An Integrated Accuracy Enhancement Method for Cable-driven Flexible Continuum Robot

Abstract

This paper focuses on an accuracy enhancement method for a cable-driven continuum robot with a flexible backbone by integrating a machine learning technique into the kinematic modeling. The continuum robot employs a segmented cable-driven scheme, which can be considered as a modular manipulator composed by a number of consecutive cable-driven segments. Based on the unique backbone structure which merely allows bending movements, a two-variable Product-of-Exponential (POE) formula is employed to formulate the kinematic model of the cable-driven continuum robot. However, such an analytic kinematic model is inaccurate due to the compliance of the backbone structure. In this paper, a machine learning approach based on the Gaussian Process Regression (GPR) is proposed to compensate the positioning errors resulting from the inaccurate kinematic model. Compared with other machine learning methods, GPR has the advantages of requiring less learning parameters and training points, which makes the learning process computationally efficient. In order to validate the effectiveness of the proposed method, computer simulations have been conducted. Simulation results indicate that the integrated approach can reduce the position error by 70.49% and the orientation error by 65.76%.