Path Planning and Nonlinear Model Selective Control Using Neural Networks: Precision Maneuvering of Unmanned Surface Vehicles

The work resulted in a paper published as

David Cartes, C. Nataraj, John Metzer, Steve Castelin, 2005, Intelligent Ships Symposium, May.


In this paper we present the hydrodynamic and control challenges associated with the navigation objective of highly robust, yet versatile USV, and provide one possible solution for precision navigation. Our discussion will focus on a medium sized utility vehicle, the eleven meter rigid hull inflatable boats (RHIB) already in the Navy Inventory. In this paper we propose a path planning and maneuvering architecture. It includes a nonlinear model selective control that uses a neural network to select from a predetermined database of performance models the most appropriate model, such that the model input/output relationship matches that of the actual USV. This appropriate nonlinear model is then used in a feedback linearization controller design, which then services the path planning algorithm. This technique has faster response than nonlinear adaptive control. The model database contains a set of reduced order, nonlinear, non-physical, parametric equations that locally represent the actual complex hydrodynamic constraints of an USV over a large operational and environmental mission space. The path planning algorithm provides the precise navigational reference for the USV to support its installed mission modules. The path planning and maneuvering architecture is demonstrated in simulation. The results show that this architecture is an improvement over traditional proportional integral control techniques.