Stochastic model predictive control for motion control of an unmanned underwater vehicle

Abstract

Recently, nonlinear model predictive control (NMPC) has been widely used for the motion control of unmanned underwater vehicles (UUVs). However, UUVs having slow operation speeds and small control fins are considerably affected by uncertainties such as unknown environmental disturbances, modeling uncertainties, and measurement errors. To overcome the limitations of existing NMPC, which involves difficulties in directly considering uncertainty, studies on stochastic model predictive control (SMPC) have recently been conducted. Sampling-based methods have been proposed to calculate future probability distributions for nonlinear models. However, real systems require computationally efficient algorithms. Therefore, in this study, we propose an SMPC algorithm based on unscented transform. In this study, an unscented transform is used to estimate the future probability distribution of a nonlinear system and the existing probability distribution of the optimization algorithm, by using the result of the previous optimization process to achieve high computational efficiency. Chance constraints are calculated analytically through linearization. We investigate time-varying ocean currents using the proposed unscented transform-based SMPC. Through path tracking and an obstacle-avoidance simulation of a UUV, we demonstrate the performance and usefulness of the proposed algorithm.