Object Recognition Using the Deformation Dynamics of Polyurethane Foam

Abstract

For the purpose of developing a versatile skin for soft robots, we propose a sensing method based on physical reservoir computing (PRC) in which we applied conductive polyurethane foam (CPF) to classify the state of a contacted object. Inspired by PRC, we exploited the natural deformation dynamics of CPF as an information processing device. By monitoring the deformation through the resistance change of the cellular foam, we demonstrate that the foam's deformation dynamics are innately capable of recognizing the different states of objects (such as shape, angle, and position). Thus, the foam acts as a natural classifier that facilitates separability through its high-dimensional dynamics. This feature is useful in practical applications and would reduce the electrical components and power consumption required to work with other systems. In this study, we examined proof-of-concept experiments using simple object conditions: three object shapes (circle, triangle, and square) at different rotations and positions. As a result, high classification accuracy was demonstrated in a number of experiments, and the possibility for enhancing the method's generalization capability was investigated. Our approach is potentially applicable to not only the foam but also to other various soft materials with large internal degrees of freedom, suggesting its universality to soft robotic platforms in general.