A System for Affective Touch in Humanoid and Social Robotics

Abstract: A system for affective touch has been constructed, in a study of humanoid and social robotics. The system detects, processes and analyses signals from touch, identifies touch types, and provides a corresponding emotional response and expression. Touch is detected on an Epi humanoid robot head through the use of conductive paint on the inside on the head shell, and the electrical signal produced is processed into a digital representation of touch. Touch types are defined and classified through the application of machine learning. Approximate touches are applied to the head, including a variation in the areas of touch, and training provides a classification of ten touch types with an accuracy above 85%. Touch types are mapped to related emotional responses, providing the basis for the selection of an eye colour expression from an Epi humanoid robot. The system is integrated with the Ikaros cognitive modelling framework and real-time interaction is made possible, enabling a dynamical and complex human-robot interaction. This further confers a consistent framework for a future experimental evaluation of the system.