You don’t expect to find robotics hardware on the sixth floor of an office building in central London. Arms, legs, torsos, teleoperation rigs, Meta Quest headsets, engineers hunched over modified joysticks training machines to pick up stuff … that all feels kind industrial, maybe suburban-warehouse … not downtown in one of the world's great cities.

But that’s exactly where I found myself a couple of weeks ago, walking the floors with Sotirios Stasinopoulos, the chief product officer of Humanoid, a UK startup that has quietly become one of Europe’s most credible entries in the global race to build humanoid robots.

That cramped-but-central setup could be a metaphor for the whole company.

While Figure, 1X and a wave of Chinese manufacturers soak up the headlines and the mega-rounds, Europe has been building its own contenders: Humanoid in the UK, Neura in Germany, a few others in Italy and elsewhere. In most cases, they operate with a fraction of the capital of their competitors.

That probably leads to a lot more urgency about actually shipping. Humanoid, founded in 2024 by former jewelry magnate Artem Sokolov and now 130+ people, has already signed deployment deals with Schaeffler, Bosch and Siemens. It’s launched several humanoid robots. They’re both called HMND 01, but it comes in a wheeled and a bipedal form factor. And it’s done that without the billion-dollar war chest its best-known rivals are wielding.

So how are they moving this fast? A few things stood out.

A bet on wheels, not legs

The company builds two platforms — a wheeled mobile manipulator, the HMND 01, and a bipedal robot.

But something like 90% of the engineering goes into the wheeled version.

“More than 85 or 90% of the use cases can be covered by a wheel-based platform,” Stasinopoulos told me. I’ve now heard some version of that line from multiple robotics companies, and it rings true. Wheels are more energy efficient. They don’t waste power just standing upright. A wheeled robot has good reach from a solid based with a low center of gravity … there’s a lot of reasons.

But a major reason is regulatory, and this is the part a lot of the hype coverage misses.

“There’s no ISO at this stage that can cover bipedal technology for industrial frameworks, because at any time you may fall,” he said. The committees working on it might have something “late next year or early ‘28.”

The wheeled platform, meanwhile, can be certified today by borrowing from two existing frameworks — the one for autonomous mobile robots and the one for collaborative industrial robots.

Humanoid is targeting a fully certified product by late 2027.

The biped isn’t dead. It’s being redirected toward home and service uses on a longer timeline … where, notably, the safety bar is different. “You definitely don’t want a big heavy metal thing falling on a pet or a baby,” Stasinopoulos said.

Hard to argue with that.

There’s something almost refreshingly unsexy about this. The flashy demos are always the walking robots. Humanoid looked at the market and essentially said: legs are a research project, wheels are a business.

The brain is where it gets interesting

If the hardware philosophy is pragmatic, the AI architecture is where the company’s engineering DNA shows.

Humanoid organizes its intelligence into four layers, numbered — in a very European way — from System 3 down to System 0. At the top, System 3 is an agentic fleet coordinator that pulls tasks from a customer’s existing warehouse or ERP software and distributes them across robots, knowing each machine’s location, charge and current tooling. System 2 is a reasoning layer running off-the-shelf vision-language models (they use Gemini, among others) that breaks each task into a checkable, step-by-step workflow. System 1 is Humanoid’s own vision-language-action model that executes the physical actions. And System 0 is the whole-body controller that actually moves the limbs.

The philosophy is fleets, not units.

“If you only want a robot to do one thing, you might as well get an old-school automation solution,” Stasinopoulos said.

The pitch is a workforce as flexible as people: stocking shelves in the morning, feeding machines in the afternoon, reorganizing inventory overnight. System 3 is in charge: all the other systems function to get the work done.

Speed is the critical question

No-one’s putting robots in a factory if they’re slow.

Humanoid says its robots now hit around 80% of human speed and success on some core tasks, up from roughly 60%.

The goal is getting faster than humans, and for that, Humanoid uses reinforcement learning.

The robots work overnight, System 2 automatically grades which attempts succeeded, and the wins get fed back into a shared model that every robot inherits. “One robot will do something better, System 2 says this is a good execution, retrain the base model, and then all the robots get the improved model,” he said. “The progress is exponential.”

