A Handheld Forceps-Like Tracker for Learning Robotic Surgical Tasks from Demonstration

This paper presents a novel approach to collecting demonstrations of surgical procedures that can be used to transfer skills from experts to robots using standard learning from demonstration. We use a handheld forceps-like magnetic tracker similar to the grasper of an Intuitive Surgical Inc. Large Needle Driver to record the demonstrations. Our approach does not depend on a surgical robot for the demonstration phase. We show that the resulting demonstrations can be executed on a surgical robot system by encoding them as Dynamic Movement Primitives. Our evaluation shows that users are quicker at demonstrating trajectories with the device rather than by teleoperating the robot, at the expense of slightly more noise due to hand tremors.