Iceland-based Hefring Marine did not start with autonomy in mind: to begin with it was all about safety.
”We started with accelerometers and a very simple computer that could capture that accelerometer data and visualise it in real time for the operator,” says co-founder and now CEO Karl Birgir Björnsson.
The problem was straightforward but largely unaddressed because vessel operators typically experience a different motion profile to passengers or crew seated elsewhere and therefore don’t realise the difference.
In high-speed vessels, particularly RIBs used for tourism and transport, this difference could translate directly into injury risk.
Early deployments focused on monitoring motion at specific locations on board. The intention was to give operators immediate feedback so they could adjust speed or heading before conditions became unsafe. However, real-time feedback alone proved insufficient.
“If we’re just showing to the operator something that happened even a millisecond ago, it’s already happened,” Karl says. “They’re still hurt.”
That limitation pushed Hefring towards prediction rather than observation. By analysing how vessels move as they interact with waves, the team began developing models that could anticipate the next impact before it occurred. This work coincided with research in Iceland on high-speed vessels, including whale-watching boats, where the data revealed large variations in impact forces across different parts of the same vessel.
“The impact force in the bow could be ten times greater than it was for the operator,” says Karl.
This research led to the formation of Hefring as a company in late 2018, with operational work beginning in 2019. While the initial project had been tied to vessel design validation, the findings highlighted a broader opportunity: using motion data to inform safer vessel operation. That principle remains the foundation of the company’s systems today.
IMAS data collection
Hefring’s current IMAS systems are deployed across approximately 400 vessels, ranging from small workboats to fishing trawlers and offshore support vessels. Data is collected automatically and continuously.
“Everything from how vessels move through the water to what kind of sea conditions they’re tackling, to what kind of fuel it uses – everything down to multiple times a second,” Karl says.
The data are anonymised and used primarily for model training, and the scale and consistency of the data are what allow Hefring to build systems that can be deployed quickly on new vessels. It also differs from the computer vision approach, which could eventually be used in tandem, Karl says.
“There are companies that do that very well, but we don’t need that for our applications,” he says. Instead, the company concentrates on vessel dynamics – like how a boat responds to waves and speed – and how those factors affect safety and efficiency. The sensors used are largely focused on motion, supplemented by data already generated by onboard systems.
“A boat already has a lot of sensors,” Karl says. “Everything from the engine to the depth sounder – we capture everything that’s on the vessel.”
Real-time information is key, when decision-making at sea can depend on current conditions, not just historical averages.
“For any future of autonomy, it needs real-time intelligence,” he says. “It needs actual information on what’s happening right now, because that’s what an operator would do.”
Human-driven autonomy
Despite frequent use of the term ‘autonomy’ in the maritime sector, it isn’t truly the case – yet.
“Nothing is really autonomous,” he says. “That’s why the word ‘unmanned’ is used quite a lot more – or ‘optionally unmanned’.”
Hefring’s systems are designed to work within this scope. Most vessels still need people on board, and most maritime operations involve tasks that cannot be automated.
“We’re not really talking about automating the process of catching fish or rescuing somebody,” Karl says. Instead, the focus is on reducing the cognitive load on operators.
By managing throttle, routing or speed recommendations, systems can allow crew to focus on other responsibilities.
“You could take one of the crew members and make them more effective on board doing those other jobs as opposed to operating the boat,” Karl says, although this would only work if the system behaved in a way that reflects experienced human judgement.
“You need that instinctive feeling that an operator would have built into a system,” he says. Without it, trust breaks down.
The models Hefring develops are trained on real operational data, capturing how skilled operators respond to different sea states. This allows the system to support, rather than override, human decision-making.
In this sense, Hefring’s work sits between manual operation and full autonomy. The systems are intended to augment human capability while laying groundwork for more automated operations in the future.
Cost and fuel savings
Fuel efficiency is one of the most measurable outcomes of Hefring’s systems.
“If you train a model to make the right decisions, you avoid human error over time,” he says.
The benefits are not limited to speed optimisation. Fleet-level visibility provides insight into how vessels are actually being used, including understanding idle time – idling consumes a lot of fuel – routing inefficiencies and operational patterns.
By identifying unnecessary idling and adjusting scheduling towards a more just-in-time approach, operators can significantly reduce consumption.
“Just structuring how the fleet is being used could yield you 30–40% fuel savings,” he says.
Beyond fuel, improved visibility reduces operational risk.
Vessels are expensive assets, and limited insight increases risk.
“Not having that level of detail is taking on unnecessary risk,” he says. Real-time data, for example, allows operators to identify vessels operating in deteriorating conditions and respond earlier if assistance is needed – resulting in the offset of the cost of deploying the systems.
Regulations and classification
While technical capability has advanced rapidly, regulatory frameworks have evolved more slowly and it’s still a bit of a mess where autonomous vessels are concerned, with no clear route set out yet.
“I think the technical capability is more advanced than the regulation that’s limiting it,” Karl says, with existing maritime rules written with crewed vessels in mind and not addressing new operating models.
A central challenge is responsibility. “There are a lot of ‘what ifs’ that haven’t been fully addressed,” he says, including questions about liability when unmanned vessels cross borders or operate without direct human control.
“Classification societies aren’t going to go out of the frame that has been set up by regulators,” he says. As a result, companies developing new technologies must fit within existing frameworks or wait for standards to evolve.
At the same time, Karl acknowledges that regulation can enable adoption. Mandated standards can reduce risk aversion by providing insurers with defined requirements, in turn affecting how quickly new systems are adopted.
Hefring’s data systems also support emissions reporting, an area of growing regulatory focus. “CO₂ emissions, NOx – we calculate that. It’s all reported in real time,” Karl says. Operators can generate reports covering specific periods with minimal manual input.
Where vessels are fitted with emissions sensors, data is captured directly. In other cases, emissions are calculated from fuel consumption.
As emissions reporting becomes more widespread, automated data collection reduces administrative burden.
“The old-school way is calling operators, getting invoices, putting it into Excel,” Karl says. “With a system like ours, it’s one button – download the report and send it off.”
Next steps
Hefring’s development strategy is still aligned with its original focus on vessel motion and prediction.
“Our focus is to keep developing, but on the same premise that we started off with,” Karl says. Navigation, fuel modelling and autonomy all build on the same underlying understanding of how vessels interact with the sea – which means real-time decision-making, again, is central.
Looking ahead, Karl expects autonomous vessels to appear gradually in specific use cases.
“I think we’ll see it in our lifetimes,” he says, while adding that adoption may follow developments in other industries, which often move before maritime.
For now, Hefring continues to focus on improving safety, efficiency, and data quality for existing vessel operations, while preparing for a more automated future.
