Investigating Neural Network Modeling of Hysteretic Forward Kinematics in Tendon-Actuated Continuum Robots

While mechanics-based models of continuum robots are intuitive and easy to interpret, they often neglect complex factors like non-linear elasticity, mechanical hysteresis, and friction, which are crucial for accurate modeling [1]. For instance, variations in joint space paths can significantly impact the positioning accuracy in continuum robots [2]. While nonlinearities can be addressed by using functional approximations and non-parametric models, modeling hysteretic effects is relatively unexplored. Neu- ral networks, such as Feed-forward Neural Networks (FNNs) and Recurrent Neural Networks (RNNs) like Long Short-Term Memory (LSTM) models, are adept at handling the nonlinear and hysteretic aspects neglected in traditional models [3]. This paper presents a comparison of three neural networks for modeling the hysteresis in the forward kine- matics of a tendon-actuated cardiac catheter. Training factors including the selection of input trajectories and the loss function are also investigated.