Online Interaction-Based Property Estimation of Objects

Abstract

The goal of this thesis is to develop online model-based parameter estimation techniques for robotics towards extracting geometrical or physical properties of an unknown (or partially unknown) dynamic environment through a tactile interaction with it. The thesis focuses on a one-dimensional case of a rigid movable object, not yet studied in the literature with parametric model-based techniques. The aim is to estimate through a tactile interaction the mass of the object and the friction between the object and the (unknown) surface where it is placed. The thesis focus on the use of Recursive Least-Squares algorithms. An explicit derivation, in a matrix form, of these algorithm is presented. The persistent excitation problem is taken under analysis since it is one of the key element for a correct parameter estimation. The thesis brings under study a novel simulated experiment where a robot finger pushes and excites the dynamics of a rigid movable mass in order to estimate the properties listed above. Because of the mobility of the object, signal discontinuities (during the pushing action) are the challenges of the estimation strategy. Possible solutions are implemented and results are discussed.