Teaching a robot a new job can take longer than the job itself. Engineers collect demonstrations, fine-tune models and test the results, often repeating the process whenever the objects or working environment change. London startup Machenta thinks much of that preparation can be replaced with a single example.
The Imperial College London spinout has raised $40 million in seed funding, co-led by Creandum and Union Square Ventures, to develop a robot foundation model that learns unfamiliar physical tasks after watching them performed once. The company announced the round on October 27 alongside its first detailed look at the technology.
Machenta’s approach begins with simulation rather than warehouses full of recorded robot movements. Its model is pre-trained entirely on synthetic data, then uses a real-world demonstration to determine what action to take. The company calls this a vision-demo-action model. Unlike systems that need additional task-specific training, it is designed to respond immediately to the example and continue adapting as the scene changes.
In demonstrations released by Machenta, robots sort objects, handle tools and work together on a packing line after being shown the relevant task. The company says the same model operates across six robot platforms with different cameras and teaching methods. Early tests also show it controlling humanoid hands, although that remains a preliminary extension rather than a proven full-body humanoid deployment.
The idea grew out of six years of research led by co-founder and CEO Edward Johns, a professor of robot learning at Imperial, and co-founder Vitalis Vosylius. Their bet is that learning how to learn can be trained at scale in virtual environments, reducing the costly collection of real-world examples that has become central to the Physical AI industry.
There are important limits. Machenta has not disclosed broad independent benchmark results, and the company acknowledges that precise manipulation and complex movements remain difficult. It is still research-focused, with no near-term commercial launch planned.
The funding will support a larger team and more computing capacity as Machenta works to expand the range of tasks its model can handle.
