Skip to main content

The drone boat of ‘Shipwreck Alley’

Meet BEN, the self-driving boat that’s been tasked with helping lay bare the long-lost secrets of the lakebed

Illustrations by Zoe van Dijk

It was just past midnight when the Ironton punched a 200-square-foot hole in the side of the Ohio. It was dark, the waters were rough, and the Ohio, a wooden bulk freighter loaded with flour and feed, was no match for the Ironton, a schooner heavy with coal. The Ohio sank within half an hour, and the Ironton soon followed, taking five of its crew down too. 

Their ghostly hulls have sat largely undisturbed at the bottom of Lake Huron since colliding in late September 1894 — just two of the many wrecks that lie in a treacherous stretch of water called Thunder Bay off Michigan’s northeastern coast. Some are so well preserved by the lake’s frigid freshwater that their unbroken masts point definitely towards the surface, rigging still intact. Others have dishes in the cupboards, a century late for dinner. A few years ago, local media reported that divers found a 1927 Chevrolet Coupe amid the wreckage of a steamship, covered with algae and barnacles, but nonetheless pristine. You can thank the rocky shoals, frequent fog, and sudden gales of Thunder Bay for turning what was once the bustling marine interstate of America’s early industrial age into a modern-day museum of Great Lakes maritime history. Locals called it “Shipwreck Alley.” 

BEN is a self-driving boat that’s been tasked with making maps

Divers flock from all over the world to see the wrecks for themselves each year — and last spring, they were joined by an unusual interloper: an autonomous boat named BEN. The boat was developed by researchers from the University of New Hampshire’s Center for Coastal and Ocean Mapping, with funding from the the National Oceanic and Atmospheric Administration’s (NOAA) Office of Coast Survey. Its name is short for Bathymetric Explorer and Navigator, but it also honors Ben Smith, the former captain of the university’s research vessel Gulf Surveyor, who unexpectedly died in 2016. BEN is a self-driving boat that’s been tasked with making maps, and it was brought to Thunder Bay to help lay bare the long-lost secrets of the lakebed.

On land, we are spoiled for maps. A few hundred imaging satellites now orbit the Earth, collecting new imagery each day, some at startlingly detailed resolution. Our maps go far enough back that we can see how the planet has changed, and how we’ve changed the planet. But on water, maps of this detail simply don’t exist. Mapping is still largely done by boat, and unlike satellites, boats need crews. It’s expensive, time-consuming work, and especially difficult in water that is shallow, rough, or remote. It’s why we know comparatively little about what lies beneath the surface of our oceans and lakes — by some estimates we’ve mapped just 9 percent of the world’s oceans to modern standards — and why BEN and vehicles like it hold so much promise. The thinking is that fleets of tireless, automated, uncrewed vehicles could one day criss-cross our waters, making maps where humans can’t or won’t. 

Ask oceanographers about our lack of maps, and they’ll tell you it’s hard to know what’s important until you know what’s there in the first place. Having the capacity to map more of our oceans, more often, and in higher detail than ever before, would give scientists an unprecedented amount of data — data crucial to our understanding of climate change, and the effects it has had on everything from melting Arctic ice to undersea life. It would also be a boon for nautical safety and navy intelligence, for deep-sea miners in search of untapped resources, and for the telecom companies unspooling undersea cables from coast to distant coast.

For now, the researchers have set their sights on the more modest locale of Thunder Bay. While the Ohio was discovered in 2017, the Ironton’s final location is still unknown. As a test of its nascent map-making abilities, BEN was tasked with looking for the Ironton’s remains. But the robotic explorer is more than just a seaworthy self-driving car. It is an ambitious little boat with its own challenges to overcome and opportunities to seize.

In our oceans, there are countless more mysteries waiting to be solved, waiting for boats like BEN.

At the local marina, there was no shortage of curious onlookers drawn to the sight of the tiny, strange-looking boat.

BEN is about 13 feet long, or the length of a compact car, and a bright banana yellow. It reminded me of an oversized jet ski — but with a tower of cameras, antennas, and other important sensors where a person would normally sit, and an array of computers packed inside.

The harbormaster, laughing from the driver’s seat of his pickup truck, asked if the research team had charged BEN’s batteries (in fact, BEN runs on diesel). Another truck pulled alongside the boat launch, three small dogs jostling for position in the open window of the back seat. “There’s no one in there?” the woman on the passenger side asked, eyes wide. The man driving it asked if we could use BEN to catch fish. 

