Hysteresis modeling and compensation of a pneumatic end-effector based on Gaussian process regression

Abstract This paper proposes a data-driven statistical learning method to identify the force-pressure hysteresis of a pneumatic end-effector based on Gaussian process regression (GPR). Given the actual complex characteristics of the hysteresis phenomena, GPR is used to establish the relationship between the input and output variables of the hysteresis as a first-order nonlinear differential equation ignoring the high-order dynamics without specifying any hyperparameters or considering the special features of hysteresis loops. The inverse hysteresis model can be derived directly. The parameters of the GPR are determined by choosing low-frequency triangle-wave pressure excitations as the training set, and then the prediction performance of the present model is tested under different types, amplitudes, and period conditions of pressure signals as the verification sets. Compared with two types of modified Prandtl-Ishlinskii model, the proposed model achieves better accuracy in both training and verification sets. Based on the inverse hysteresis feedforward compensator, comparative experiments of the force-tracking control are conducted in the form of the open-loop and closed-loop controllers, of which the results indicate the effectiveness and superiority of the proposed model.