And unlike imitation learning, which caps out at human performance, “reinforcement learning does not have this limit. This is unlocking even superhuman performance.”

Second-mover advantage

Stasinopoulos credits Humanoid’s pace of development to two things.

First, relentless commercial focus:

“We don’t develop technology and see how it can be used.” From day one the company chased big industrial customers with a slide deck and designed to their requirements.

Second, what he called “second-mover advantage:”

Humanoid’s team includes 50-plus veterans from Boston Dynamics, Sanctuary AI, 1X and others. “We’re making use of all the knowledge and all the mishaps that other companies had to pay millions or billions of dollars to learn.”

That’s kind of a cheat code.

I’ve been sort of sitting on this one for a month, and I’m so glad I can finally talk about it.

Back in June I interviewed Dar Sleeper, head of product design at 1X. He showed me video of the company’s new robot hands and told me to keep it under wraps. Hard to do, because what he showed me was genuinely stunning: fast, precise, weirdly human motion. I dropped a hint in my story and then had to shut up about it.

Now I can talk about it. So here’s what NEO’s new hands can actually do … click play;

Most robot hands are pretty numb. You tell them where to go, they go, but no data comes back. NEO’s are different. They run on tendons at very low gearing, which means every joint can feel.

Push on a finger and it kinda pushes back and tells you exactly how hard you pushed. 1X calls it “force transparency.”

What that buys you: NEO can pick a grape off the stem, install a light bulb, use a screwdriver, plug in its own charger, zip a jacket, and pick a single screw off the floor.

That’s impressive.

It can also lift something like 20 pounds and push a loaded cart. Delicate and strong, which is a combination that’s always been hard. There’s durability too: testing to 2 million cycles under load, 1X says.

Maybe my favorite detail, though?

These hands are washable.

IP68 sealed, food-safe, the works. NEO can make you an egg, chop your salad, and then go rinse its hands at the sink. I don’t know of another robot on earth that can do that. This is genuinely some kind of holy grail (yeah, there’s lot of them in robotics) for a robot.

It’s a big update from the days when your Roomba saw your pet’s you-know-what on the floor, tried to clean it up, and ended up just spreading it all over your house.

One line from Dar stuck with me.

He said NEO will ship “with more hardware capability than the AI currently knows how to use.” Translation: the hands are ahead of the brain, and software updates will keep making your robot smarter over time.

That’s a good place to be.

I’ve been covering robot hands for years, and this is the closest I’ve seen anyone get to the real thing. After watching these work, “a robot that does your chores” starts to sound more like a promise than a wish.

I hope so, anyway.

More soon.

- john

This is literally the year when humanoid robots start to leave the lab and appear in our homes. Our living rooms. Our kitchens.

Just sit and think about that for a moment.

Crazy times, no?

Neo, from 1X, will start shipping to homes this year. So I sat down with Dar Sleeper, head of product and design at 1X (and former Tesla), one of the very few people on…

Read more

I heard today that some universities are going back to old-school paper and pen exams. No phones, no laptops, nothing digital. We’re back to walking uphill both ways 😀

That would totally kill my hand, but it would force me to actually learn stuff and be able to express it on demand.

Is that what we should be doing in education?

Last week on TechFirst, I s…

Read more

You probably didn’t know you needed a robot dog to do yoga with. I didn't know that.

But the man who has shipped more robots into homes than anyone alive is betting that half the projected $5 trillion physical AI market will be for companions, not drone-like machines that just do work.

That dog, btw? It’s a quadruped, not a dog.

Maybe closer to a bear tha…

Read more

I’m standing in a 300-year old warehouse in Boston where the sailcloth for the USS Constitution (yeah, that Constitution) was made. I’m with about a hundred humanoid robots.

Most of them are failing miserably at their jobs.

They’re called Sonny. None of them have been here longer than two months. Some arrived a few days ago. They’re picking, packing, sort…

Read more

That pic I’m using for this newsletter? Yeah, totally misleading. Because AI in the factory doesn’t mean a factory with all machines, robots, and humanoids doing all the work and AI doing all the thinking.

That’s still science fiction.

