Development of Pneumatic Artificial Muscle Manipulator with Feedforward and Feedback Control

Abstract

A McKibben artificial muscle is a good candidate as an actuator of man-machine systems although it has quite slow response. This study proposes a control method, which integrates feedforward control and feedback control, to achieve rapid response. The feedforward input is given from a database acquired by global identification since a robot arm with McKibben artificial muscles has a complicated model. An identification error is mainly caused by mechanical hysteresis on the McKibben artificial muscle. Therefore, a technical solution to reduce the effect of mechanical hysteresis is introduced. Experimental results show that the convergence performance is much improved by the proposed method. Input shaping by a sin curve reduces oscillation and further improves the convergence performance. The main issue of the method is that it takes much effort for the global identification on many sample points. Hence, experimental results with reduced number of identification points are shown and their control performances are compared.