A tensor-based approach to touch modality classification by using machine learning

Abstract

The use of piezoelectric sensor arrays to measure contact forces has been extensively studied in connection to robotics. In this research, Polyvinylidene Fluoride (PVDF) has been used for direct measurement of the mechanical stress and large bandwidth electromechanical transduction. Additionally, a machine learning algorithm has been specifically designed to deal with the inherent tensor morphology of raw tactile data. An experiment involving 70 participants has been organized to collect the output signals under different modalities of touch. The proposed pattern-recognition system showed good accuracy in performing touch classification in a three-class classification experiment, opening interesting scenarios for the application of tensor-based models to support human-robot interactions.