Machenta, a robotics spinout from Imperial College London, has raised $40 million in what Imperial describes as the largest seed round raised by a UK robotics startup, to scale a robot foundation model that lets a robot learn a new physical task from a single demonstration. The round, announced on October 27, was co-led by Creandum and Union Square Ventures, and the company will use the money to scale its model and expand the team at a new office and lab in King’s Cross, London.
Learning From One Demonstration
Today’s AI-powered robots usually need a task demonstrated many times before they can do it reliably, a process that can take days and still leaves a robot able to do only the tasks it was specifically trained for. Any new skill typically means more demonstrations and another round of model updates by developers. Machenta says its model changes that sequence: show it a task once, and it can perform that task without further task-specific training.
The company calls the approach instant robot learning. It has demonstrated a single model learning tasks including packing and tidying across different kinds of robots, including humanoid robots, rather than training a separate system for each machine or each job. For a factory or warehouse, that distinction matters: a robot shown a new job on the spot is far more useful than one sent back to engineers whenever the work changes.
“Intelligence is not so much about what you can already do, it’s about how quickly you can learn to do something new,” said Professor Edward Johns, Machenta’s co-founder and chief executive, who is also Professor of Robot Learning in Imperial’s Department of Computing. “Humans don’t need to see the same thing over and over before they can do it. If you take some objects, and show somebody what to do with them, they’ll probably then be able to do the same task immediately. We’ve spent years trying to understand what it would take for robots to learn at the speed of a human. That is what instant robot learning is about.”
Trained In Simulation, Not On Real Footage
Machenta’s second bet is on how its model is pre-trained. Most robot models today are pre-trained on huge datasets taken from the real world, particularly human demonstrations, which are expensive to collect and raise practical and privacy questions when they involve filming people at work. Machenta says its model is pre-trained on simulated data, in computer-generated environments that mimic how objects move and interact.
That approach is designed to make pre-training far more scalable. Instead of collecting footage for every new task, the team generates simulated experience at scale, and the model learns how to learn from a demonstration rather than memorising specific jobs. The company says it used no real-world data in that pre-training, unlike rivals racing to amass real-world robot data, as in LG’s fivefold increase in its robot training-data target.
Six Years Of Research Behind The Launch
The work behind Machenta began six years ago at Imperial, where Johns and his team at the Robot Learning Lab began investigating whether a robot could one day learn new physical skills as quickly as a human. After the techniques developed in the lab began to show strong potential, Johns and his former PhD student Vitalis Vosylius founded Machenta with colleagues from the lab to scale the work.
Imperial’s vice-provost for research and enterprise, Professor Mary Ryan, said Machenta’s early success pointed to a wider opportunity for the UK. “Machenta’s technology, and the vote of confidence it has received from investors, is testament to the quality of the research by Ed and his Robot Learning Lab,” she said. “Machenta’s early success signals an opportunity for the UK to become a global leader in AI-enabled robotics.”
What Still Has To Be Proved
For all the ambition, Machenta is still a small, research-focused company, with an 11-person team according to Dealroom, and no near-term plan to commercialise the technology. Its demonstrations so far cover tasks such as packing, assembly and tidying, and the company says it is now working to scale the model to more precise and complex tasks, which remain harder than the simpler jobs it has shown publicly.
Machenta has also declined to disclose its valuation, so the $40 million figure is the clearest public measure of investor confidence. For a seed round, it is strikingly large, ranking in the 99.6th percentile of all-time UK robotics seed rounds in Dealroom’s sample of 271 deals, which suggests investors are betting that learning speed, not data volume, will decide the next phase of robotics.
Why It Matters
This matters because the bottleneck in robotics is shifting. Hardware has improved steadily, and robot foundation models can now handle many tasks, but teaching each new task remains slow and expensive. If a robot can genuinely learn from one demonstration, deployment changes from an engineering project into something a worker on a factory floor could do in minutes, and the economics of automation change with it, as in South Korea’s plan to put 1,700 AI robots into public service.
Background
Machenta grew out of Imperial College London’s Robot Learning Lab, where its founders spent six years studying how quickly a robot can learn. The new funding will scale its instant-learning model and grow the team at its new King’s Cross office and laboratory.
Source: Imperial announcement and Dealroom report
0 Comments