London startup Machenta raised a $40 million seed round to train robots entirely in computer simulation. Imperial College London reported the round, co-led by Creandum and Union Square Ventures, as the largest-ever seed raise by a UK robotics startup Imperial announcement.
Machenta is a 2025 spinout of Imperial College London founded by robot learning professor Edward Johns and former PhD student Vitalis Vosylius. It is building a robot foundation model for Instant Robot Learning. A foundation model here means one large pre-trained model that can handle many different tasks.
Table of Contents
- What is Instant Robot Learning?
- How does simulation-only training work?
- What have the robots been shown doing?
- What will the $40 million pay for?
What is Instant Robot Learning?
Instant Robot Learning means a robot learns a new physical task from a single demonstration with no further training. Machenta describes the goal as showing the task once and deploying immediately company explanation.
That would replace hours or days of per-task data collection and retraining. For a reader running automation, the practical change is simple:.
- show the desired handling task one time
- deploy the robot on that task without a new training cycle
How does simulation-only training work?
The model is pre-trained entirely in computer simulation without any real-world training data. Machenta says it learns how to learn from synthetic object interactions rather than memorizing fixed tasks company explanation.
In practice, simulation supplies many varied grasps, placements, and contacts. The robot then uses that broad preparation to interpret one real demonstration. The demonstration tells it what to do, while simulation has taught it how to adapt.
What have the robots been shown doing?
Machenta demonstrations include packing and closing assembly-line boxes. Other demonstrations show tidying tasks across several robot types, including humanoids.
The cross-robot detail matters for buyers and integrators. The same learned approach is meant to transfer, not stay tied to one arm or body. Packing, closing, and tidying point first toward pick-and-place work.
What will the $40 million pay for?
The company plans to expand hiring and scale computing power and simulated training data rather than commercialize immediately. SiliconReport reported those plans and noted that Machenta declined to disclose valuation funding report.
The team now stands at 11 people. More compute and more simulated interactions come before a product sale.
