Lesson 04
DRIFT
A fleet of autonomous ships that sail the high seas to find new sources of energy sounds like science fiction. DRIFT is the visionary startup using AI to make it a reality.
In a landlocked corner of England,
Ben Medland walked across a field and saw the ocean.
It was a calm summer evening in the rolling countryside, and he was out with his six year-old son, James. As Ben spoke to him, he realised that his son was deeply concerned with the climate crisis he had heard about on the news. James pointed to a wind turbine on the horizon, its blades sitting motionless in the still air. ‘They need to turn it on.’
‘But there’s no wind,’ Ben explained.
‘Then why don’t they make one that follows the wind?’ James asked, with a child’s innocent logic.
Ben paused. Why don’t they?
Mobile wind-powered energy farms
Often Faculty are asked to apply AI to existing businesses. Sometimes, they get to help build new businesses from thin air. But for one memorable client, thin air is the business.
DRIFT Energy is the company that Ben founded as a direct result of that conversation with his son. It’s just the latest step in a career that’s taken him from advanced projects at BAE Systems, to building Accenture’s Digital and Data Strategy unit into a $100m business. A tall 43 year-old with a mop of black curls and an infectious smile, Ben has been solving problems his whole career, whether those problems were engineering, business or technological. Now he’s set his sights on the biggest challenge of them all.
The race to reduce the carbon we’re pumping into our atmosphere requires every solution humanity can throw at it. Incremental improvements won’t be enough. It needs new and creative approaches that rethink the fundamentals of how we produce, transport and consume energy. Like a turbine that can go where the wind blows, instead of just waiting for it to appear.
Ben’s vision is to build an unmanned sailing vessel that will operate as a mobile wind-powered energy farm. But the idea’s grown more complex since that lightbulb moment with his son. If you’re imagining a classic three-bladed windmill lashed to the back of a ship, think again. The power comes from a turbine slung under the vessel. As the ship speeds through the water, driven by the wind, its kinetic energy pushes water through the turbine’s rotor to generate the electricity: a windmill and a watermill all rolled into one. Or as Ben puts it: ‘A ship with a propeller where the energy goes the other way.’
But electricity generated in the middle of the ocean doesn’t really have anywhere to go. So the second piece of Ben’s design is to have the electricity from the turbine power an electrolyser, a neat piece of equipment that sucks in water from the sea and splits it into hydrogen and oxygen. The oxygen is released, while the hydrogen is pumped into storage tanks in the ship’s hull. When the tanks are full, the ship cruises back to port, offloads its cargo, and sails out to do it all over again.
Hydrogen has long been forecast to play a major role in efforts to reduce carbon emissions. Unlike most renewables, it can be stored long-term and transported easily. It can be used instead of oil in hard-to-electrify industries like heavy transport, aviation and - relevantly - shipping; and it can decarbonise industrial processes like steelmaking and cement production. If produced using clean electricity, it has virtually no damaging emissions.
The drawback is that there are precious few natural sources of pure hydrogen on Earth. While hydrogen is the most abundant element in the universe, on our planet, it is almost always bonded with something else. It has to be split out from water, biomass or fossil fuels - and, under the unyielding laws of thermodynamics, it will always take more energy to extract the hydrogen than you will get from using it.
Which is why it makes so much sense to produce it in places where there’s unlimited free clean energy. Places where you don’t get bogged down in difficult planning or permitting issues, and where you don’t need any costly fixed transmission and distribution infrastructure. Places where the natural feedstock - water - is available in almost unlimited quantities. Places like a vessel in the middle of the ocean.
Getting the ‘right wind’
When Ben approached Faculty to discuss the idea, they were cautious. ‘I could see them thinking, “This is a bit mad”,’ Ben recalls. Andrew Perry, head of Faculty’s Energy Transition and Environment business unit, puts it more diplomatically. ‘We thought it sounded amazing, but also extremely ambitious,’ he recalls. ‘How could so many different technologies be combined and optimised to work together? How would it work commercially?’
