AI helps a moving boat choose where to release young corals

A young coral can reach the reef inside a small ceramic device, attached to a tile and protected by its new home. The shape matters: a grazing fish that helps a reef by clearing algae can also damage a tiny coral. Australian Institute of Marine Science researchers have been refining these devices to balance protection, retention on the reef and the practicalities of making and deploying them.[1]

Once a boat carries the devices out to sea, another decision remains. Where, precisely, should each one go? AIMS’s Deployment Guidance System combines cameras, navigation and AI to help crews make those choices. The larger restoration area is selected with ecological modelling; the view from the boat helps determine the placement within it.[2]

Finding coral can mean holding back

The dataset’s labels are Deploy, No Deploy and Coral. Coral also means withholding a device: an existing colony is not somewhere to drop another coral. A boat-mounted camera looks down at the seafloor; AIMS marine scientist Amy Coppock supplied the released dataset’s expert annotations.[3]

The Reef Guidance System is detailed in an arXiv paper, updated on 28 September, using Great Barrier Reef imagery. Its patch-based configuration divides an image into pieces and considers their arrangement before release. Field processing reached 5.5 frames per second on a Jetson Orin. A typical deployment boat travels about one metre per second: the decision has to arrive before the target passes.[4]

The equipment has to fit the boat

AIMS engineer Benjamin Moshirian described the wider field problem in a November 2025 account. Trials used five-metre crewed vessels on inshore reefs and smaller tenders farther out, where deck space is scarce. Camera equipment, computing and dispensing hardware have to work together aboard those boats. The long-term ambition is automated guidance and release, but the development work includes crews, vessels and equipment configurations, rather than a single standard boat.[2]

That practical variation is central to the project. A system suitable for a research vessel will have limited reach if it cannot travel with the people who do the next deployment. Getting the decision onto the boat is part of making the science usable.

What the release decisions achieved

Table III gives spatial patch aggregation 77.78% overall agreement with expert deploy-or-withhold labels at a 0.3 threshold. Deploy precision was 59.77%—the share of proposed releases experts approved—and recall 67.16%, the share of approved opportunities found. Whole-image prediction achieved 76.59% agreement, 65.21% precision and 38.02% recall. These measure image-based decisions, not coral survival.[4]

Patch processing found more expert-approved opportunities, with lower precision. The paper does not specify the exact decision count behind this comparison.[4]

AIMS’s Pilot Deployments Program keeps the later biological question in view. Deployment locations are recorded so teams can check device retention and coral growth and survival after six months and a year. Its trials are also testing how the work fits into operations with Traditional Owners and delivery partners.[5]

A release instruction is immediate. The answer to whether that was a good place for a coral takes longer. Recording the location gives the team a way to return to the same patch of reef and find out.[5]

Report an error or suggest a correction