Most robotics founders are hoarding real-world video. Edward Johns went the other direction and deleted it entirely.
Machenta, the London company he co-founded with former PhD student Vitalis Vosylius, emerged from stealth with $40M in seed funding co-led by Creandum and Union Square Ventures. It is the largest seed round a UK robotics company has raised. The company declined to give a valuation.
The pitch rests on one claim: train a model purely inside simulated environments, then hand it a robot that can copy a physical task after a single demonstration. In footage shared with reporters, three machines pack and seal boxes on a line after watching the motion once.
Rivals are buying data
Competitors have taken the opposite path. Physical Intelligence is valued near $11B, FieldAI is raising at $10B, and Skild AI has passed $14B while reporting $100M in annual recurring revenue from more than 60 customers.
Johns, who spent six years on the question at Imperial College, argues data volume is the wrong lever. His lab’s early robots needed roughly 30 minutes of extra sensing before they could scoop an object with a spatula. The gap he wanted to close was the wait itself.
His framing is that intelligence shows up less in what a system can already do than in how fast it absorbs something new.
A dozen people and no product
Machenta employs about 12 people in King’s Cross and is not rushing to sell. Precise movements and complex assembly still defeat the model, which learns instant tasks reliably but struggles with fine control.
The cash goes toward compute and simulation breadth, plus headcount. Johns describes the method as physical vibe coding, a phrase that lands better with venture investors than with factory line managers.
