Pomegra
Machenta, an 11-person Imperial College London spinout, has raised a $40 million seed round co-led by Creandum and Union Square Ventures to build a robot model that learns new tasks from one demonstration.
Key Takeaways
- Machenta raised a $40M (about £30M) seed round, announced October 27, 2026, co-led by Creandum and Union Square Ventures.
- Imperial College London calls it the largest seed round ever raised by a UK robotics startup. The valuation is undisclosed.
- The 11-person team trains its model entirely in simulation and says it needs one or two real demonstrations per new task.
Lead
Machenta, a London robotics startup founded in 2025, announced a $40 million seed round on October 27, 2026. Creandum and Union Square Ventures (USV) co-led the financing, which is equivalent to roughly £30 million. The company did not disclose its valuation.
The money is meant to expand the company's team, buy more computing power, generate more simulated training data and fit out a new office and laboratory in King's Cross, London. Machenta employs 11 people. That works out to more than $3.6 million of fresh capital per employee, an unusually heavy ratio even for a seed-stage AI lab.
What Does Machenta Actually Build?
Machenta builds a robot foundation model, a single pre-trained AI system meant to control different machines, that learns a new task from one demonstration. Most competing approaches depend on large volumes of real-world data, such as teleoperated robot recordings or human video. Machenta pre-trains its model entirely in simulation instead, with the stated aim of teaching it how to learn rather than how to perform any single job.
The company has shown the system on several robot types, including humanoids, doing tasks such as packing boxes on an assembly line and tidying a table. The approach grew out of research at Imperial's Robot Learning Lab, where CEO Edward Johns, a professor at the university, has worked on the idea for roughly six years. Co-founder Vitalis Vosylius is Johns's former PhD student.
Johns frames the goal as speed of acquisition: intelligence, he has said, is less about what a system can already do and more about how quickly it can learn something new.
How Does the Round Compare With Earlier Funding?
The seed is roughly seven times larger than Machenta's earlier financing. A smaller 2025 round of about £4.3 million included Hoxton Ventures as an early backer. Details of the other participants in the new round were not disclosed.
The size is a signal about the market rather than the company's traction. Machenta has no announced customers and, by its own account, no near-term commercialization plan. Investors are paying for a research thesis and a founding team, not revenue. A seed of this scale also means the next round will be judged against a high bar, since the company will need to show that simulation-trained models transfer to physical hardware outside demo conditions.
Kathi Wilhelm of Creandum said very few teams have repeatedly changed what people thought possible in robotics. Rebecca Kaden of USV said robots become far more useful when they can adapt instantly to new tasks. Both statements describe the pitch more than they test it.
Why Is Simulation-First Training Contested?
Simulation-first training is contested because the gap between simulated and physical environments has frustrated roboticists for years. Contact-rich work such as handling deformable objects, inserting parts with tight tolerances or gripping slippery items is hard to reproduce accurately in software. Machenta's own limits reflect this: precise, complex movements remain difficult, and early versions needed about 30 minutes of additional data collection per task.
That detail matters. A "one demonstration" claim sits alongside a measurable fallback to extra data on harder jobs. The commercial question is how often that fallback is needed, and whether the per-task setup time shrinks as the model scales.
Who Else Is Competing for Robot Foundation Models?
Machenta enters a field with far better-capitalized rivals, including Physical Intelligence, FieldAI and Skild AI. Skild AI has been valued at more than $14 billion after a $1.4 billion round in January. Against those figures, a $40 million seed buys a small slice of the compute those companies can deploy.
The bet is therefore on efficiency. If a model can learn from a handful of demonstrations, it needs less real-world data collection, which is the most expensive input for the data-hungry approaches. If simulation does not transfer well, Machenta's smaller budget leaves less room to compensate by brute force.
What Does the UK Angle Mean?
The round gives the UK a rare large robotics financing at the earliest stage. Imperial's Vice-Provost Mary Ryan said the result signals an opportunity for the UK to become a global leader in AI-enabled robotics. University spinouts have historically struggled to keep scale-up capital in Britain, and this deal pairs a London lab with a Stockholm-founded firm in Creandum and a New York-based fund in USV.
Outlook
Machenta now has the capital to test whether simulation-trained learning holds up on real hardware at scale. The milestones to watch are named customers or pilot deployments, a reduction in per-task data requirements and any follow-on round that discloses a valuation. Until then, the company is a well-funded research bet, priced against rivals that have raised tens of times more.
