Autonomous underwater robot : vision and control

Autonomous underwater vehicles have been an important tool for many undersea applications including scientific, commercial, and military applications, such as exploring the ocean floor, studying marine life, and inspection and repairing underwater structures. Vision is an important resource that enable autonomous underwater vehicle to do several tasks. Vision is particularly useful for underwater navigation since it can provide accurate depth measurements at short ranges. This thesis is concerned with the development of a visually-guided autonomous underwater vehicle. The goals for this visually-guided robot are to be able to visually track and maintain a fixed position relative to a stationary target, and to visually track and swim after a dynamic target. These tasks, where vision is used to provide feedback to the controller, are usually called visual servo. The implementation of the system consists of the development of many modules, namely: the vehicle model and vehicle controller, the thruster model and thruster controller, and the vision system and visual servo controller. The developed vehicle model uses Euler parameters for attitude representation, and thus avoids singularities, that occurs with the use of conventional Euler angles, in the system’s kinematic equations. A computed torque controller is developed based on the knowledge of the system dynamics. The good performance of the developed computed torque controller is demonstrated through a simulation study. Thruster dynamics have been neglected in the development of the vehicle model and vehicle controller. For thrust control, a steady-state thruster model has been adopted. This thruster model is demonstrated to give good thrust estimates at steady-state as well as reasonably good thrust estimates during the transient response. The steady-state thruster model is hence used in the design and implementation of the thruster controller, as there are no raw thrust measurement data available. A nonlinear PI control using saturation type integrator has been proposed, in conjunction with the steady-state thrust model, for thrust control to overcome the integral-windup phenomenon. The resulting thrust control system shows a good performance in controlling the estimated thrust. The vision system consists of a pair of stereo cameras which, once calibrated, give accurate 3D information. Normalized cross-correlation target tracking has been developed to locate and track targets. The target position is estimated with information from the stereo cameras, and sent back to the vehicle control system. The integration of all subsystems provides visual servo control system. In the current implementation, some sensors have not yet been implemented into the system, which led to the use of a slightly modified control scheme (PI plus gravity control) instead of computed torque control. This controller receives only the visual position feedback and sensory attitude feedback. The result is a blend of a direct position-based visual servo structure for position control, and a dynamic look-andmove structure for orientation control. The direct position-based visual servo control in position control is an achievement for autonomous underwater vehicles research, since most underwater vehicles use some sort of sensory feedback other than vision for position control. The vehicle successfully performs visual servo with both stationary and moving targets. The vehicle can maintain a fixed distance to the stationary target as well as move around it. In chasing a moving target, the vehicle can visually track the target and move along with the target as it moves. In summary this thesis provides two major contributions: 1) a visual servo control system for autonomous underwater vehicle that can track and follow dynamic targets as well as keep stationary with fixed targets, and 2) a direct position-based visual servo control that uses only visual feedback for position control. Acknowledgements My two years at the Robotic System Labs will always be a wonderful experience. I have spent much of my time in the past two years in the Lab doing my first research degree in English. A lot of people have helped me trough the research. I would like to thank my supervisors David Wettergreen and Samer Abdallah, who introduced me to such a great project, Kambara the yellow submarine; who have been helping and guiding me through my research; who have been teaching and guiding me a lot in writing this thesis as well as other papers. Kambara will be an unforgettable project. I joined the group of submariners when Kambara was about one year old. It could not swim until it was almost three years old. We have spent a lot of time in get it up and running (swimming). Another thank you to my advisor Thomas Brinsmead for helping me finishing up this research and drafting my thesis. I cannot forget Peter Brown, our software engineer. Without his great effort, Kambara would still be a baby submarine that cannot swim. A big thank you to him. Thank you to our Lab leader, Alex Zelinsky, who has been waiting patiently for this project to come to the point where it is today. And thank you to everybody else in the lab who have been so kindly supportive: James Ashton who has always been a great sys. admin.; Jenni and Rita who support everyone in the lab; Chris, Matthew, Ian, Jordan, Sunil, and Terence who were once on the same yellow boat. Thank you to all my Thai friends in Canberra who have always been there when I needed company and friendship, in particular, P’Ying, Om, Muay, Tip, Tui, Pom, Chap, Praew, and N’Jayne. Lastly, to my parents and family, I would like to thank them for such a great support.