A London robotics start-up that most people have never heard of has just raised more money at the seed stage than any British robotics company before it. Machenta, a spin-out from Imperial College London, came out of stealth this week with a $40m seed round, roughly £30m, co-led by Stockholm's Creandum and New York's Union Square Ventures. It describes the round as the largest seed investment in UK robotics to date.
The money matters, but the idea behind it matters more. Machenta says it has built an AI model that can teach a robot a new physical job after watching a person do it once, and that it got there without feeding the model a single minute of real-world robot data.
Who is behind it
Machenta was founded in 2025 by Edward Johns, a professor of robot learning at Imperial College London, and his PhD student Vitalis Vosylius. The research goes back about six years, to Johns's lab, where the guiding question was whether a robot could learn a new physical skill as quickly as a person can.
The company is small. The Cap Table puts the team at 11 people and Tech.eu at around a dozen. The new funding will go on hiring, more computing power and the generation of more simulated training data. Machenta is also moving into a new office and lab in King's Cross, in London. It has not said how many people it plans to take on, and it has not disclosed its valuation or who else invested.
What it claims to have built
Most robots are still taught slowly. Typically, engineers collect many demonstrations of a task, often using real hardware, and train a system to repeat it. That is expensive and time-consuming, and the result tends to be good at one job only. Several companies in the sector now pay contractors to film themselves at work so there is enough footage to learn from.
Machenta has gone the other way. Its model was trained entirely in computer-generated environments. Once trained, the company says, it can direct a robot to do a new task, such as packing, assembling or tidying, after seeing it performed just once or twice. Johns calls the approach "instant robot learning" and likens it to "physical vibe coding", a nod to the way AI tools now let people describe software rather than write it line by line.
His own explanation is worth quoting: "Intelligence is not so much about what you can already do, it's about how quickly you can learn to do something new. Humans don't need to see the same thing over and over before they can do it."
Machenta says a single model has learned everyday tasks on several different robots, humanoids included. That flexibility is the point of what the industry calls a foundation model: one large model, trained once, then put to work across many machines and jobs.
The investors' view
Creandum partner Kathi Wilhelm said the team had worked on the problem "long before there was a company around it". Union Square Ventures managing partner Rebecca Kaden pointed to how much capital the industry is spending on real-world data collection and task-by-task training, which Machenta's approach is designed to sidestep.
That is the bet in a sentence: if simulation can do most of the work that rivals pay people to film, the cost of teaching robots could fall sharply.
The caveats
This is a funding announcement, not a product launch, and it deserves to be read that way. Machenta describes itself as a research company and has no near-term plans to sell a commercial product. It also acknowledges that precise movements and complex motions remain difficult. The claim that one demonstration is enough has not, as far as we can see, been independently tested in public, so treat it as a promising claim rather than a proven result.
The competition is also well funded. US groups Physical Intelligence and FieldAI are working on general-purpose robot models, and Skild AI was valued at more than $14bn after a $1.4bn round led by SoftBank in January. Crunchbase figures cited by The Cap Table put global venture funding for physical AI at $47.4bn across 521 deals in the first half of 2026. Against that, $40m is a modest sum, which is rather the point of calling it a record for Britain.
The UK angle
British universities have long produced excellent robotics research, and just as long been accused of watching it grow up elsewhere. A $40m seed cheque, led by a Swedish fund and an American one, shows international money is willing to back a UK team from day one. It also shows where the money thinks the value lies: in the software brain rather than the metal body.
For UK manufacturers, warehouses and care providers, the practical promise is robots that need less specialist programming. Many small and mid-sized British firms have held back from automation because setting up and reprogramming a robot cell costs too much for short production runs. A machine that can be shown a task once would change that sum, if it works outside the lab. That is a large "if", and Machenta itself is not yet selling anything.
It also fits a wider pattern. We have written about the British companies in the humanoid race, and about the hubs trying to get robots into UK workplaces. Machenta adds a research-led name to that list.
What's next
The things to watch are straightforward. First, whether Machenta publishes results that outsiders can check, ideally with demonstrations on real robots doing tasks they have not seen before. Second, whether it moves from research towards partners who will actually deploy the technology. Third, whether it can hire quickly in a market where the biggest US labs are competing for the same people.
If the approach holds up, the big winners may be the firms that build robot bodies and need a capable brain for them. If it does not, Machenta will still have shown that British robotics can attract serious early money. For a sector that has often struggled to find it at home, that is a useful result in itself.