4-DOF Visual Servoing of a Robotic Flexible Endoscope With a Predefined-Time Convergent and Noise-Immune Adaptive Neural Network

Abstract

Endoscopes provide views inside the patient's body during minimally invasive surgery. Although various robotic endoscopes have been developed to reduce surgeons' workload in manual endoscope operations, autonomous endoscope manipulation remains challenging due to the misorientation effect and different disturbances. In this work, we developed an intelligent endoscope system to steer the surgical view automatically. To keep the target (i.e., the tip of an instrument) at the center of the camera view with a suitable size and orientation, an image moment-based 4-degree-of-freedom (DOF) visual servoing method is implemented. We propose an error learning-based sliding mode controller to realize fast and smooth error convergence. It is specially constructed to improve convergence rate without causing undesirable system chattering. Moreover, the kinematic modeling of the endoscope results in a quadratic programming problem, which is solved by a novel adaptive noise-immune zeroing neural network accelerated to predefined-time convergence by a newly constructed activation function. The experiments show that the proposed control strategy guarantees a superior convergence rate and robustness compared with existing methods. Lab tests show the application potential of the proposed endoscope system in clinical practice.