Funding : Ifremer
E-mail : Clementin.Boittiaux (at) ifremer.fr
Engineering degree ESIEE Paris 2020
Thesis director : Vincent HUGEL
Co-supervisor : Claire DUNE
Co-supervisor : Aurélien ARNAUBEC
Co-supervisor : Ricard MARXER
To this day, the vast majority of the ocean floor remains unexplored. Indeed, exploring great depth is difficult due to many scientific challenges. In order to better know and understand this environment, many tools have been developed over the years to collect samples, physical data and videos from the abyss. Recently, in 2020, the French oceanographic fleet expanded with Ulyx, an autonomous underwater vehicle (AUV) capable of diving up to 6,000 m.
The particularity of a robot like this one is to be able to carry out some tasks on its own while in depth to successfully fulfill the mission it was previously assigned to. To achieve this, Ulyx will need, among other things, to be able to position itself with precision and recognize scientific points of interest that were indicated to it. However, currently used positioning systems such as USBL have an uncertainty of about 10 m when they are deployed at great depths.
Therefore, this thesis aims to reduce this uncertainty by integrating artificial intelligence on board of the AUV. With latest technological advances in deep learning, particularly in the field of computer vision, our research focuses on the recognition of already visited points of interest and the location of the vehicle in these same areas. We pay particular attention on computer vision approaches, which present many challenges in this environment. One of them, for example, is to make our models robust to topological or physical changes that can occur year after year.