Learning the inverse kinematics of tendon-driven soft manipulators with K-nearest Neighbors Regression and Gaussian Mixture Regression

Due to the urgent need for Minimally Invasive Surgeries (MIS), all kinds of surgical robots have been developed and investigated intensively in last decades, which can both release the fatigues of surgeons and speed up the process of wound healing. Tendon-Driven Serpentine Manipulator (TSM) maybe among the most widely adopted and promising ones to turn robot assisted MIS into reality. But due to the high nonlinearities and model uncertainties in the TSM system, it is extremely difficult to precisely control the robot. In this paper, we develop and investigate two approaches from machine learning domain, Gaussian Mixture Regression (GMR) and K-Nearest-Neighbors Regression (KNNR), to learn the Inverse Kinematic (IK) model of our TSM robot. Then we compare the performance of GMR and KNNR with that of an IK model derived in previous literatures. Experimental results conducted on a real world TSM robot performing trajectory tracking tasks validate the superior performance of the proposed methods over traditional analytical IK models.