Shape Reconstruction of Soft-Body Manipulator: A Learning-Based Approach

Abstract

This work explores the use of machine learning to model the curvature of a soft-body continuum robot. Because of their compliant structures, such robots are subject to strains and deformations that are uncharacteristic of their rigid-body counterparts, giving rise to infinite degrees of freedom. Traditional modelling approaches as applied to rigid manipulators – based on Euler-Bernoulli beam theory – are therefore not quite adequate to the task of modelling soft continuum manipulators. Equally, most alternative approaches that have been tried are predicated on the constant curvature assumption, which suffers from limiting assumptions. To enhance model flexibility, we apply a Bayesian learning technique, namely the Gaussian process, for interpolating soft-robot shape from sparse data.