Radhen Patel, Nancy Ouyang, Branden Romero, Edward Adelson
International Symposium on Experimental Robotics, 2020

GelSights are visual-tactile sensors that give robots a sense of touch by capturing high-resolution surface geometry via a camera embedded in a soft gel fingertip. This work designed a novel miniature wedge-shaped GelSight — the Digger Finger — customized for a specific application: identifying objects buried underground in sand and other granular media.

I helped prototype the physical sensor design and developed the initial shape detection and annotation algorithm, which used a YOLO-OBB model to classify objects from the tactile images captured while the finger digs through granular material.

The project was featured on MIT News in May 2021.