Sea Technology

DEC 2018

The industry's recognized authority for design, engineering and application of equipment and services in the global ocean community

Issue link:

Contents of this Issue


Page 15 of 52 December 2018 | ST 15 erties in the water column and on the seafloor influ- ence, or are influenced by, the distribution of ani- mals in these environments. Using AE2000f as a scout, the team created ini- tial 1-cm-resolution, 3D visual reconstructions by flying the AUV at 10 m above the seafloor at 4 km per hour in order to identify areas of interest. This is higher and faster than typical AUV activity for color seafloor photography, allowing for a large expanse of seafloor to be covered. Once biological and chemical hotspots, such as bacterial mats and bub- ble plumes, were found, the team prepared for the next phase using artificial intelligence to determine where to do higher resolution photo mosaicking at 10 times the resolution of phase one. During the second phase, two AUVs and ROV SuBastian were deployed to carry out more detailed surveys. The underwater robots hovered close to the seafloor, covering a smaller, targeted area and generating several 1-m-resolution, 3D visual recon- structions. These 3D photomosaics were obtained using a unique camera system designed for the AUVs and data-processing algorithms, which allow scientists to understand the data being obtained much faster than would otherwise be possible. The team also adapted ROV SuBastian with addi- tional instruments, such as a Raman spectrometer, which tested pore water and the sediments within the hydrate fields. With each vehicle deployment, Thornton and his team simultaneously began processing the new data that they received. Each robot brought about 200 gigabytes of imagery, or 20,000 images per dive, plus chemical and other measurement data. If the team were to look at all of that data manually, it would take a lot of time and effort. Using unsupervised clustering/unsupervised learning, a type of artificial intelligence, on the images allowed the team to make quick and in- formed changes to the expedition deployment plans and focus their investigations on specific areas of interest. Color and texture provide a way for the algorithms to search for the main features desired by the team and cluster pictures that look similar. These groups of photos are then used by the scientists and engineers on board to determine where to sample next. Evolving Toolset "At the moment, humans generally make the deci- sions, but if we can prove that interpretations made by the vehicles can be trustworthy, then we will be confi- dent enough to hand over the responsibility without jeopardizing valuable opportunities to get important in- formation," said Thornton. Because of this innovative underwater technology and algorithms, the team is confident that the detailed obser- vations and in-situ chemical analysis were made exactly where needed in the field. The collected measurements gave the team detailed knowledge to make informed a platform to collect data, which humans need to inter- pret and understand in order to build new insight." With the use of artificial intelligence (AI), underwater robots can boost human understanding by carrying more of the load of interpretation. Robot Fleet For this expedition, there were three AUVs (TUNA- SAND01, TUNASAND02 and AE2000f), developed by the University of Tokyo, and Schmidt Ocean Institute's 4,500-m-depth-rated ROV SuBastian. The team under- took data-gathering deployments with these robots to look at how the chemical, physical and biological prop- (Credit: Schmidt Ocean Institute) (Credit: Schmidt Ocean Institute) (Top) TUNASAND01, TUNASAND02 and ROV SuBastian on Falkor's aft deck. The robots worked together in coor- dination, acquiring high-resolution imagery of the seafloor during the Adaptive Robotics expedition. (Bottom) ROV SuBastian being recovered to Falkor's aft deck after a suc- cessful dive.

Articles in this issue

Links on this page

Archives of this issue

view archives of Sea Technology - DEC 2018