Online Dual Neural Network Receding-Horizon Tracking Control for Magnetic Actuated Endoscopic Robots

Magnetic actuated endoscopic robots (MAER) are more and more favored in minimally invasive surgery (MIS) because they can get rid of the shackles of the mechanical transmission medium and can be controlled remotely by force or torque. However, it is still a challenging research topic to realize the safe and accurate control of MAERs due to their high nonlinearity and safety requirements. Therefore, to realize the safe and autonomous tracking control of MAER and reduce the mental burden of the surgeons in MIS, this article proposes an online dual neural network receding-horizon tracking controller (ODNN-RHTC) for MAERs by comprehensively considering the MAER's tracking performance and physical constraints such as joint acceleration, velocity, and position, etc. First, the MAER's motion/vision coupling model and the prediction model are established. Next, a new visual tracking controller for MAERs is designed by comprehensively considering the MAER's tracking performance and physical constraints, and a novel dual neural network (DNN) method is introduced to solve the control scheme quickly. The fast convergence of the control system, whose Lyapunov candidate functions are continuous but not locally Lipschitz, is rigorously proved by using Zubov's theorem. Finally, experimental and numerical simulation results show that the proposed methods can well control the MAER tracking target task. Compared with the existing technologies, the MAER's tracking trajectory is smoother and the trajectory fluctuation is smaller by using the proposed neural network method, and the MAER's stability and tracking accuracy can be further improved by using the proposed ODNN-RHTC.