MEAM Seminar: “Towards Understanding Tactile Sensing Across the Robot Manipulation Stack”

Humans are exceptionally skilled at manipulating a diverse range of objects, apparent from the order of magnitude difference in sizes, weight distributions, stiffnesses, and geometries of items we use on a daily basis. For robots to seamlessly integrate into a world centered around human dexterity, they must also possess comparable manipulation abilities. My research aims to address this challenge by exploring and understanding how tactile sensing can contribute to robot manipulation across the stack in sensor design, perception, and control. First, we leverage a selectively transmissive soft membrane for a novel tactile and proximity sensor that offers uncompromised spatial resolution in both modalities. Then, we explore how to fuse the tactile and proximity modalities to extract contact patches from extreme deformations of the soft sensing surface, extending functionality to new deformation regimes. Finally, I will conclude with our work on an RL-tractable tactile skin model that enables zero-shot sim2real transfer of ternary shear and binary normal forces, which we use to develop an RL policy for dexterous in-hand translation.