<title>Nonlinear optimal predictive control of rubber artificial muscle</title>

Abstract

This paper presents the use of nonlinear optimal predictive control on rubber artificial muscles. A neural network which can learn dynamic characteristics of rubber artificial muscles is employed in this control scheme in order to obtain the model. The actuator has strong nonlinear characteristics due to compression of both rubber and air. It is difficult to control with precision using the linear control theory. A suitable control method for this actuator has yet to be established. Computer simulations of a common pneumatic actuator, which is modeled quite easily, were carried out to verify the proposed control scheme. This control scheme was adapted to a practical one dimensional robot manipulator which was developed for this study. The theory was compared with PID control in order to explain availability of the theory. When desired targets are changed as a step response, this actuator can be controlled precisely in these simulations and experiments. The actuator can be controlled adaptively under changes of characteristics if the learning of the neural network continues. It is obvious from these results that the proposed control scheme is effective for rubber artificial muscle.