Classification of Textures using a Tactile-Enabled Finger in Dynamic Exploration Tasks

Abstract

Reproducing human-like dexterous manipulation in robots requires identifying objects and textures. In unstructured settings, robots equipped with tactile sensors can detect textures by using touch-related characteristics. The use of a tactile-enabled finger for texture categorization is investigated in this article. Four machine learning methods are used to recognize textures from the data of pressure, gravity, angular rate, magnetic field sensors embedded in the compliant structure of a multimodal tactile sensing module. Machine learning models trained on data retrieved during 2-dimensional exploration of sample textures achieved classification accuracy rates of more than 90% for all features.