Doctoral Thesis: Tactile Sensing for Contact-Rich Dexterous Robotic Manipulation

By: Megha Tippur

Presenter’s Affiliation: CSAIL

Thesis Supervisor: Edward Adelson

Details

• Date: Thursday, May 7
• Time: 1:00 pm - 3:00 pm
• Location: Hewlett Room 32-G882 (Stata Center)

Additional Location Details:

Abstract: Achieving human-level dexterity in robotic manipulation will require systems that can perceive and reason about physical contact. While tactile sensing has long demonstrated its value in manipulation, its full potential remains largely underexplored within modern robot policy learning. This thesis explores the role of camera-based tactile sensing in contact-rich dexterous manipulation. We present novel curved tactile sensor designs (GelSight360 and RainbowSight) that enable high-resolution sensing at fingertip scale for integration onto dexterous robotic hands. We also demonstrate the value of tactile feedback through two manipulation systems: a bimanual visuotactile system for folding and hanging garments from occluded configurations and a system for localizing and retrieving objects through touch alone in vision-free environments. Together, this work takes steps toward making camera-based tactile sensing a practical and scalable component of dexterous robotic manipulation.

Host

• Megha Tippur
• Email: mhtippur@mit.edu