Neural-network-based adaptive observer design for autonomous underwater vehicle in shallow water

It is often unavailable to obtain direct measurements of the underwater vehicles´ velocities in actual implementations. A neural-network-based adaptive observer system is designed to solve this problem in this paper. Since the dynamics of autonomous underwater vehicle (AUV) are highly nonlinear nature and the hydrodynamic coefficients are difficult to be accurately estimated, a dynamic recurrent fuzzy neural network (DRFNN) is employed in the observer to estimate the unknown nonlinear characteristics in the vehicles´ dynamics. The proposed observer can estimate AUV´s low-frequency motion and slowly varying environmental disturbance from the measuring signals, which include high-frequency motion signals and the noise of sonar. The network weights adaptation law are derived from the Lyapunov stability analysis. With the Lyapunov stability theory, the convergence of these estimations is global and exponential.