We might get there in 5 or 10 years, hopefully with an evolved social consciousness about work, jobs, careers, and univer…

Read more

Fourteen months ago, quantum navigation didn’t exist. Not in any real sense: it hadn’t been validated, hadn’t been tested, hadn’t been proven outside a pristine lab with floating concrete slabs and enough vibration isolation to survive a plane crash next door.

So anytime someone wanted to hack navigation on a battlefield, in a city, or over a country, th…

Read more

I’ve been covering tech for a long time. Long enough to have watched the hype cycle eat itself, spit itself out, and then demand a second helping. (Thirds, anyone?) AI-washing, crypto-washing, metaverse-washing: brands always want to present themselves as part of the hot new wave. So do insta-gurus on LinkedIn.

The newest hype leader?

Agent-washing.

I rece…

Read more

You can’t tell a robot how tight to hold a potato chip. That’s the whole problem.

Try this right now: place your fingers exactly 23.4 millimeters apart.

You can’t do it. There’s no ruler in your brain, no internal coordinate system, and no precision positioning system running in your motor cortex.

And yet you can pick up a potato chip without breaking it o…

Read more

Everyone is talking about AI agents.

My local one is named Claw (so original!), lives in a Mac Mini (also so original!) and is about to start a long and hopefully productive career in fetching news articles and building ecosystem maps for me.

Most of the agentic AI conversation is vague, though: buzzwords and bold promises without a ton of substance. Or …

Read more

Pretty obviously, AI agents are super-hot. There’s so much they can do, it’s literally a game-changer. My Mac Mini is on order.

But the security is a nightmare.

You give an AI system access to your tools, your files, and your accounts, then ask it to complete real work. The AI begins acting on your behalf. It reads your emails, writes code for you, calls …

Read more

Humans are ridiculously good at learning.

I mean, it doesn’t feel that way while we’re trying to learn algebra or solve the quadratic equation, but show someone how to do something once and there’s a decent chance they can repeat it. Show them twice and they’ll probably get it right most of the time. By the fifth try they’re totally got it and they’re te…

Read more

They plugged the AI character into the robot. It could talk. It could move. It could be in space.

After, it said:

“It felt good to feel.”

Is that just a chunk of math and code? Or is it the beginning, possibly, of something much more?

Check out our conversation here (and subscribe to the TechFirst YouTube channel to never miss an episode):

Read more

Is your AI agent operating a humanoid robot to build airplanes in France while you sleep?

If not, you might just be living in the past.

That’s the intentionally-hyperbolic opener to my recent TechFirst with Jansen Teng, CEO/cofounder of Virtuals Protocol.

He’s connecting agents to robots … so AI agents are gonna start doing doing real physical work in the …

Read more

I’m currently vibe-coding a fitness app, which also comes with a website. I generated the website in about an hour of work. Because I’m not a designer, I cheated and got AI to design it for me.

The result?

Instantly average.

That sounds bad, but think about it. That’s actually a big win for me … average is way better than I could achieve on my own feeble a…

Read more

Machine learning, machine learning, machine learning. We’ve spent the last few years obsessing over what AI can learn. What about machine UNlearning?

Machine learning and AI have exploded in our massive current LLMs. That’s meant bigger models. Trillion-plus parameters. More data, more compute, more capabilities … more more, essentially.

But there’s an un…

Read more

I had one of the most enjoyable podcasting experiences ever this week: doing a live podcast on the show floor at Web Summit Qatar with Amy Peck, CEO of EndeavorXR. Live just feels different. In-person just feels more real. And I’ve been wanting to dig into the future of XR for a while now …

Maybe the biggest take-away from our conversation?

XR isn’t tryin…

Read more

Look, the pitch for humanoid robots in the home has been pretty simple and pretty attractive: they’ll do the work so we don’t have to.

Fold the laundry. Mop the floors. Load the dishwasher.

Perfect, right?

A guilt-free labor droid. Life without the drear and grind of daily chores like cooking and cleaning. Basically, everyone would be able to live like th…

Read more

AI accelerates creativity. AI accelerates production. AI accelerates distribution. (And yeah, sure, AI also democratizes mediocrity.)

Let’s be honest: AI is hitting entertainment like a sledgehammer. From algorithmic gatekeepers and AI-written scripts to digital actors and entire movies generated from a prompt, everything is changing.

I recently had a cha…

Read more