When marine biology meets robotics
To study marine life in the Indian Ocean, the BUBOT project is applying big data to biology. The project involves automating underwater observations and then developing the algorithms needed to analyze the thousands of hours of video footage collected. This project is supported by I-SITE MUSE as part of its 2018 research support program.
Why use robots when diving is what makes exploring underwater life so special? For marine biologists, robotics holds great promise for unprecedented observations, as it allows them to explore marine habitats for hours on end at depths exceeding twenty meters. Through new observation tools, the BUBOT project (Better Understanding Biodiversity changes thanks to new Observation Tools) thus opens a new window onto the marine biodiversity of the Indian Ocean.
Low-cost robot
The first step is to build a robot that will automate observations at depths ranging from ten to one hundred meters, thereby enabling the study of coral reef biodiversity across this vertical gradient. “Biologists provided us with precise specifications to create the customrobot,” explains Karen Godary-Déjean of LIRMM (Montpellier Laboratory of Computer Science, Robotics, and Microelectronics). Specifically, the robot must be capable of autonomously traversing a transect while avoiding turtles, nets, and other drifting objects. Equipped with sensors, the device will therefore need to navigate around obstacles, slow down, and turn around. Another constraint is cost, because while underwater technology exists, the state-of-the-art versions available come with prohibitive price tags, running into the hundreds of thousands of euros.
"Marine Video Surveillance"
The robot isn’t the ultimate solution. The cameras installed on the seafloor are a complementary tool for observing the seafloor over the long term. “It’s like a real video surveillance system for the sea,” says Camille Magneville, a doctoral student at the Marbec (Marine Biodiversity, Exploitation and Conservation) laboratory, as she describes the system she installed at two sites in Mayotte—one in a protected area and the other in a fishing zone. Using twelve cameras, she collected nearly 500 hours of video footage over four days, which will help better assess the diversity and ecological roles of fish at each site.
However, big data is only useful for research if researchers are able to make use of the vast amounts of data produced by automated systems. BUBOT therefore includes a component focused on artificial intelligence: teaching machines to recognize fish species so they can automatically analyze all images recorded by cameras, whether mounted on a robot or fixed in place. “Thanks to all the videos collected in Mayotte, we hope to be able to recognize more than 100 species within two years (which should account for more than 95% of the individuals visible in the videos),” explains Sébastien Villéger, a researcher in functional ecology at MARBEC. To train the deep learning algorithms, the project is supported by students, mainly from the University of Montpellier, who are applying their taxonomy skills to annotate the fish in the videos. “Our goal is to have a wide variety of postures, conditions (lighting, apparent color), and backgrounds for each species, so that ultimately the algorithm will be effective in all situations,” notes the researcher.
In addition to identifying species, assessing the size of each individual is also a crucial aspect of studying underwater ecosystems. This measurement is not currently possible with the GoPro cameras mounted on the robot. The next step, therefore, is to equip the robot with a video system featuring stereo cameras.
Chatty fishermen
In keeping with its transdisciplinary approach, BUBOT also draws on the humanities to understand the environment through the insights of its primary observers: the fishermen. A geographer and an anthropologist thus round out the team. In Mayotte, they conducted nearly thirty interviews and gathered accounts of declining fish stocks in certain areas and shifting fishing grounds—all valuable information for selecting study sites. And the artisanal fishermen are talkative. “The interviews last an average of just over three hours, and the fishermen don’t hesitate to share information about their fishing grounds,” says Esmeralda Longépée of UMR Espace-Dev.
Due to COVID-19, BUBOT has fallen slightly behind schedule. Another unexpected change to the plan: of the three areas originally covered by the project, the fieldwork in Mozambique had to be canceled, as the country was recently classified as a high-risk area. It will be replaced by the Comoros. As for the first site, the Scattered Islands, the initial underwater expeditions proved promising, revealing a wealth of biodiversity. However, given the challenging diving conditions on a steep drop-off, the team is eager to send its robot there.