Dynamic model-based path planning optimization and control for USV in inland waterways

Abstract

The aim of this thesis study is to apply the Guidance, Navigation, and Control (GNC) system for autonomous marine technology to a marine vehicle in the form of a catamaran with a differential drive system. The Model Predictive Controller (MPC), which is a model-based control approach, is used for controlling the system. Within the scope of the study, a simulator including the dynamic model of Otter USV, a vehicle owned by Maritime Robotic AS company, has been utilized. Firstly, within the scope of the study, the system model of the used vehicle has been defined, and the components of the vehicle's mathematical model have been described. In order to use it in the model-based controller, a system identification approach has been employed to determine the coefficients of the vehicle's mathematical model. For system identification, various maneuvers were performed on the vehicle's simulator model, and navigation data was collected. Using this collected navigation data, the values of the parameters of the vehicle's dynamic model were determined through the non-linear least square method. Secondly, the development of the guidance system for the vehicle has been carried out. A system has been developed that can reach the target point while avoiding collisions in an environment with static obstacles, taking into account all input and differential constraints of the vessel model. For global path planning, a method called Kinodynamic RRT has been developed, which plans the path by considering the dynamics of the vehicle. In addition, utilizing the optimal control problem approach, an optimization-baed path planning has been performed. To enable the vehicle to follow the generated overall path, the implementation of a path following algorithm called Line of Sight (LOS) has been applied to this system. In this study, since the focus was on the realization of vehicle trajectory tracking in narrow waterways, different approaches were taken in the calculation of the lookahead distance in this method to improve the performance of the classical LOS method. The following of the generated path by the guidance system of the vehicle and the following of the reference state values generated for path following have been performed using a nonlinear model predictive control (NMPC) system. To find the optimum values of the parameters that affect the performance of the controller, the performance of the controller has been tested in different scenarios, and the most suitable values have been determined. In order to create this model-based controller and to have a suitable software architecture for real-time requirements, a fast solution method was needed, so the algorithm was developed in the CasADi optimization framework. To observe the performance difference between the developed controller and conventional control methods, a suitable PID controller has been developed for the Otter USV. The performance of these two controllers has been compared in scenarios with and without disturbances.