PhD Candidate · Robotics
Technical University of MunichApplied and Theoretical Aspects of Robot Intelligence (ATARI) Lab · MIRMI
Supervised by Prof. Dr. Majid Khadiv
Generative action policies and safe robot control.
Research
“The leap from model intelligence to robot intelligence is a leap in action generation.”
My work asks how robots can turn richer observations into actions that are safe, physically feasible, and reusable across tasks.
I study how richer sensing can shape action generation, from thermal perception to multimodal vision-language-action (VLA) policies. My goal is to ground robot behavior in physical context, including conditions where visual observations alone are unreliable.
I combine diffusion and flow matching with control barrier functions and trajectory optimization. I work toward action generation that respects safety limits and robot dynamics, connecting learned behaviors to motions a robot can execute.
I explore how generative policies can discover and learn diverse behaviors, including rare but valid actions. This motivates my work toward a growing repertoire of skills that robots can reuse and combine across tasks.
Publications
Diffusion policies, flow matching, certified constraints, thermal VLA, and deployment-facing robot systems.
2026
Preprint
Shafeef Omar, Sebastian Sanokowski, Elham Daneshmand, Dian Yu, Daniel Cremers, Cordelia Schmid, Majid Khadiv
arXiv:2610.24847
2026
CoRL 2026
Dian Yu, Sebastian Sanokowski, Majid Khadiv
arXiv:2606.08743
2026
Preprint
Mohammad Hasan Yeganegi, Dian Yu, Andrea Del Prete, Majid Khadiv, Matteo Saveriano
arXiv:2607.24296
2026
IROS 2026
Dian Yu*, Qingchuan Zhou*, Bingkun Huang, Majid Khadiv, Zewen Yang
arXiv:2603.05754
2026
IEEE RA-L
Victor Dhedin*, Ilyass Taouil*, Shafeef Omar*, Dian Yu, Kun Tao, Angela Dai, Majid Khadiv
arXiv:2602.06827
2026
IEEE T-RO
Zewen Yang*, Xiaobing Dai*, Dian Yu*, Zhijun Li, Majid Khadiv, Sami Haddadin, Sandra Hirche
arXiv:2506.02955
2025
Under review
Xiaobing Dai*, Zewen Yang*, Dian Yu*, Fangzhou Liu, Hamid Sadeghian, Sami Haddadin, Sandra Hirche
arXiv:2504.08661Experience
ATARI Lab, supervised by Prof. Dr. Majid Khadiv.
Physically consistent system identification via convex optimization and LMI constraints; sampling-based control and MPC for flow-matching training and inference; VLA models for humanoid loco-manipulation.
ROS2 control stack development and electronics integration for a lower-limb exoskeleton platform.
Robot learning research and prototyping in Bosch Corporate Research.
Prototype learning-based control for an autonomous legged-wheeled rescue robot; selected as a TUM IDEAward finalist.
Connecting wires together.
Zurich, Switzerland
Real-time conversation and physical interaction with a humanoid robot.
View repository
Munich, Germany
Generative robot throwing with reinforcement learning and flow matching.
View repository
Education
PhD Candidate, ATARI Lab. M.Sc. Robotics, Cognition, Intelligence, GPA 1.3, graduated with high distinction.
Dual B.Eng. background in electrical engineering and automation, ranked first in both programs.
Tools
Languages
Chinese native · English full professional · German professional working proficiency.
Contact