Yu Deng

About

I am Yu Deng, a Ph.D. candidate at the Artificial Intelligence and Machine Learning (AI/ML) Lab, hessian.AI, and the JUPITER AI Factory (JAIF) in Darmstadt. My supervisor is Prof. Dr. Kristian Kersting.

I work on embodied and agentic AI, with a focus on systems that can perceive, remember, reason, and act in the physical world. My research connects object-centric 3D perception, geometry-aware robot learning, executable knowledge, and proactive agent memory.

I am particularly interested in building generalizable AI systems that can turn visual observations and accumulated experience into structured representations, reusable skills, and reliable actions. This includes learning from 2D and 3D visual signals, grounding policies in geometry and objects, and designing agents whose memory and knowledge can be inspected, repaired, and extended over time.

Publications

• Nautilus: From One Prompt to Plug-and-Play Robot Learning. arXiv 2026. [arXiv]
  Yufeng Jin^*, Jianfei Guo^*, Xiaogang Jia, Yu Deng, Zechu Li, Han Liu, Weiran Liao, Vignesh Prasad, Mathias Franzius, Gerhard Neumann, Georgia Chalvatzaki.

• Cognifold: Always-On Proactive Memory via Cognitive Folding. arXiv 2026. [arXiv]
  Suli Wang^*, Yiqun Duan^*, Yu Deng^*, Rundong Zhao, Dai Shi, Xinliang Zhou.

• Kintsugi: Learning Policies by Repairing Executable Knowledge Bases. arXiv 2026. [arXiv]
  Teng Cao^*, Yu Deng^*, Hikaru Shindo^*, Quentin Delfosse, Lanxi Wen, Suli Wang, Jannis Blueml, Christopher Tauchmann, Kristian Kersting.

• Robot-DIFT: Distilling Diffusion Features for Geometrically Consistent Visuomotor Control. arXiv 2026. [arXiv]
  Yu Deng^*, Yufeng Jin^*, Xiaogang Jia, Jiahong Xue, Gerhard Neumann, Georgia Chalvatzaki.

• STORM: Segment, Track, and Object Re-Localization from a Single Image. ICML 2026. [arXiv]
  Yu Deng^*, Teng Cao^*, Hikaru Shindo, Jiahong Xue, Quentin Delfosse, Kristian Kersting.

^* Equal contribution.

Education

• Ph.D., TU Darmstadt, Apr. 2026 -
• M.Sc., TU Darmstadt, Oct. 2023 - Feb. 2026