Despite recent advancements in robotic manipulation and heightened interest in pick and place applications, robots still struggle with basic manipulation tasks. With millions of industrial robots worldwide performing manipulation tasks that rely on computer vision, it is crucial to identify and improve upon weaknesses in their systems. To improve a robot’s tactile perception ability, we want to augment traditional computer vision with tactile data from a GelSight touch sensor. The first part of this project will focus on developing motion planning methods to construct a labelled database of what different objects feel like to a robot with a GelSight sensor. The second will focus on migrating from recognizing objects with computer vision to recognizing them with only the GelSight.

I am participating in SuperUROP because I am excited about being able to combine my background in both computer science and mechanical engineering to a robotics project where the software that the robot is running is equally important to how the robot is physically interacting with the world.