In his line of work, Andrew hears a lot of blue-sky thinking from visionaries who think AI is a magic wand they can wave at the climate crisis. But as Ben laid out his ideas, Andrew started to buy into it. ‘The more we discussed and understood the concept in Ben’s mind, the clearer it became that his vision had real substance, and the model made sense.’ He pauses. ‘So long as you could get the right wind.’
Ben’s hunch was that his sailing ship would be able to significantly outperform the efficiency of fixed offshore wind turbines, maybe even doubling it. But his whole idea rested on the vessel being able to plot a course to catch the best wind for the longest time. If it couldn’t do that, the economics of the project would never stack up.
Testimonials
"Each ship will deliver roughly 100 tons of green hydrogen every year. Run through a fuel cell to produce electricity, that’s enough to power up to 1000 UK homes, or for hydrogen-powered cars to drive 7.1 million miles."
‘First and foremost it’s a data science challenge,’ says Ben. ‘We looked at all the off-the-shelf options, nautical routing software and so on. For what we needed, there’s nothing like it out there.’ Which is unsurprising: most routing software is focussed on getting the vessel from A to B as efficiently as possible. It doesn’t cope well if, for example, A and B are in the same place. For DRIFT, it would be all about the journey.
That’s when Andrew understood why Ben came to Faculty. ‘He wasn’t just building a ship. He was building an autonomous vessel, one that could optimise its route in real time in the chaotic weather conditions of the North Atlantic.’ He needed AI to fill in the details.
Designing for speed in all weather conditions
The first job for Faculty was to test Ben’s preliminary hypothesis about how efficient the turbines could be if the ship was sufficiently intelligent in its routing. Instantly, they ran into a new challenge.
A sailing ship like Ben was planning had never been built before. Without a legacy design constraining them, DRIFT had a unique opportunity to take a completely fresh direction, to wring every knot of speed out of it. The faster the boat could go, the more kinetic energy it would transfer to the turbine and the more efficient it would be at making hydrogen. It would take its cues from the high-performance yachts that race in competitions like the America’s Cup and SailGP, boats which go so fast they literally fly over the water on foils. Only this time, the technology is directed at making the ship go greener.
But working from a blank sheet of paper is also incredibly daunting, especially when you’re developing something as complex as a hydrogen-producing sailing vessel. Starting from scratch throws up a multitude of choices. Small or large? Single- or double-hulled? How many sails? How many turbines? What shape? What materials? And those are just the basics. There are multiple layers of decisions, that each knock on to each other, and you have to resolve them all before the final ship can take shape.
To guide their design choices, DRIFT needed to understand how different options would affect the ship’s core purpose: to get the most speed from the widest range of weather conditions. From mild, balmy days where the air is still, to wild nights when storms sweep across the ocean, the vessel design had to deliver the best average speed across all the journeys it might take. And it had to be affordable and practical to build.
It is possible to model how a given boat might perform in different weather conditions. But it’s not trivial to optimise the vessel’s performance model at the same time as optimising the route it should follow. Suddenly things get very complicated, very quickly.
The best sequence of moves for one particular shape of ship might be no good at all with a different design. It’s a version of the classic ‘travelling salesman’ problem, except that you’ve got no idea how the salesman’s getting about and you don’t know where the customers will be.
Ultimately, Faculty’s modelling and DRIFT’s design loops fed into each other. Faculty could test how different types of vessels would perform, and DRIFT could test the trade-offs between performance and commercial feasibility to develop their understanding of what to build. The digital and the physical sides of the project were inseparable.
DRIFT’s first ship would cost tens of millions of pounds to build, so there was no room for error. Not in the construction of the vessel, and not in the algorithm.
The vessel design would take its cues from the high-performance yachts that race in competitions like the America’s Cup and SailGP, boats which go so fast they literally fly over the water on foils. Only this time, the technology is directed at making the ship go greener.
