Learning a Library of Surgical Manipulation Skills for Robotic Surgery
Keywords: robotic manipulation, surgery
TL;DR: Deep imitation learning to learn a set of contact-rich surgical manipulation skills
Abstract: In this abstract, we present preliminary progress towards learning a library of surgical manipulation skills using the da Vinci Research Kit (dVRK) system. Such a library of skills may include picking-up various surgical tools, driving a needle into tissue, tying a suture knot, or cutting tissue, among many others. These skills may be composed in various ways to accomplish a long-horizon surgical task such as suturing, and each skill may be resused in other downstream applications in various novel settings. Our goal is to learn these skills using one of the most effective tools for robot learning, deep imitation learning. Initial results on two tasks are demonstrated: 1) picking-up a suture needle and 2) picking-up a suture needle and handing the needle over to the opposing arm. Promising initial results show that learning such diverse skills may be possible in simple settings. However, further innovation is necessary in adapting these systems to work in truly novel scenarios or even learn to recover from simple mistakes. Towards this end, a few points of improvement and promising future directions are discussed.
Submission Number: 11
Loading