Global Adaptive Neural Network Control of Underactuated Autonomous Underwater Vehicles with Parametric Modeling Uncertainty

This paper focuses on global adaptive neural network control for a class of underactuated autonomous underwater vehicles in the presence of possibly large modeling parametric uncertainty. As the control inputs cannot directly act in the sway and heave directions, two virtual velocities defined here, plus three actual control actions provided by the thrusters and rudders, are used to achieve the convergence of the system errors to around zero. Motivated by real-time characteristics in the trajectory tracking, the proposed controller presents a significant advantage because it contains only one adaptive parameter to be updated online rather than the neural network weights. In addition, we also consider the practical situation that the velocities of the vehicle may experience sharp speed jumps when the position tracking errors initially change suddenly, which always results in thruster saturation. The biologically inspired model is introduced to smooth the virtual velocity commands such that the vehicle satisfies the control input and velocity constraints. Finally, comparison simulations are given to show the effectiveness of the proposed scheme.