To model the vessel's likely hydrogen output, Andrew’s team dug deep into the science of seafaring to understand which factors contribute to performance. But the model wasn’t predicting weather. ‘There are supercomputers out there that forecast the weather better than we ever could,’ says Andrew, ‘and those feed straight into the algorithm.’ The model’s job was to find the ‘Goldilocks’ zone, where the wind is neither too strong nor too calm, but just right. A light wind is clearly no good for propelling the ship at the speeds it requires, but excessively windy conditions might damage the vessel or wreck it completely.
Key to the problem is the fact that the same wind has very different effects depending on which way the ship is pointing. Every sailboat has its optimum point of sail, the angle relative to the wind that generates the most speed. A boat facing straight into the wind will go nowhere. As it changes direction, it can gain or lose speed; for most vessels the optimum course is at about a 90-degree angle to the wind.
A sailing ship going from A to B will often have to choose a sub-optimal point of sail in order to get where it’s going. But an autonomous hydrogen-producing energy yacht can go in whichever direction it chooses. It’s got nowhere it needs to be.
But of course, at sea conditions are always changing. As well as getting the most out of the wind conditions in the moment, the ship also had to be planning ahead. The whole point of Ben’s vision was to follow the wind. And that wasn’t just about forecasting the next best action to go to the nearest good bit of wind. The algorithm had to think many steps ahead, strategically assessing thousands of possible route options many hours in advance, making sure it wasn’t sacrificing long-term performance chasing near-term gains. Like a game of chess, except with a board as big as the ocean and almost no constraints on moves.
And even then, it couldn’t necessarily just take the best route through the Goldilocks zone. Any time the ship is at sea with a full tank of hydrogen is time wasted. So the algorithm had to find a route that would take the ship back to port, ideally arriving at the exact moment that its tanks hit capacity.
Having developed the prototype algorithm after an intensive seven-week sprint, Andrew’s team tested its performance against thousands of potential voyages. The results were unequivocal. On its simulated journeys, the boat was able to achieve a load factor - effectively the proportion of the time the turbine is able to operate - of 70-80%, compared with 35-40% for fixed offshore wind turbines.
Ben’s hypothesis was right. Now the hard work could begin.
An algorithm for ‘the imperfect realities of the physical world’
Demonstrating the potential of DRIFT’s concept helped the business close its seed funding round, led by the aptly-named Octopus Ventures and supported by Blue Action Accelerator, which invests in novel ocean- and climate-friendly technologies. £4.65m was unlocked to continue detailed design work on the first ship, and to improve the model.
‘Our initial prototype was a relatively rough and ready model,’ explains Andrew. ‘It was necessarily pragmatic, to prove the concept at minimal cost. It focussed exclusively on wind conditions, and how they would impact on speed and hydrogen generation.’
But to go to the next stage, Ben needed the algorithm to be bulletproof. The initial modelling had shown that a 58-metre catamaran was the optimum design, but there were literally thousands of detailed design decisions that flowed from that. Those choices would be made based on how the different options performed in the route optimisation model, so Andrew’s team needed to make sure the model accounted for every relevant factor. This included things like wave height, sea state, tides, currents and swell - as well as how the interplay between all of those factors would affect the ship’s ability to harvest energy.
They’re all represented on a visualiser, which has a kind of hypnotic beauty when you look at it. Hundreds of arrows make whorls across the screen, their length and colour changing to show the wind’s speed and strength. The ship, a little purple dot, leaves its base on the west coast of Scotland and strikes out over the top of Ireland, zig-zagging dotted lines across the north Atlantic. A pulsating red blob sweeps in, representing dangerously high waves. The boat retreats, around northern Scotland and back to the Orkneys, taking shelter in the lee of one of the islands where it’s shielded from the worst of the conditions, as mariners have for centuries. When the colours subside to a more agreeable yellow, it heads back to port.
