A robot that can follow animals in the deep sea

A new breed of robotic vehicles promises to open up exciting avenues of research in the deep sea by tracking swimming and drifting animals. This vehicle, called the Mesobot, was developed by engineers and scientists from the Woods Hole Oceanographic Institution (WHOI), MBARI, Stanford University, and the University of Texas Rio Grande Valley.

Mesobot underwent its first tests in 2019—first in MBARI’s test tank, then at sea when it was launched from MBARI research vessel Rachel Carson. Mesobot is designed to track and study swimming and drifting animals to 1,000 meters (3,281 feet) deep and for up to 24 hours at a time.

The Mesobot will extend and amplify previous midwater work by MBARI and other institutions. For decades, MBARI research (led by Bruce Robison, Steven Haddock, and others) has used remotely operated vehicles (ROVs) to study midwater animals. Though seldom seen, these creatures help support major fisheries such as tuna and billfish, provide food for other large animals such as sharks and whales, and help regulate Earth’s climate by moving carbon from the surface to deep waters. These animals are incredibly important to many processes in the ocean, but we have a limited understanding of their behavior and function in the difficult-to-access midwaters.

The multi-institution team designed the Mesobot to be less intrusive to deep-sea animals compared to most ROVs. It is equipped with low-light, ultra-high-definition (4K) cameras, red lights that are less visible to animals in the mesopelagic zone, and large, slow-turning propellers that minimize disturbances in the water. Cutting-edge onboard software also allows the vehicle to track animals or objects underwater for up to 24 hours at a time as they move through the water column, even during vertical migration.

The Mesobot is a hybrid between an ROV, which is powered and controlled using a tether attached to a surface ship, and an autonomous underwater vehicle (AUV), which is programmed at the surface and operates untethered without human intervention while underwater. At the beginning of each dive, operators control Mesobot from the surface like an ROV, using a thin fiber-optic tether. The researchers then sever the tether, and the freed vehicle continues on its own as an AUV.

MBARI staff have been involved in several aspects of the project. Principal Engineer Kakani Katija, Senior Scientist Bruce Robison, and Mechanical Engineer Brett Hobson were involved in the conceptual design of the vehicle. The vehicle’s main computer system is based on MBARI’s long-range AUV (LRAUV). The animal-tracking algorithms are being developed by Software Engineer Mike Risi, along with Katija and MBARI Adjunct and Stanford University Professor Steve Rock, and are based on early versions of software that have been tested on several other vehicles. Re-engineered versions of these tracking algorithms were recently demonstrated on MBARI’s MiniROV.

Given its size and flyaway capabilities, the MiniROV provided a strategic platform to demonstrate the stereo cameras, vehicle tracking, and vehicle-control algorithms that would eventually be integrated into Mesobot. While in MBARI’s test tank, the MiniROV tracking and control algorithms were put through their paces, tracking a number of targets through a wide range of scenarios. These early tests also provided valuable input on the configuration and selection of stereo cameras and illumination for Mesobot.

Once vehicle control and tracking algorithms and hardware were successfully tested in MBARI’s large indoor seawater test tank, the MiniROV was deployed at sea from the R/V Rachel Carson. During these field excursions, which took place in 2018 and 2019, the engineering team demonstrated successful tracking of several midwater animals. These included a Phronima amphipod in a salp barrel during different illumination conditions and a siphonophore whose repeated predation behavior was observed for more than five hours. The team believes this was the longest continuous observation of any animal in the midwater. Once these tracking developments were demonstrated, the underlying technology was transferred to Mesobot.

During a 2019 field program, the Mesobot was deployed in Monterey Bay off MBARI’s R/V Rachel Carson for several days of open-water field trials. To ensure the safety of the Mesobot, MBARI’s SmartClump, a system developed as part of the MiniROV with a camera that provides a birds-eye view of the Mesobot and facilitates data transfer from the vehicle to the ship.

During the initial field tests, the Mesobot made several dives to roughly 200 meters (about 650 feet) below the surface. These forays helped engineers test vehicle systems under real-world conditions and refine their methods for launching and recovering the vehicle off the side of a rolling ship. The deployments resulted in several successful episodes of tracking midwater animals, albeit for only minutes at a time. Efforts are currently underway to extend tracking duration and robustness to meet the Mesobot’s promise of uninterrupted, 24-hour “hands-off” observations. Further refinement of the tracking algorithms using machine learning to detect different biological targets is planned for 2020 using the MiniROV as part of a separate effort—funded by the National Science Foundation (NSF)—with Katija, Rock, Risi, and Paul Roberts, and collaborators at CVision AI.

Many marine animals (including delicate, soft-bodied jellies) migrate hundreds of meters toward the ocean surface at twilight and then descend back into the depths at dawn. This is the largest animal migration on Earth and, yet, the phenomenon is still poorly understood. Eventually, autonomous vehicles and tracking algorithms will enable scientists to perform prolonged studies of individual animals or groups of animals as they take part in these diurnal vertical migrations. In the near term, researchers hope that the Mesobot will enable a step in that direction by following individual animals during their journeys.

The NSF, WHOI, MBARI, and the Audacious Project housed at TED generously support this project.