Adaptive model predictive control of unmanned underwater vehicles

Abstract

Unmanned underwater vehicles (UUVs) are increasingly essential for a variety of underwater tasks, with a primary emphasis on achieving autonomy. Autonomy is critical for enhancing safety, flexibility, expanding operational capabilities, and reducing expenses. However, developing effective and robust control algorithms for UUVs is challenging due to nonlinear dynamics, uncertainties, constraints, and environmental disturbances. Model Predictive Control (MPC) is a well-established technique for UUV control, with the key challenge lying in obtaining precise prediction models to enhance controller performance. This thesis primarily introduces two enhanced MPC approaches that enable a UUV with partially unknown dynamics to autonomously navigate complex marine environments. The first approach is a Disturbance Observer-based MPC (DOBMPC). The DOBMPC integrates unmodeled dynamics and environmental disturbances into a disturbance term estimated by an Extended Active Observer (EAOB). While the proposed DOBMPC effectively enhances disturbance rejection, the thesis also addresses handling unknown dynamics more meticulously. Subsequently, the second proposed control method is an Adaptive MPC with an online system identification algorithm. This online system identification method is constructed using an Extended Active Observer (EAOB) and the Recursive Least Squares with Variable Forgetting Factor (RLS-VFF) algorithm to estimate environmental disturbances and identify uncertain hydrodynamic parameters. The estimated disturbances and parameters are continuously updated in the MPC's prediction model to generate optimal control inputs based on real-time environmental and vehicle conditions. These proposed methodologies are validated within the Gazebo and Robot Operating System (ROS) simulation environment, illustrating their effectiveness in managing uncertainties and disturbances for UUV control.