Co-Chairs: Dmitry Berenson and Nima Fazeli

Abstract:
My PhD research focuses on representation learning with force and touch for contact rich, dexterous robot manipulation. Recent robot learning has made major progress in vision and language, but those modalities alone cannot fully capture physical interaction in robot manipulation tasks. Tasks like cleaning, assembly, and assistive care require robots to reason about contact and force, especially in the presence of sensor noise and visual occlusion. My work develops learning based methods that integrate force and tactile sensing into robot learning and control. I build object centric multi-modal (vision and touch) representations to infer object state and contact under noise and occlusion, design contact aware planning and control frameworks with learned dynamics to regulate interaction that effectively integrated with language modality, and develop shared tactile representations that align human and robot touch signals for cross embodiment policy transfer. Overall, this dissertation studies how force, touch, and contact can improve robotic dexterity in contact rich manipulation.

Zoom:
https://umich.zoom.us/j/95094568449
Meeting ID: 950 9456 8449
Passcode: 837399