Robust Control of Underwater Drones by Extended Model Predictive Control and Sliding Innovation Filter

Abstract

In recent years, underwater drones have been actively developed and have been used in industrial applications such as marine debris collection, facility inspection at hydraulic power plants and sunken ship exploration. In many cases, underwater drones are remotely controlled by humans. Although in extreme environments such as caves or deep sea due to communication stability or cable length limit, remote control by humans is not available, and an autonomous control is required. However, the autonomous control of an underwater drones is difficult in an underwater environment with many problems. As typical problems in the underwater environment, there are unknown external disturbances such as waves and water currents, sensor noise caused by poor sensors and modeling errors that occur during system identification. Although Extended Model Predictive Control (EMPC) provides robust tracking control against unknown external disturbances and modeling errors, it is difficult to address sensor noise. On the other hand, well known Kalman filter can deal with sensor noise, however it is difficult to address modeling errors. In this research, we aim to develop a robust tracking control method for underwater drones with EMPC, Sliding Innovation Filter (SIF) and Bayesian Optimization (BO) against unknown external disturbances, sensor noise and modeling errors simultaneously. SIF is a robust sensor noise filter against modeling errors and has a simple control structure, requiring no assumptions. Also, BO is applied to efficiently and accurately estimate parameters that affect the estimation accuracy of SIF. The effectiveness of the proposed method is shown by numerical simulations using a model that assumes experimental device.