Data-Driven Nonlinear Model Predictive Control of an Autonomous Uncrewed Surface Vessel

Thumbnail Image
Cooper, Michael Riley
ASV , USV , Uncrewed Surface Vessel , Data-Driven Control , Systems Control , Nonlinear Model Predictive Control , Robotics , Autonomous Vehicle , System Identification
Small surface vessels are a vital part of many Canadian and global operations including transportation, environmental surveying, and emergency response. Increasing the autonomy of these vessels could provide benefits by reducing on-board labour requirements, reducing travel time, and increasing fuel efficiency. With traditional control techniques, models for Uncrewed Surface Vessels (USVs) do not take into account evolving disturbances such as the wind, waves, and water current due to the complexity of these environmental forces. However, knowledge of these disturbances can improve controller performance and safety strategies for control. The Sparse Identification of Non-linear Dynamics (SINDy) is a data-driven model identification technique that provides human-interpretable system models in short training times using low amounts of training data. In this work, SINDy was used to create a USV model from same-day data, capturing up to date changes in the vehicle performance and environmental conditions. That model was then combined with real-time Nonlinear Model Predictive Control (NMPC) to create a data-driven control strategy for USV path following. Field experiments in the Great Cataraqui River (Kingston, Ontario) were performed with an Otter USV to compare the performance of various controllers over a predefined path, including an NMPC controller with an idealized model and a controller with a model generated using the SINDy algorithm.
External DOI