Leaning Impedance Distribution of Object from Images Using Fully Convolutional Neural Networks

Abstract

Robots have been introduced into industrial factory automation. It is necessary to consider interactions between the robots and environments to expand executable tasks of the robots. In the interaction, impedance is an essential factor for the robot to contact with the environment, whereas the impedance is unobservable without contact. In this study, we introduce a concept of affordance for impedance estimation without contact. We propose the impedance estimation method from an RGB image input using deep learning. In this paper, we show that the proposed method can extract pixels corresponding to sponges with its impedance composed of stiffness and viscosity, including the distribution of the impedance. We conducted the experiments to validate the proposed method.