A Model-Free Method-Based Shape Reconstruction for Cable-Driven Continuum Manipulator Using Artificial Neural Network

摘要

The robot-assisted natural orifice transluminal surgery (NOTES) typically involves applying the lengthy and slender continuum instruments/manipulators to get access to the target lesion sites, and then performs complex operations. However, due to the strong-nonlinearity of continuum robotic manipulators and the confined and tortuous anatomical paths, it's difficult to establish accurate inverse kinematics (IK) model to achieve precise motion control and real-time shape sensing for accurately modeling their shapes. To tackle such difficulties, a model-free method based on neural network have been proposed to solve the IK problem and reconstruct the shape of a continuum manipulator at the same time using the training results from the electromagnetic (EM) tracking approach. For the IK problem, the relationship between the tip position and the corresponding cable lengths can be learned and for the shape estimation problem, the mapping from the cable lengths to the shape of the continuum robot can be established. A dataset of 500 random continuum manipulator postures was used to train the neural network, with recorded EM sensors-tracked positions logged synchronously to the cable lengths. Experiment results show the proposed model-free method could achieve high accuracy and reliable inverse kinematics and shape reconstruction outcomes.