Where we've been, and where we're going.

In time, a robot will be developed that can learn any physical task just like a human. As a team, we have wondered for some time when that moment will be, and how it may come about.

In Machenta’s formative years at Imperial College, we studied the foundations of one-shot robot learning huddled together in a university lab. There, we all felt that something wasn’t quite right about robots needing hundreds of manual demonstrations to learn just a single task.

So whilst the world set off on a physical data collection race, we aimed for something more intriguing: a robot brain that could learn instantly. Every experiment and every paper got us closer. And when we found early signs of life in a recipe that could scale with compute alone, by pre-training entirely in a simulated world, we doubled down on our mission and formed Machenta together.

As a company, we now see this same vision through a new lens. Whilst those formative years were driven by academic curiosity, these first few months at Machenta are driven by conviction around impact: that the greatest opportunities in robotics will come from making robots radically easier to teach.

We often remind ourselves of what the early days of LLMs were like, and how curiosity grew precisely because their intelligence was so quick and easy to access. Something extraordinary then followed. Curiosity gave way to experimentation, experimentation gave way to adoption, and adoption has led to an unprecedented era of imagination and creativity. We believe that to now create the same story in physical AI, instant robot learning is where we begin.