It’s here in Alpena, Michigan, a small town of 10,000, that the Thunder Bay National Marine Sanctuary is based. The sanctuary is overseen by NOAA, and it protects some 4,300 square miles of freshwater — basically, the top half of Lake Huron on the American side. Like the world’s oceans, much of it has never been mapped.

There are 99 known wrecks in the sanctuary’s waters, but at least 100 more that have yet to be found

“If you can believe it in this day and age of technology, we have only surveyed about 16 percent of the sanctuary,” said Stephanie Gandulla, the sanctuary’s research coordinator. Gandulla told me there are 99 known wrecks in the sanctuary’s waters, but at least 100 more that have yet to be found — the Ironton among them. That’s not even including the countless wrecks that lie outside the sanctuary, which litter the lake’s Canadian side. “There’s lots of work yet to be done,” she said.

Leading BEN’s sojourn on Lake Huron was Lindsey Gee, the mapping and science coordinator of the Ocean Exploration Trust, the ocean research nonprofit founded by explorer Robert Ballard of Titanic discovery fame. Gee and his colleagues don’t typically map freshwater lakes, but they decided to collaborate with the sanctuary, and the University of New Hampshire researchers, in anticipation of using BEN at sea.

The boat’s size makes BEN well-suited to coastal waters, and regions too shallow for larger boats yet too deep for divers. They planned to spend two weeks in and around Alpena mapping points of interest to the sanctuary’s staff — the Ironton among them. The hope is that BEN — tireless, automated — will eventually be able to collect more data for analysis than the sanctuary’s own crewed research vessel Storm could collect on its own. When I visited, the researchers were preparing to map some shallower shipwrecks that were close to Alpena’s shores. It was a dry run of sorts for the Ironton search to come.

BEN’s minders sat across the marina, inside a small white tractor-trailer parked by a break wall — the mobile command and control center that is crucial to BEN’s operation. It is much more spacious on the inside than it seems from outside, crammed with computers, tables, tools, and a trio of giant screens that let the researchers monitor BEN’s vitals and see what its cameras and radar see. Blessed with a day of clear weather in an otherwise dreary week, the researchers offered to show me how BEN makes maps.

BEN’s automatic identification system declares itself a “pleasure craft”; there’s no option yet for “self-driving boat”

Val Schmidt, the university research engineer who leads BEN’s development, helped ease BEN down the boat launch and into place alongside one of the marina’s docks. BEN’s automatic identification system declares itself a “pleasure craft”; there’s no option yet for “self-driving boat.” Fully fueled, it weighs about 2,000 pounds and can run for around 16 hours.

Should they ever lose contact, there’s also a kill switch on the side of the boat — a simple lanyard of red string tied to a cap. Pull the string, the cap comes off, and the fuel stops flowing. That way it can’t run away to Canada, one of Schmidt’s colleagues joked.  

They turned the boat on, and Schmidt used his foot to push BEN away from the dock. For the sake of expediency — and to minimize any chance of damage before reaching open water — a colleague back in the trailer manually guided BEN out into the lake using a knock-off Xbox controller, like a very expensive remote-controlled boat. Once BEN is free of the break wall, they let the ship’s onboard computer take control.

“Mowing the lawn” is what oceanographers call the slow, tedious craft of making maps at sea.

You drive your boat in a straight line while your sonar repeatedly pings the seafloor below with sound. At the end, you loop around and start a new line, going back the other way next to the line that was just completed. With each line, you collect more data until you’ve covered the area you want to map — like filling the outline of a shape in a coloring book.

BEN, however, can do all of this on its own, and neither waves nor wind can conspire to push the boat off course. The whole process is mundane, but the researchers have to remain alert, continually looking for any potential hazards that might require them to take manual control. Though BEN may be able to drive itself, it is still learning how to understand and respond to the world around it.

“Mowing the lawn” is what oceanographers call the slow, tedious craft of making maps at sea

The idea is that, eventually, BEN will not only be able to tell the difference between a sailboat and a container ship, but also decide how to alter its path in response. BEN only tops out at about 5 and a half knots — if it were a runner, it could race a 30-minute 5K — whereas big merchant ships might move at a swift 20 knots. Realistically, BEN would only have a few minutes to identify a potential hazard — its location, what it is, whether it’s moving — and then figure out where to go.

