Transferring autonomous reaching and targeting behaviors for cable-driven robots in minimally invasive surgery

Abstract

Cable-driven mechanisms have been widely used as compliant actuators in surgical robots over last decades due to their superior performance of dexterous operation in confined workspace. And minimally invasive surgery (MIS) is one of the most important applications of such systems. In MIS, point to point reaching and targeting behavior is the most fundamental movement primitive, typical examples include leading the robot tool to the target lesions. Currently, the motion control of such surgical robots in MIS are realized by clinical staff with haptic devices. However, due to the totally different configurations of the robot and the haptic device, and the friction loss, viscoelasticity, hysteresis and nonstationary behaviors inherently in the cable-driven mechanisms, the teleoperation procedure is difficult and uncomfortable, and may cause severe fatigues of the surgeons. In this work, a novel motion control approach, learning from demonstration (LfD), is used to transfer autonomous reaching and targeting skills from human experts to a cable-driven surgical robot. A modulated first order dynamical systems model is used to encode human demonstrations and generate executable paths for the robot. Experiments have been performed on a seven degrees-of-freedom KUKA LBR robot and a three degrees-of-freedom tendon-driven serpentine manipulator (TSM) to validate the proposed methods.