Three-dimensional Deformation Control System for Fabric Actuator

This paper presents a control system fora fabric type actuator consisting of 12 extra thin pneumatic artificial muscles and enabling three-dimensional movements. The basic structure of the fabric-type actuator has proposed in previous researches, however, the controller of it has designed heuristically and realized only six patterned deformations. In order to control its three-dimensional deformation, it is necessary to make a control system whose input is objective deformation. Modeling of the relation between pneumatic inputs and deformations is needed to make the control system. The general formula approaches are difficult to apply because the actuator deforms flexibly and nonlinearity. In this paper, a control system based on a black-box modeling methodology is proposed. In order to describe the appearance of three-dimensional deformation, deformation vector, a group of transition vectors on segmented areas of the actuator, is used. The controller has designed by gathering the data pair of pneumatic inputs and deformation vectors and learning relation from deformations to pneumatic inputs with a Deep Neural Network. Even there are some differences between objective and observed deformations in steady state, the errors are acceptable to control by considering the flexibility of the actuator.