Physical Reservoir Computing in Tensegrity with Structural Softness and Ground Collision Dynamics

Abstract

This paper describes the effects of body structures and environments around the bodies on their computational abilities. This type of computation outsourced for body structures is called morphological computation. Physical reservoir computing, which uses the body itself as a neural network-like system, is one approach to this computation. In this research, a tensegrity-like structure, designed by the compressive elements and the tensile elements that are made of springs, is used as a body structure. We demonstrated that the structure has a higher computational ability when the structure is easier to move with a smaller spring constant, or when the structure's input amplitude was set to be smaller. In addition, the environments around the body were shown to have large effects on the computational ability. The results of these previous experiments were summarized into four features. First, the softer body has a higher computational ability. Second, the input amplitude can control the memories of the system. Third, the best input position exists which maximizes the computational ability. Fourth, the environments around the structure have effects on the system and its computational ability. Furthermore, this paper suggests that a parallel link structure (i.e., closed-loop link structure) is a prerequisite for physical reservoir computing, and tensegrity structures have ability to learn different functions at the same time.