Recognition of Contact States in Robotized Assembly Using Wavelet Transform and Support Vector Machines

Abstract

This paper presents a method for recognition of contact states in robotized assembly. As a basic feature a generalized contact force between parts is chosen. Recognition machine teaching is carried out off line based on quasi-static contact force model. Offline features extraction using Discrete Wavelet Transform and classification using Support Vector Machines is carried out. From results of Support Vector Machines classification, fuzzy rules are extracted and Fuzzy Inference Machine for on line recognition of contact states is generated. Thus obtained fuzzy inference machine and features are used for recognition of contact states on-line. Proposed method is elaborated and tested in intensive real world experiments using an example of cylindrical peg into hole part mating.