Path following of autonomous underwater vehicles in the presence of unknown disturbances

Abstract

First of all, this study can be briefly explained as follows. It is about the path following of autonomous underwater vehicles (AUVs) when exposed to external environmental factors such as currents in predefined paths. In this study, algorithms were developed to solve the relevant problem based on the dynamic model of the Remus-100 autonomous underwater vehicle (AUV). In this context, the performances of three guidance methods are compared to solve the problem, these guidance methods are line of sight (LOS) guidance, integrated line of sight(ILOS) guidance, and adaptive line of sight(ALOS) guidance. Additionally, a linear quadratic regulator (LQR) controller was designed to control the AUV. The states of the autonomous underwater vehicle, which is the input of the autopilot, were estimated using the Extended Kalman filter(EKF). The reason for this study is that AUVs are being used a lot today and path following is critical in these vehicles. To briefly mention the usage areas of AUVs, these technologies are used in scientific research about oceanography, marine biology, mapping the seafloor, environmental monitoring about pollution levels, and geology research of the seafloor. In addition, AUVs may be used in military and defense for underwater surveillance and anti-submarine warfare. These vehicles are used in oil and gas exploration to inspect and map underwater pipelines and infrastructure. Additionally, they are operated in search and rescue operations to find missing ships and airplanes. Finally, they can be used in commercial fishing to help fishermen determine the best fishing locations and underwater archaeology to explore and map the underwater archaeological sites. These usage areas bring with them an important problem that needs to be solved with high accuracy, and this is the path-following problem. These vehicles must follow predefined paths with high accuracy, even with limited position information and in situations where they are exposed to current disturbance. Otherwise, these vehicles cannot successfully perform the defined tasks. In this study, a solution to the 2-D path following problem is presented. The sub-topics focused on solving the problem are the design of the navigation, guidance, and autopilot algorithms of theAUV. Briefly, the navigation algorithm is the algorithm that calculates where and in what orientation the underwater vehicle is at the moment. The guidance algorithm calculates the heading angle that the underwater vehicle should apply to follow the path defined in 2-D. The autopilot algorithm ensures that the underwater vehicle is kept in the desired orientation. As a result, these algorithms must work properly simultaneously for the AUV to follow a defined path. When first focusing on the navigation algorithm, its output is also the reference input of the autopilots of underwater systems operating in real-time. In this context, these inputs need to be estimated and measured accurately. In addition, if a state that is not directly measured is to be used as an input to the autopilot, this state needs to be estimated. In addition, for the autopilot to produce proper outputs, the noise level of noisy measurement data produced from sensor models must be reduced, and this can be made possible with EKF. In addition, this AUV is exposed to the effects of underwater currents during its operations. This disturbance can create major problems in docking missions where location accuracy is important and in long-term missions such as military applications or mapping. Therefore, these disturbance components need to be estimated. In this context, using the EKF, the AUV's position components, velocity components, acceleration components, Euler angles, angular rates, and sea current components will be estimated. The sensors to be used when making these predictions are an ultra short base line (USBL) acoustic positioning system, inertial measurement unit (IMU), depth sensor, compass, and Doppler velocity log (DVL). The reason for using EKF is that it can be easily applied to nonlinear systems, with its prediction accuracy, adaptability, and simplicity. In this study, autopilots were designed using the LQR technique. The LQR method offers a robust framework for control design, optimizing control inputs to minimize a specified cost function while considering system dynamics and constraints. This study applies LQR to the nonlinear dynamics of AUV by first linearizing the model using numerical techniques. This allows for the formulation of the control problem in a linear framework, where LQR can effectively compute feedback gains to steer the system towards desired states. The primary advantage of utilizing LQR lies in its ability to provide stable and efficient control across a range of operating conditions. By optimizing a quadratic performance index, LQR ensures that the AUV achieves the desired performance while adhering to system constraints. Furthermore, the straightforward implementation process of LQR simplifies the design and facilitates real-time deployment on AUV platforms. This enables the development of robust autopilot systems capable of navigating complex underwater environments. As a result, when the LQR controller was tested under various conditions, it was seen that it worked with high performance and low control effort even in disturbed environments. Finally, it is briefly mentioned that the guidance methods used in the path following strategy, LOS, ILOS, and ALOS guidance methods were used in this study. These guidance algorithms take as input the waypoints of the path they need to follow and the currently calculated position of the AUV. As output, they calculate the heading angle the AUV should apply to follow this path. The calculated heading command gives satisfactory results in path following in 2-D for autonomous underwater vehicles. In these methods, the location information given as input is the location information that is the output of EKF. In addition, the vehicle's states used as reference in autopilots are also components calculated by EKF. The dynamic model output of the vehicle was not used as input in the autopilot and guidance algorithms, these inputs were the outputs of the EKF. If dynamic model outputs were used, reliable results could not be obtained in testing the usability of these algorithms in real life. Therefore, the outputs of the dynamic model were used as ground truth and the actual states of the vehicle were converted into sensor outputs using sensor models. States that were not directly measured by sensors were estimated with EKF. With this approach, an attempt was made to simulate a real AUV operation. During the testing phase, 4 different paths were defined and tested. The first of these paths is a straight path, the second is a circular path, and the third is a path where both a straight route and turning maneuvers are made. The last path is sinusoidal. While testing these paths, current disturbance was added to the environment, and in this way, it was aimed to test situations where the vehicle was exposed to current disturbance. In this context, tests were carried out with 3 current speed components and 4 current direction components. These tests compared the performance of 3 different guidance algorithms in different current components and path definitions. The cross-track error component, which calculates the distance from the path, was used as the comparison parameter. Finally, it was concluded that some guidance laws are more advantageous in certain situations. Briefly speaking, ILOS guidance law shows superior performance in circular and sinusoidal paths, which shows that ILOS guidance gives better results in paths containing curvature. It has been observed that LOS guidance giving good results on straight and semi-rectangular paths, which shows that LOS guidance gives better results on flat paths compared to other methods. In this case, it can be concluded that since the defined paths are predetermined paths, if there are straight lines in the planned path, LOS guidance can be used during this process, and ILOS guidance can be used actively in the parts of the path that contain curvatures, that is, in places where there are turns and maneuvers. This ultimately enables maximum path-following performance to be achieved in all designed paths. Since one guidance law performs better than the other according to the path features, automatic guidance law switching can be performed in AUV tasks by establishing a simple switch-case logic and achieving minimum cross-track error. One of the most important results of this study can be explained as follows: Depending on the characteristics of the defined path, the guidance law may change during the mission because the currently used guidance laws calculate the yaw angle command as a result of instantaneous calculations, that is, they do not contain any retrospective terms. Therefore, transitions between these algorithms do not create any discontinuity. Finally, all of the algorithms developed within the scope of this study have been developed to be used in real underwater vehicles, so I believe that the algorithms to be developed for Remus-100 in the future can be used as a reference source.