Chenyu Zhang

Cambridge, MA

chenyuzhang@g.harvard.edu

I’m an S.M. student in Computational Science and Engineering at Harvard University (GSAS / SEAS). I hold a B.S. in Statistics from the School of Mathematical Sciences at Peking University.

My interests include: Robot Learning, Generative AI, 3D Vision for Robotics, and Computer-Use Agents.

I like building the pieces that let a system be measured honestly — benchmarks, calibration, simulation — and then the models that have to pass those tests. Recently that has meant converting real robot video into physics-consistent simulation, generative models of 3D object-particle dynamics, and long-horizon office workflows for GUI agents.

Most recently I was a research intern in the Qingyun Program at Tencent, where I was a core contributor to UI-Mate, the Hy Team’s foundation GUI agent, and owned task construction for its OSWorkerBench benchmark.

Before that, I was a research associate with Dr. Katerina Fragkiadaki in the Machine Learning Department at Carnegie Mellon University, and a research assistant at the OV³ Lab under Dr. Jiahuan Zhou in the Wangxuan Institute of Computer Technology, Peking University.

Away from the screen you’ll usually find me on a tennis or pickleball court, at a piano, or over a chessboard.

news

Sep 01, 2026	Started my S.M. in Computational Science and Engineering at Harvard University. Hello, Cambridge!
Aug 28, 2026	UI-Mate is out — Tencent Hy Team’s open-weight foundation GUI agent, released with OSWorkerBench, a benchmark of 100 long-horizon office workflows across 41 applications whose task construction I owned. Code · Report.
Jun 30, 2026	Graduated from Peking University with a B.S. in Statistics from the School of Mathematical Sciences. Four wonderful years — thank you to everyone along the way.
Jan 22, 2026	RobotArena ∞ has been accepted to ICLR 2026! 🎉
Oct 27, 2025	RobotArena ∞ is on arXiv — scalable robot benchmarking by translating real teleoperation video into simulation. Project page.

selected publications

arXiv

UI-Mate: Advancing Open-Weight Foundation GUI Agents with In-Context Demonstrations

Tencent Hy Team

In arXiv preprint arXiv:2608.15930. Core contributor to the technical report; owned task construction for OSWorkerBench , 2026

Abs HTML Code

UI-Mate is an open-weight foundation GUI agent that pairs an environment-grounded training stack with in-context demonstration learning. Released alongside it is OSWorkerBench, a benchmark of 100 long-horizon office tasks spanning 41 applications, evaluating both general computer-use ability and the ability to follow a procedure supplied by a multimodal demonstration.
ICLR

RobotArena ∞: Scalable Robot Benchmarking via Real-to-Sim Translation

Yash Jangir, Yidi Zhang, Kashu Yamazaki, Chenyu Zhang, Kuan-Hsun Tu, and 4 more authors

In International Conference on Learning Representations (ICLR), 2026

Abs HTML

RobotArena infinity shifts vision-language-action policy evaluation into large-scale simulated environments built automatically from real robot video. The pipeline converts demonstrations from widely used robot datasets into physics-consistent simulated counterparts using vision-language models, 2D-to-3D generative modeling and differentiable rendering, requiring only a single RGB image from a static-camera demonstration. Policies are scored by both VLM-guided automatic evaluation and scalable human preference judgments.
CVPR

SCAP: Transductive Test-Time Adaptation via Supportive Clique-Based Attribute Prompting

Chenyu Zhang, Kunlun Xu, Zichen Liu, Yuxin Peng, and Jiahuan Zhou

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

Abs

SCAP preserves vision-language model accuracy under deployment distribution shift with no labels and no retraining, adapting the model dynamically based on cliques of visually similar test samples. Accuracy improves further as the test batch grows.