Working to tackle this problem is Coral Moreno, a PhD student on BEN’s development team. Her specialty is sensing and perception. Moreno has been taking all of BEN’s various sensors — cameras, LIDAR, radar, GPS, and sonar — and attempting to fuse the data together into a comprehensive picture of potential hazards above the water, and eventually, below. “There is no single sensor that can provide you all the information that you need,” Moreno said. “They really complement each other because they are good for different ranges, and they provide you [with] different kinds of information. So you really need to use all of them.” 

While there’s lots to learn from the world of self-driving cars, it’s not as simple as putting car technology on a boat. Water is rarely still, and BEN is constantly moving. There are no stoplights, and no clearly marked lanes. Getting good data to train BEN’s image recognition algorithm has also been challenging. Images taken by BEN’s cameras are sometimes distorted by splashes and glare on the surface of the water. Existing image sets — what researchers use to train their neural networks to recognize, say, faces — weren’t created with the marine environment in mind.

A small window on Moreno’s laptop flashed possible matches, giving me a glimpse at what BEN thinks it’s seeing. Close to shore, it seemed to work, correctly identifying dogs and their owners walking along the pier, the boats in the marina, and the trucks that trundle along in the distance with a high degree of confidence. But out on the lake, it’s mostly false positives. Much to the researchers’ amusement, BEN mistook lighthouses for fire hydrants during early tests. Less amusing is the possibility that BEN could misidentify a potential hazard, and meet the same fate as the wrecks it’s supposed to hunt.

Getting good data to train BEN’s image recognition algorithm has also been challenging

BEN is so small that — here, Moreno made a splat noise — a larger boat could run into BEN “like it was nothing, and not even notice.”

While Moreno and her colleagues keep an eye out for any splat-worthy boats, they also have their eyes on the sonar data BEN is sending back. BEN is equipped with a multibeam sonar, which uses sound to ping the seafloor in a wide, fan-like area, and then measures the reflection of each ping. The time it takes for a ping to return is used to measure depth, and the strength of the ping’s reflection — the backscatter — can be used to characterize the makeup of the lakebed or seafloor. Those measurements are then rendered, roughly, and visualized in real time on one of the trailer’s screens.

We could see what’s in the water column directly below BEN — that is, everything the pings hit on the way down — and the current depth. In another window, an isometric, rainbow-colored cutaway of the seafloor slowly extruded, in cool colors for the valleys, and warmer ones for the peaks. The operators are constantly watching the data to ensure the sonar is properly configured. Shallow water requires different settings than deeper water.

Temperature and salinity can also cause sound to bend as it moves through the water, resulting in inaccurate readings, so any environmental changes — measured as soundspeed — must be accounted for too. The idea is that BEN will eventually be able to set and correct these values itself, so it can not only drive — and successfully avoid hazards — but also make maps by itself. Another graduate student, Lynette Davis, has been working on the feature, called “Don’t run aground BEN.” They plan to test it this spring, but for now, the researchers set the values themselves.

It’s all very interesting, but I was mesmerized by the backscatter the most. New data slid into view like a side-scrolling video game, or the way images used to load over dial-up modems, line by line. Rocks and mud reflect sound differently — as do the ghostly hulls of long-lost wrecks — and these differences can shed light on what makes up a lakebed or seafloor (or, in this case, what lies on top).

My eyes scanned the incoming telemetry, rendered in different shades of gray, and tried to make sense of the data. I looked for tell-tale ripples and anomalies in the backscatter, any beams or fragments that might suggest a wreck. As we passed over one of the sanctuary’s chosen sites, I saw what I thought was a hull. But it’s easy to see ghosts in the backscatter — to my untrained eye, a lot of things looked like a wreck — and we won’t know for sure until later. What we could see in real time is only a rough approximation of the polished data to come.

Once BEN is done here, the team’s mapping specialist, Erin Heffron, will process the collected data, and render it into a higher quality, more magnificently detailed map of the lake floor. Until then, I looked for ghosts in the backscatter, imagining how it would look to see the Ironton slowly emerge, largely intact, like traveling back in time.

