Autonomous Underwater Vehicle Photographic Surveys in Complex Terrain

Photographic seafloor surveys are a valuable tool for studying ecosystems in the benthic zone. The vehicle carrying the camera must navigate at close distance to the subject. Flight style vehicles are energy efficient and thus capable of long, cost efficient surveys. Their limited manoeuvrability however brings a high risk of terrain collision.To improve vehicle safety and survey success, the component of the altitude tracking sensor and control system with largest impact needs to be identified. Since tests on larger vehicles are expensive, tests for comparing different configurations in simulation and on small scale test platforms need to be established.This thesis focuses on the two flight style vehicles Autosub6000 and Delphin2. Delphin2 is equipped with additional thrusters for hover capable actuation. Both vehicles detect obstacles in the vertical plane ahead with a mechanical scanning sonar. The terrain following behaviour of Autosub6000 is analysed using data from recorded missions in combination with case studies of a vertical plane simulation of Autosub6000. An altitude tracking method based on that of Autosub6000 is implemented on Delphin2, and studied in repeated experiments over a terrain step in a lake. The addition of the vertical thrusters for enhancing the altitude tracking is tested. To compare and evaluate experiment results, measures for vehicle risk and photographic survey success are introduced. Since vertical plane obstacle detection is identified as the limiting factor, improved detection methods for mechanically scanning sonars are developed.The simulation is validated for studying terrain following behaviour, but uncertainties about the buoyancy change at depth, and terrain detection limit the accuracy for simulating a specific vehicle. The lake experiments with Delphin2 are repeatable, and show distinction between different configurations. The vehicle risk and mission success are estimated more accurately using the known terrain profile. The tests of hover capable actuation at speed show a negative impact on the flight style actuation. The studies of Autosub6000 experiments, simulations, and Delphin2 experiments highlight that the forwards looking detection has the largest impact on success and repeatability. The causes behind the noise responsible for false detections and reduced detection range are identified, and filters for removal are applied. A Gaussian terrain prediction method is developed, increasing the detection range and the reliability thereof.Using these results, adjustments on existing systems can optimise survey outcomes, providing researchers with better data. To enhance flight style vehicles with thrusters, selective operation at lower speed is recommended. The development focus for future systems must be on the forwards detection and better estimation of buoyancy changes