Machenta

Instant Robot Learning.

[Machenta @Machenta_]

Steerability with Instant Robot Learning.

[Edward Johns @Ed__Johns]

Steerability is important in robot learning!

When teaching robots using a pre-trained model, it's important to show that the model is "steerable".

This means that, given the same set of objects, you can teach the robot different interactions just by providing different demonstrations.

If this is possible, it shows that the model is genuinely conditioning its behaviour on the teaching demonstrations, rather than simply performing one of the tasks learned during pre-training.

At @Machenta_, we've built our model to be instantly steerable from a single demonstration.

The video below shows this: using the same objects, we teach the robot three different interactions just by changing the demonstration each time.

https://t.co/P1woKrXtjH

[Machenta @Machenta_]

One-Shot, One-Person Factory with Instant Robot Learning.

[Edward Johns @Ed__Johns]

We’ve built a one-shot, one-person factory!

We taught a robot assembly line using Instant Robot Learning at @Machenta_.

Normally, this would take weeks of system integration, but it took us two minutes to teach.

See the video below 👇

Three different bimanual robots are collaborating on a packaging task. All three robots:
• Have different camera setups.
• Were taught using different teaching modes.
But they all run the same end-to-end model.

The model has never seen these objects before. It was trained 100% in simulation, and these objects were not in the simulation.

I was also asked about this one-person factory at @MTSlive. Check out our conversation below!

[Machenta @Machenta_]

Reactivity & Recovery with Instant Robot Learning.

[Edward Johns @Ed__Johns]

Quick video showing some of the "Reactivity & Recovery" capabilities of Instant Robot Learning at @Machenta_:

1. One real demonstration: putting blue object in grey box and closing box.
2. Robot reacts in real time when object and box are moved around, with a closed-loop end-to-end policy.
3. Robot recovers when object is pulled out of gripper, and re-grasps it.
4. Robot notices that object is removed from box, and places it back in.
5. Robot notices that box lid is opened, and re-closes it.

The important part:
1. None of this reactivity and recovery needed to be shown in the demonstration: it all comes from the intelligence of the pre-trained model.
2. Model has never seen these objects before. It is trained 100% in simulation, and these objects were not in the simulation.

https://t.co/P1woKrXtjH

[Machenta @Machenta_]

Instant Robot Learning on @MTSlive.

[Edward Johns @Ed__Johns]

Thanks @MTSlive for the chat last night about Instant Robot Learning! Here's a little clip about why we're excited at @Machenta_ to be building AI for robots that learn instantly.

[Ilir Aliu @IlirAliu_]

A robot learns a new task. Instantly.
From... one demonstration.

Yes, you read that right. One❗️

But what's really crazy:

It's never seen these objects before!!

@Machenta_ developed Instant Robot Learning, using a Vision-Demo-Action foundation model trained solely in simulation.

The model enables robots to learn manipulation tasks instantly from one demonstration with no further training, supporting reactivity, recovery, steerability, and cross-embodiment across varied robots including humanoids.

If you watch the full video (which you should), demonstrations show one-person automation of assembly lines in minutes, compositional generalization to novel objects, and rapid iteration via physical prompts.

Where is this tech coming from?

@Ed__Johns and his team (He is an Imperial College professor and Machenta CEO) were building on prior lab work like Instant Policy for efficient sim-to-real transfer.

Earlier today, they announced the company's $40M seed round co-led by @creandum and @USV.

Congrats Edward Johns, to you and your team; looking forward to what you guys are building!!

[Katharina Wilhelm @Kathi_Wilhelm]

Physical vibe coding is here.

@Machenta_ brings instant learning to robots: show a robot a task once and it gets to work. You prompt it with a demonstration the way you prompt ChatGPT with text.

While most of robotics races to collect more real-world data, @Ed__Johns and @vitalisvos19 spent the past years at @imperialcollege on the opposite bet: a model trained entirely in simulation that learns how to learn.

That means it’s bounded by compute, not data, and it will only get better as it scales.

We believe the hardest problems are often solved by people who bring a different lens. Ed is a true polymath: after his PhD, he spent a year as a singer-songwriter, playing gigs with a then still unknown Ed Sheeran. In a band, when someone shows you a chord, you rarely need to see it twice. Perhaps that’s why Ed asked a different question from most of his field: not how to make robots learn more, but how humans learn so fast.

At @creandum, we're proud to co-lead Machenta’s $40M seed with @usv, backing Ed, Vitalis and the team.

[Business Insider @BusinessInsider]

Machenta, founded by Imperial College London researchers, raised $40 million from Union Square Ventures and Creandum to build AI that teaches robots new tasks from one demonstration. https://t.co/NE2TXLDJUG

[Rebecca Kaden @rebeccakaden]

Intelligence ultimately isn’t about breadth of knowledge but how well we learn something new. The same will be true in robotics.

Robotics has been in a scale of data dynamic. Thrilled to work with @Ed__Johns and the @Machenta_ team as they bring Instant Robot Learning to life and fast intelligence to the physical world. Show a robot something once and they know how to do it. From the best in class team that invented this method together. Let’s go!

@usv

[Lukas Ziegler @lukas_m_ziegler]

That’s sick. 🔥

Robots that learn a new task in a minute.

When I first met @Ed__Johns in London, he described it to me as “physical vibe coding".

Demonstrate once and the robot just learns it.

Machenta just raised a $40M seed, the largest robotics seed round ever in the UK.

Proud to back another outstanding physical AI company as an angel investor among @creandum & @usv

UK IS COOKING IN ROBOTICS RECENTLY!

[Machenta @Machenta_]

Here we go...! 🚀

[Edward Johns @Ed__Johns]

Today we’re out of stealth, and we’ve raised a $40m seed round co-led by @creandum and @usv, to bring Instant Robot Learning to life.

For the past few months, we‘ve been busy building a new company.

Machenta.

This follows results from a new kind of robot foundation model we’ve built, which I’m so excited to share with you today!

Usually, it takes hours/days of data collection and training to teach a robot.

We’ve built a model that learns tasks instantly: one demonstration, no further training.

A model trained entirely in simulation, without a single pixel of real-world data.

A model where teaching robots is so quick and easy, it feels like physical vibe coding.

A model for robots that learn like humans do, from someone like you.

Welcome to the era of Instant Robot Learning.

Thank you @Kathi_Wilhelm, @rebeccakaden, @hkanji, @jonno_evans and all of our angels for your support!

@Machenta_

Learn more: https://t.co/P1woKrXtjH

See below for what you can do with Instant Robot Learning!