Research Intern
BAAI · Embodied vision-language-action model pretraining
Hey, I am happy you are here
Thanks for stopping by!
I am a Ph.D. student at the Institute of Automation, Chinese Academy of Sciences. My research interests include prompt learning, domain generalization, and representation learning for vision-language-action models.
My current projects explore generalizable action representations for embodied intelligence, including covariant action modeling, semantic-action manifold alignment, and human-to-robot transfer for robotic manipulation. Learn more about my research interests in publications.
Get in touch by sending me a note.
About me
Belief. Intelligence emerges when generalizable multimodal representations become stable.
My current work centers on VLA training and representation learning for embodied intelligence. I am interested in how vision-language-action models acquire compact, transferable action spaces that support robust manipulation and generalization.
I am a Ph.D. student at the Institute of Automation, Chinese Academy of Sciences, supervised by Changsheng Xu. I am broadly open to research collaborations, visiting opportunities, and industry research roles around embodied AI, robot learning, and multimodal foundation models.
Ph.D. Student
CASIA · Pattern Recognition and Artificial Intelligence
B.S. in Computer Science
News
- May 2026
- Our GAM and LAST have been accepted by ICML. (First Author)
- Feb 2026
- Our paper H2R-BM has been accepted by Pattern Recognition. (Corresponding Author)
- Jul 2025
- Our paper EgoPrompt has been accepted by ACM Multimedia. (First Author)
- Mar 2025
- Our paper on multiple local prompts distillation has been accepted by IEEE Transactions on Multimedia. (First Author)
- Feb 2025
- Our paper on bi-modality individual-aware prompt tuning has been accepted by IEEE TPAMI.
- Oct 2024
- I joined BAAI as a research intern working on embodied vision-language-action model pretraining.
- Sep 2023
- I started my Ph.D. study at the Institute of Automation, Chinese Academy of Sciences.
Lately ...
General Covariant Action Modeling
Constructing generalized manifolds for action learning through spatio-temporal decoupling.
See project ->
Lie-algebraic Action Space Tokenizer
Bridging vision-language and action manifolds via Gromov-Wasserstein alignment.
See project ->
Human to Robot for Bimanual Manipulation
Leveraging human videos to improve generalization in robotic bimanual manipulation.
See project ->