Path-Planning, Guidance and Navigation Tools for Docking Underactuated AUVs
MetadataShow full item record
Summary This thesis is motivated by the growing usage of autonomous underwater vehicles (AUVs) and, particularly, how providing them with the ability to reach and dock onto underwater docking stations can positively impact and transform the way underwater missions, are done today. Enabling an AUVs to autonomously reach and connect to a docking station where it can charge batteries, report the results of a mission and receive newer ones would reduce the cost of AUV missions and increase their overall safety. Moreover, subsea infrastructures allow AUV to reside underwater and to be permanently available, further extending their uses. However, the underwater environment imposes communication and navigation challenges, making autonomous docking difficult undertaking. The actuators and sensors that are usually fitted in an AUV allow it to move, explore, and perceive the surrounding environment. Since the payload of the vehicle is often limited, the sensors and actuators are optimized for the set of tasks that the AUV is mostly used for. This can make performing a precise task, such as docking, challenging because the AUV becomes an underactuated vehicle. For instance, docking in the presence of cross-currents can become difficult since the vehicle does not have lateral thrusters to compensate for the current. Docking can also originate in the navigation, because producing reliable navigation information might not be possible if the vehicle is unable to permanently maintain the docking station inside the field-of-view of its sensors. Thus, rather than exploring the best sensor and actuator configurations to simplify a docking maneuver, the thesis aims to provide solutions that enable docking for common configurations of existing AUVs. This thesis, titled Path-Planning, Guidance, and Navigation Tools for Docking Underactuated AUVs, seeks to provide a solution to some of the challenges behind docking an AUV autonomously, and it does so by analyzing the path planning, navigation, and guidance, and how each can be designed and optimized to best serve the docking maneuver. The first part of the thesis proposes tight combinations between path planning and guidance control that are specially made for docking, and to enable the AUV to partially overcome the sensor and actuation limitations. The viability of such approaches is validated through stability proofs, numerical simulations, and/or experimental results. The second part of this thesis proposes, large datasets of sensor measurements that are obtained in the area surrounding the docking station by a system with an accurate positioning system. The data is used to train a convolutional neural network (CNN), generating an end-to-end solution that encompasses the detection of the docking station, the navigation, and the guidance in a closed solution. The trained system allows a vehicle equipped with fewer sensors to reach and dock onto the docking station. The proposed approach is experimentally validated. The last part of the thesis proposes using the probability density functions within the Bayesian filters to give an assessment of the probabilities of successfully docking adapting the controllers to the uncertainties in the navigation. This part aims to provide the vehicle with a better awareness of its situation and improve the management of the uncertainty.