If you want to find methane leaking out of the seafloor, you go looking for cloudy water. Gas works its way up through the sediment, stirs the mud on the way, and the haze it leaves behind is something a sensor can actually measure. Scientists call that haze turbidity, and it’s one of the more reliable signs that there’s an active seep under you.
Woods Hole Oceanographic Institution has been teaching its deep-sea drone to spot that haze by itself instead of waiting on instructions from the ship. The drone managed it. It also found the shortcut, which was to run itself into the bottom and make a cloud right where it was sitting.
WHOI put the account out on July 8 in its own Oceanus magazine, in a piece by Amy Nevala. Sean Kelley, the WHOI engineer leading the push to make the vehicle smarter, said the reaction on board was to stop and think it through again. “This is not as easy as we thought,” he said.
So was that a failure?
I’d say no, because the drone did exactly what it was told to do. You point a machine at turbidity and it doesn’t care who made the turbidity. Doing that the honest way means holding a survey grid in cold water, reading a pretty faint signal and deciding it’s real. Bumping the bottom gets you the same reading in about a second. That’s a software problem, and software problems are a lot easier to fix after you’ve watched one happen than before.
This drone gets booked like a research ship
The vehicle’s called Sentry, and it’s a science tool rather than a weapon. It belongs to the National Deep Submergence Facility at WHOI, paid for by the National Science Foundation, the Office of Naval Research and NOAA, and scientists book time on it for expeditions the same way they’d book a research ship.
It’s also an AUV, and the difference between that and an ROV is worth having straight. An ROV hangs off a cable, with a pilot on the ship driving it and watching a video feed. An AUV goes over the side with a mission already loaded into it and nobody holding the other end. There’s an acoustic link, so the ship can check on it and even hand it a new job while it’s still on the bottom, but nobody’s flying it.
Sentry weighs 2,750 pounds without extra science gear, runs on lithium-ion batteries and moves itself around with four brushless electric thrusters mounted on wings that pivot. Those pivoting wings are the interesting bit. They’re what let it hover, climb a scarp and hold a tight grid close to the bottom, instead of blowing past everything like a torpedo with a camera bolted on.
All of those come off the facility’s own published spec sheet for the vehicle. WHOI’s article rounds the weight up to a 3,000-pound car, which is close enough for a magazine and worth knowing if you’re comparing numbers across the two.
Nobody knew these seeps were there
Methane seeps on the US Atlantic margin are a genuinely recent discovery, which surprised me, because that’s a busy stretch of water. Before 2014, fewer than five cold seeps had been mapped along the whole margin. Then a team led by Adam Skarke of Mississippi State went back through water column sonar data that NOAA’s Okeanos Explorer had been collecting anyway, and pulled more than 570 gas plumes out of it between Cape Hatteras and Georges Bank. A USGS inventory published ten years later counts roughly 1,139 unique sites.
Methane is a potent greenhouse gas, and most of what leaks out down there gets eaten by microbes well before it reaches the air. How much gets past them, and from where, is still an open question. Gas working through sediment can also weaken it, which is the kind of thing you’d want mapped before anybody puts something heavy on that slope.
Sunita Shah Walter, a biogeochemist at the University of Delaware and a former WHOI postdoc, handles the chemistry end of the project. Her team reads the traces microbes leave behind when they eat methane, which sorts a stable methane hydrate from a live seep. She’s said the seepage on this margin has no good explanation yet, which is a fairly unusual thing for a researcher to admit about their own patch of ocean.
That’s the job Sentry was doing in 2025 off the Mid-Atlantic coast, flying tight grids from the University of Rhode Island’s research ship Endeavor and taking pictures of the bubbles so the team could put a location on each vent.
The rewrite is a year of work
Kelley’s team is spending the next year with software engineers on Sentry‘s mission control system, according to WHOI. They want a vehicle that can notice a feature like a seep, drop the survey it was handed, go investigate the thing, then get back on the line it was given. None of that exists yet.
Teaching the drone to judge a signal for itself is the hardest thing on Kelley’s list, and that test result is why I take the project seriously. If a machine can’t tell a real plume from one it stirred up itself, it’ll happily hand you a full dataset of nothing. Better to find that out on the Atlantic slope than on a mission that costs a lot more to run twice.
WHOI hasn’t said when the bottom strike happened, how many times the team ran that test, or which sensor was reading the turbidity, and I can’t find those details anywhere else. We’ve got Kelley describing the whole thing on the record in his own institution’s magazine, which is more candor than you normally get about a robot doing something dumb. Underwater drones tend to get written up in press releases where nothing ever goes wrong, whether they’re hopping across the Pacific floor taking cores or sitting on the seabed listening for submarines.
The seeps aren’t done with visitors either. WHOI says Alvin, the crewed submersible, goes back to the same Atlantic sites next year so scientists can look at them through a window, and a third expedition after that will run a big shipboard coring rig to pull sediment up to 90 feet long and read the methane history stored in it.
Sentry has been diving for science since 2008, and it’s spent most of that time flying the grids the ship handed it. Nevala’s account of what happened when they let it decide for itself went up on July 8.
