Numerical Methods for Model Predictive Control of Autonomous Surface Vehicles

Autonomous surface vehicles (ASVs) are seeing growing use in fields such as environmental surveys, weather data collection, military, and water transport. Given the high nonlinearity of ASVs, traditional linearized model predictive control (LMPC) struggles when the ASV veers from its design point. Nonlinear model predictive control (NMPC) is becoming a favored choice for these nonlinear systems, especially with the rise of powerful computing resources. Yet, the extensive computational demand, largely from intricate Hessian matrix evaluations and the Newton method, restricts its wider application. In this paper, we present an NMPC program harnessing a symbolic computation approach for the Hessian matrix. A central highlight of our work is the evaluation of three distinct numerical methods, each applying a different level of Hessian matrix simplification, for the task of two-dimensional ASV trajectory tracking.