Recurrent Neural Network With Preisach Model for Configuration-Specific Hysteresis Modeling of Tendon-Sheath Mechanism

The tendon-sheath mechanism (TSM) is widely used for flexible surgical robots owing to its ability to transfer motion through complicated paths and small volumes. However, precise motion control of the robots is difficult owing to the nonlinear characteristics of the TSM due to hysteresis, friction, deformation, backlash, etc. The hysteretic behavior exhibits a certain tendency depending on the shape of the TSM. In this study, our goal is to derive a hysteresis model that applies to the arbitrary geometry of a single-curve TSM. A hysteresis model was formulated by combining the Preisach hysteresis model, which has the structure of a weighted sum, and a recurrent neural network (RNN). Hysteresis output data were experimentally gathered for several representative shapes of the TSM to train the model. During data preprocessing, the data in the section where the hysteresis characteristic was prominent was augmented. Hyperparameter tuning using the grid search method was conducted to improve the performance of the model. The results indicate that the proposed model successively learned the tendency of configuration-specific hysteresis and adapted to arbitrary shapes of a single-curve TSM, even with shapes that the model did not learn.