Accelerated Dual Neural Network Controller for Visual Servoing of Flexible Endoscopic Robot With Tracking Error, Joint Motion, and RCM Constraints

Aiming at the requirements for endoscope in minimally invasive surgery, a dual neural network (DNN) controller is designed for a flexible endoscopic robot (FER) with 10 DOF. First, the FER's kinematics model with remote center of motion (RCM) constraints is established. Then, a quadratic programming control scheme involving the tracking error, joint motion, and RCM constraints of the FER is proposed. A DNN solver of the control scheme, which is activated by the sum of linear and sign-bi-power activation function with adjustable parameters (LSB-AF-AP), is designed, and its convergence rate (CR) and antinoise ability can be improved via adjusting the parameters of LSB-AF-AP. The DNN can converge rapidly in finite time and it is proved by utilizing the Lyapunov theory. Compared with the previous linear AFs applied to the DNN, theoretical analysis indicates that the DNN activated by the LSB-AF-AP has an accelerated CR. Meanwhile, it is proved that the solution obtained by using the designed DNN is the optimal solution of the control scheme. Finally, simulations using the ROS and Gazebo, and an experiment performed on a real FER are conducted. The simulative and experimental results verify that the control scheme can complete the target tracking task well, and the CR and antinoise ability of the algorithm can be further improved via adjusting the parameters of LSB-AF-AP.

Automated summary

This study focuses on the requirements for endoscope in minimally invasive surgery and proposes a dual neural network (DNN) controller for a flexible endoscopic robot (FER) with 10 DOF. A control scheme involving tracking error, joint motion, and RCM constraints is established and a DNN solver with adjustable parameters is designed to improve convergence rate and antinoise ability. Simulations and experiments verify the effectiveness of the control scheme.