous manipulation in real-world scenarios. 我们提出了 CordViP，一种新颖的基于对应关系的视觉运动策略，显著提高了真实场景下灵巧操作的鲁棒性和泛化能力。

PDF Project

See Once, Then Act: Vision-Language-Action Model with Task Learning from One-Shot Video Demonstrations

G Chen, M Wang, Q Shao, Zichen Zhou, W Mao, T Cui, M Zhu, Y Deng, L Yang

arXiv preprint arXiv:2512.07582, 2025. arXiv preprint arXiv:2512.07582, 2025.

This paper introduces a Vision-Language-Action (VLA) model capable of learning new tasks from just one-shot video demonstrations, greatly reducing the data requirements for robotic imitation learning. 本文介绍了一种视觉-语言-动作 (VLA) 模型，该模型能够仅从单次视频演示中学习新任务，极大地降低了机器人模仿学习的数据需求。

arXiv

View full publication list on Google Scholar 查看完整的出版物列表 (Google Scholar)

Awards荣誉奖项

2026.06

🏆 1st Place — ManiSkill-ViTac Challenge 2026 🏆 冠军 — ManiSkill-ViTac Challenge 2026

Won 1st Place at the ManiSkill-ViTac Challenge 2026 at the CVPR 2026 Embodied AI Workshop. Team: BIT IN²Bot. 在 CVPR 2026 Embodied AI Workshop 的 ManiSkill-ViTac Challenge 2026 中获得冠军。队伍：BIT IN²Bot。