Structured task-and-motion reasoning
Grounding semantic goals into reusable skills and closed-loop plans for reliable, long-horizon tasks.
Mathematics · Embodied AI · Robot Learning
I work on foundation-model-based robot learning for reliable, long-horizon manipulation.
I am now at THU. Previously, I was at SJTU and HKU.
Let’s talk
Have a research question, project idea, or something interesting to discuss?
View my availability and choose a convenient time to meet.
Current focus
Grounding semantic goals into reusable skills and closed-loop plans for reliable, long-horizon tasks.
Connecting perception and control through effective action representations for reliable manipulation.
Adapting foundation models to physical behavior through in-context learning and policy-controller co-design.
A long-term personal benchmark: enable robots to sit at a table and play mahjong with humans. I support Slow Science.
Research output
I am fortunate to be advised by Mengdi Xu, Jianping He, and Edith C.H. Ngai, and grateful to collaborate with many wonderful colleagues. See the full list on Google Scholar. † denotes equal contribution.
Academic path
University of Nottingham Ningbo China
Manipulation · In-Context Learning · Robot Learning
Robot Learning · Control · Manipulation
Federated Learning · Algorithmic Fairness · Spatio-temporal Graphs
Updates
Aside