Probabilistic Identification of Unmanned Surface Vehicles Using Efficient Gaussian Processes with Uncertainty Propagation

Research on unmanned surface vehicles (USVs) has become the focus of people's attention with the increasing demand of marine activities such as ocean surveillance, search, rescue and military operations. The motion simulation, controller design and dynamics analysis of a USV all need an accurate ship maneuvering model. However, traditional ship maneuvering models and modeling methods are not completely suitable for various USVs with different shapes. Here, we present an efficient probabilistic identification method for USVs to address this problem based on Gaussian process (GP). A moment-matching-based approximation approach is used to consider the uncertainty propagation of GP in multi-step prediction. The sparse technique is introduced in GP to improve computing efficiency. The proposed identification and prediction scheme is verified by the USV experimental data from Hamburg Ship Model Basin. The results demonstrate that the developed method is a powerful and fast probabilistic identification tool for ship dynamic systems.