Humanoid Robots, VLAs & World Models: Where Robotics Really Is (October 2026) | Industry Pulse EP1

Voice-over: AI clone of Lukas Timm's voice, script written by Lukas.

Cold open

Okay, so this is robotics. Robots falling on stage. Falling while dancing. Falling at their own premiere.

And then this happened. Figure trained its old robots to jump. Into a furnace. Because Arnold said so. And honestly, the jumps are better than most of the walking we just saw.

[Arnold Schwarzenegger, Figure video: "Hasta la vista, Figure 02."]

Same year. Robots doing backflips and two-metre jumps. A robot cleaning a kitchen alone for four minutes. And Tesla building robots on the old Model S line.

So which one is it? Terminator, or robot butler? Let's probe.

[Jensen Huang, NVIDIA, CES 2026: "Look at all these friends that we have building robots."]

One year. From around the corner, to here. Let's see if he's right.

Probe 1 · The brain

Probe one: the brain. Basically, someone took the idea inside ChatGPT and changed the output. Text in, text out, became camera plus sentence in, motor commands out. They call it a VLA. Vision, language, action.

Figure's Helix two. Four minutes alone in a kitchen, sixty-one actions, nobody remote controlling it. Figure says one network replaced more than a hundred thousand lines of hand-written code.

I spent like ten years in automotive compliance, tracing requirements for code like that. So, that's basically my old job. Deleted.

Physical Intelligence, pi zero point seven. It runs an air fryer it was never trained on, by combining skills it already has. And Google DeepMind, Gemini Robotics two. A whole humanoid, from toe to fingertips.

But be honest. Picking something up from the floor? Less than half the time. So no robot butler yet. But the direction is clear.

Probe 2 · World models

Probe two: world models. A model that predicts what happens next. Push this, it falls. Pull the drawer, it opens. So the robot can practice in its head, before it touches anything.

Google DeepMind's Genie four, announced this Tuesday. NVIDIA's Cosmos, with open weights. And Meta's V-JEPA two, whose robot version learned from under sixty-two hours of robot data.

Why this matters? Language models had the whole internet to learn from. Robots don't have an internet of movements. So either you film millions of hours of humans. Or the robot dreams them.

This is my version. My office, scanned with an Ouster lidar, as a simulation. So my robots can mess up in there first.

Probe 3 · The garage

Probe three, and that's my favourite. You don't need a billion dollars anymore to play.

Hugging Face just open-sourced humanoid legs. Mostly 3D printed, about two and a half thousand dollars. Just legs, for now. And robot brains are now an API. Physical Intelligence opened theirs on October thirteenth. Dream Labs, six days later.

Built this month

And here's what we built ourselves this month. Our own iToF depth camera, calibrated against RGB. The SLAM map, snapped into a Gaussian splat of the room. And Heron. Started as a four-legged robot. Now it wants to be a flamingo.

And this is Ecko. A robot arm with an old Samsung phone as a head, in my office in Hamburg. I tell him, Kartier den Raum. Map the room.

First try, I'm on the joystick. Disengage, twice. Second try, same thing. Third try...

He made it. That's what building actually looks like. Not the trailer.

Probe 4 · The speed

Probe number four: the speed. Just look at October.

Robot brain as an API. Ninety million dollar seed round. Seed! Tesla's first production robots. A forty million seed for robots that learn a task from one single demo. And AGIBOT wants to measure robots in work hours, like employees.

That's one month. When I say exponential, I mean the robotics news cycle now feels like LLMs in twenty twenty-three.

Probe 5 · What's real

Last probe: what's real? Because the hype is real. The revenue, mostly, isn't. Yet.

[Elon Musk, CNBC: Optimus could be "the biggest product ever of any kind."]

Tesla's Optimus robots at the Cybercab event? Remote controlled. The viral rogue robot from Indonesia? Staged. And the furnace? Part IP protection, part marketing genius.

So when you see a robot video, ask three things. Is someone controlling it? How many takes? And can I buy it? Most of the time: yes, a lot, and no.

Where this is going

So where is this going? Everybody is chasing one thing. A robot brain that generalizes. New task, new room, new body, and it just figures it out. Like you, in a stranger's kitchen, still finding the cups.

Terminator or robot butler, that's being decided right now. By labs worth billions, and by people in their garage. And the tools are open, and on an API.

So I'm building it. In my office. With a phone as a head. This was Industry Pulse. Every month: what happened, what's real, what's hype, and what I'm building with it. Tell me which lab I should probe next.