Using Neural Networks to Model Hysteretic Kinematics in Tendon-Actuated Continuum Robots

Abstract-- The ability to accurately model mechanical hysteretic behavior in tendon-actuated continuum robots using deep learning approaches is a growing area of interest. In this paper, we investigate the hysteretic response of two types of tendon-actuated continuum robots and, ultimately, compare three types of neural network modeling approaches with both forward and inverse kinematic mappings: feedforward neural network (FNN), FNN with a history input buffer, and long short-term memory (LSTM) network. We seek to determine which model best captures temporal dependent behavior. We find that, depending on the robot's design, choosing different In contrast, the modeling of hysteretic effects has received much I. INTRODUCTION While hysteresis models such as the Preisach Since continuum robots produce a workspace through flexure and Bouc-Wen models [10] have been developed explicitly of their components, modeling their kinematics is substantially to reproduce hysteretic effects, it remains challenging to more complex than for robots comprised of rigid links estimate model parameters based on data sets [11]. Furthermore, since the flexure depends With the explosion of interest in deep learning, neural on the robot design, the modeling equations vary with robot networks are being applied as an alternative technique to type, e.g., concentric tube robots [1] versus tendon-actuated mechanics-based modeling of continuum robot kinematics robots (Figure 1) [2].