Speaker: Xinyu Zhang

Bio

Location : Room 402, 4th floor, 1 Spring Street, Downtown New Brunswick

Committee: 

Professor Abdeslam Boularias

Professor Kostas Bekris

Professor Chengzhi Mao

Professor Ruoshi Liu (external)

Event Type: PhD Defense

Abstract: World models have received significant research attention as a pathway toward generalizable robot learning. In this talk, I first discuss how to design efficient and accurate world models, and explore their applications in robotics, based on my recent work, Learning Visual Feature-based World Models via Residual Latent Action. Next, I introduce a new paradigm for dexterous hand data collection based on my CVPR 2026 Highlight paper, Glove2Hand: Synthesizing Natural Hand-Object Interaction from Multi-Modal Sensing Gloves, and briefly present a novel policy design that treats robot actions as language (RA-L 2025). Finally, I outline my research proposal to scale up data and learning for robotic manipulation.

Organization: 

Contact  Professor Abdeslam Boularias

Zoom Link: https://rutgers.zoom.us/j/9331476253?pwd=ckZKVDVOWkY3Tm9DZzBCNTA1b3hhQT09