Pseudo control hedging-based adaptive neural network attitude control of underwater gliders

Underwater gliders are widely used in ocean survey operations due to their unique capability of long-range cruising with low energy consumption. As a critical technique, accurate attitude control is challenging due to the highly nonlinear, heavily coupled system model of an underwater glider with uncertain hydrodynamic parameters and suffering from environmental disturbances. To investigate this problem, a dynamic inversion-based nonlinear model reference adaptive controller is designed in this paper. The pseudo control, as the desired acceleration, is constructed by a proportional-derivative (PD) controller and an adaptive controller, and a single-hidden layer (SHL) neural network is employed to compensate for dynamic uncertainties of gliders. The actual control signal is derived by the pseudo control through the dynamic inversion of the approximate model. The attitude tracking error is proved to be ultimately uniformly bounded using a Lyapunov-based method. To avoid the possible failure of neural network training caused by actuator saturation and inner dynamics, a pseudo control hedging signal is added to the reference model as the amount of the pseudo control that cannot be achieved due to actuator saturation and dynamics. Finally, the diving and surfacing maneuvering of a glider is simulated to validate the effectiveness of the proposed adaptive neural network attitude controller.