An Adaptive Observer for an Underactuated Unmanned Surface Vessel

In this paper, we report an experimental evaluation of an adaptive, nonlinear observer that can handle drastic changes in the dynamics of an Unmanned Surface Vessel (USV). The observer is based on an extended Kalman filter (EKF) with a gradient descent based adaptation law with implicit regularization. The adaptive observer is compared to a traditional, non-adaptive observer using real data from on water tests with an USV. Results show that both observers perform well under nominal conditions. However, the adaptive observer outperforms its non-adaptive counterpart when the dynamics of the USV changes due to a drastic disturbance. With this, we show how an adaptive observer can be beneficial when there are changes to the underlying nominal model, or no such model exists.