Learning Motion and Stiffness Behaviours from Demonstrations for a Blunt Dissection Task

Abstract

Blunt Dissection is the task of exposing a structure of interest (e.g. an artery) by the separation of the surrounding tissue. This particular type of dissection is blunt, which means that no sharp instruments are used, deeming the method more gentle with a lower risk of atraumatic injuries. Blunt dissection is regularly performed during Laparoscopic Cholecystectomy, e.g. when the surgeon exposes the Calot triangle. The task is often repetitive, in the sense that it consists of a set of common manoeuvres that repeat over the course of the operation, yet it is time consuming and must be performed with care since otherwise complications could occur. In this work, we aim to develop a robotic solution that can automate the task to some extent. To this end, Learning from Demonstrations (LfD) is an efficient method that can be used instead of manual programming. Recent works in the surgical robotics domain have attempted to use LfD to great effect in tasks such as tying knots, suturing and cutting. On the other hand, there has been less work on how to introduce robotics into blunt dissection. This was only considered in, yet the authors used conventional programming to execute the task. Furthermore, no effort has been put on shaping the robot behaviour appropriately during contact, via e.g. the modulation of robot force and impedance. In this master thesis, we try to take a first step towards including LfD into a robotic blunt dissection task, in order to learn appropriate strategies to perform the task.