BEN is about 13 feet long, or the length of a compact car, and a bright banana yellow.
BEN is about 13 feet long, or the length of a compact car, and a bright banana yellow.
Photo by Matthew Braga

BEN isn’t the only autonomous boat in operation, nor even the only boat to have emerged from the University of New Hampshire’s engineering department. An international team led by researcher Rochelle Wigley of the Center for Coastal and Ocean Mapping won first place in the Ocean Discovery XPRIZE, sponsored by Royal Dutch Shell. The multiyear challenge required participants to map a 250-square-kilometer patch of seafloor in less than a day, without any human intervention. Rather than map from the surface, Wigley’s GEBCO-Nippon Foundation team deployed an underwater mapping vehicle from an autonomous boat. They were awarded a cool $4 million for their work.

Students at Denmark’s Arctic Research Centre, part of Aarhus University, have also been developing an autonomous vehicle similar to BEN for the purpose of researching ocean currents near icebergs and glaciers, which pose safety risks for larger crewed vessels. There’s an ambitious project to build fleets of wind-powered boats, called Saildrones, that could rove the oceans in fleets for months at a time — mapping among their many potential capabilities. Another company, SeaMachines, demonstrated an autonomous firefighting boat in 2018, and an autonomous oil spill skimmer in 2019. The company said it’s currently testing its navigation assistance and perception technology on an A.P. Moller-Maersk container ship, where it makes more sense to augment the crew’s ability to safely navigate a busy port than automate them out of existence.

As for oceanographers, some believe that even a handful of these vehicles set loose on the ocean could fill a sizable gap in our seafloor maps

As for oceanographers, some believe that even a handful of these vehicles set loose on the ocean could fill a sizable gap in our seafloor maps. Roland Arsenault, a software engineer on the BEN research team, recalled the time he spent on a NOAA research vessel in the summer of 2018. Each day, the NOAA crew sent a few people out on a smaller boat to do mapping surveys. They would come back at night, process the data, and do it all again the next day. But what if they had a fleet of boats like BEN they could send out instead? A small crew could run five or six boats at once.

“I’m not talking about the whole ocean filled with them yet,” he told me, “but heading in that direction, right?”

The data collected would aid in the study of our changing climate and the prediction of storms, yield safety improvements for fishing and freight vessels, and help oil and gas companies cut their survey costs. An international organization of ocean mapmakers — the General Bathymetric Chart of the Oceans (GEBCO) — has estimated that a collaborative effort between commercial shipping operators, international hydrographic organizations, oil and gas surveyors, fishing boats, scientific research vessels, and, yes, autonomous boats, could yield a complete map of our oceans by 2030.

Back at the Thunder Bay National Marine Sanctuary, it will be a while before researchers can say if the Ironton is present among all the data collected last spring. The sanctuary’s own research vessel Storm covered an area of nearly 80 square kilometers in ten days, while BEN covered just over 73 square kilometers over 11 days — and the post-processing required to make sense of it all has been delayed by other mysteries.

After their time in Alpena, the researchers took BEN to sea aboard the Ocean Exploration Trust’s research vessel Nautilus. In August, BEN aided in the search for another wreck — the long-lost plane of storied pilot Amelia Earhart. They spent two weeks around Nikumaroro, a remote island in the western Pacific, but they came up empty this time, too. Like the wreckage of the Ironton, it’s not clear where, exactly, Earhart went down, and searches have been limited by cost and time. It’s the kind of mystery that would be perfectly suited for a fleet of autonomous boats like BEN.

It’s the kind of mystery that would be perfectly suited for a fleet of autonomous boats like BEN

I knew I couldn’t leave Alpena without seeing a wreck myself, so I visited one of the few you can see from shore: the remains of the Joseph S. Fay. It lies about an hour north of Alpena, behind a lighthouse on the beach, a lattice of wood and bent iron rising from beneath the surf.

When the waves fell back, they revealed the twisted metal and weathered, blackened wood of the century-old wreck. Though it was swept onto the rocks in 1905, there’s still a remarkable amount left. It stretches like a scar down the beach, only a fraction of the ship’s total length.

I had a few seconds at a time to study the wreck before it was obscured by the waves, like an Etch A Sketch the length of the shore. Then the wreck emerged again, and my eyes had a few seconds to adjust, to focus anew on a different part — like the backscatter from BEN’s sonar, looking for signal amid the noise.

Update March 23rd, 8:35AM ET: Added information about primary funder of BEN’s development, NOAA’s Office of Coast Survey.