Chengyang Li

Chengyang Li (李承阳)

I am a first-year jointly supervised Ph.D. student at SJTU and EIT, where I am advised by Prof. Wentao Zhu. Previously, I interned at Galbot, under the guidance of Prof. Li Yi. I received my bachelor's degree from ShanghaiTech University, where I was fortunate to work closely with Prof. Lan Xu, Prof. Jiayuan Gu, and Prof. Jingyi Yu. My research interests lie in Physical RSI.

Email / CV / Google Scholar / Github

Research

* Equal contribution. † Corresponding author.

Thea: Towards the Harness of Embodied Agents
Qi Wang*, Tianyi Wang*, Chengyang Li*, Shikun Ban, Yurun Chen, Yizhong Ge, Jason Qin, Chengtai Li^†, Wentao Zhu^†
Technical Report, 2026
Paper/ Code/ Project/ Video

We present Thea, a harness that adapts the core components of coding agents to the physical world, where an agentic loop orchestrates robot capabilities wrapped as callable tools.

GazeVLA: Learning Human Intention for Robotic Manipulation
Chengyang Li*, Kaiyi Xiong*, Yuan Xu, Lei Qian, Yizhou Wang, Wentao Zhu^†
CoRL 2026 ★ Spotlight
Paper/ Code/ Project/ Video/ Data

We propose a novel learning-from-human framework that explicitly models intention to capture the causal structure of manipulation behavior.

SemanticMN: Semantic Anchoring for Robotic Action Representations
Yuan Xu*, Youheng Shi*, Chengyang Li, Wentao Zhu, Yizhou Wang^†
CoRL 2026 ★ Spotlight
Paper/ Code/ Project/ Video

We anchor robotic action representations to a semantic manifold, improving task performance and generalization without changing the deployed model.

SMGDiff soccer motion generation preview

SMGDiff: Soccer Motion Generation using diffusion probabilistic models
Hongdi Yang*, Chengyang Li*, Zhenxuan Wu, Gaozheng Li, Jingya Wang, Jingyi Yu, Zhuo Su, Lan Xu^†
ICCV 2025
Paper/ Project/ Code/ Bibtex

We introduce SMGDiff, a novel two-stage framework for generating real-time and user-controllable soccer motions.