This is still a work in progress. Ultimately, the team is aiming to get the model of the vessel and its environment to the sort of level that Formula 1 teams operate at, where every last detail can be simulated and tested. That even includes coming up with a pit stop strategy to minimise the time the boat is stuck in dock while the full hydrogen tanks are unloaded and replaced.
But unlike a Formula 1 car, there’s no driver in the cockpit: DRIFT’s vessel will operate autonomously. That means the algorithm is far from an academic exercise. It’ll have to perform not just in the calm waters of Faculty’s data lakes, but out in the real ocean. Using only video, radar and sensor inputs, it’ll have to plot a course that follows the optimal wind, while dealing with every hazard to navigation the oceans can throw at it.
For Faculty, that means forensically examining every aspect of the simulated journeys, to understand where refinements need to be made. Boundaries need to be set for the shallow waters near coastlines. The vessel has to be aware of obstacles. Even routine sailing operations like tacking and gybing - changing direction, in layman’s terms - create inefficiencies that need to be accounted for.
‘It’s all the imperfect realities of the physical world,’ Andrew explains. ‘The list isn’t endless… but it is long!’
And it’s absolutely essential the algorithm can handle them all. Because Ben doesn’t want to launch one ship: he wants to launch whole fleets of them, hundreds or even thousands strong, that can deliver hydrogen to the four corners of the globe in quantities big enough to tip the scale of global warming. His goal is that one day his catamarans will be as synonymous with renewable energy production as the world’s 340,000 wind turbines.
‘Each ship will deliver roughly 100 tons of green hydrogen every year,’ says Ben. ‘Run through a fuel cell to produce electricity, that’s enough to power up to 1000 UK homes, or for hydrogen-powered cars to drive 7.1 million miles.’ And it’s 1.2 million kilograms of CO2 that won’t go into the atmosphere.
The current plan is to lay the keel for the first vessel in late 2025, and to build it within 18 months for a potential launch by summer 2027. But long before it sets sail, DRIFT has already been on an incredible journey against the odds, travelling over 15 million miles on its simulated voyages. ‘And the technology on DRIFT will only get better,’ Ben points out. ‘The sail performance, the turbine performance, the hydrogen plant, the energy chain - and the costs of all of those - are going to improve over time as the market evolves. The data and the speed of the compute will improve. So there's an awful lot of tailwinds behind the company.’
There’s a long way to go, but Faculty is proud to have supported Ben and the DRIFT team this far, and hopes to keep helping them every step of the way ahead.
Wherever the wind takes them.
The lesson in summary
AI is a feature, not a product. But it can define a product.
- The most advanced language models are very impressive at what they do. But despite this, there are few occasions where the things they can do out of the box - summarising or creating new text - correspond to the things that are most valuable. The same is true for other types of AI; from time-series forecasting to computer vision.
- As a result, there are few occasions where an AI model alone makes a full product. Use cases that start with ‘some impressive thing that AI can do’ and try to narrowly slot it into a business process will disappoint.
- Instead, it is best to think of AI as a cog in a machine, rather than the machine itself. A piece of functionality that can be connected together with others into a piece of software whose functionality goes much further than any of the component parts.
- However, AI is an unusually powerful cog that can make new kinds of machines possible. In much the same way that an engine alone isn’t a car, but it was the catalyst that made the whole paradigm of automotive transport possible.
- As a result, this is a good time to seek new ways of solving old problems. Even where AI plays only a small role in a piece of software, inside a business process, it may be the unlock that allows you to change the entire way the process runs.
Did you enjoy this story? There are nine others just like it, told from the perspectives of nine of our other inspiring customers, in the full book 'Ten Lessons From Ten Years of Applied AI'. Just leave your details below to get instant access to your copy of the book.
With contributions from:
Ten Lessons From Ten Years of Applied AI
Download the eBook
Get instant access to ten examples of AI solving the world's biggest challenges, told through the stories of ten of our most brilliant customers.