Nye's Digital Lab is a weekly scribble on creativity in an age of rapid change.
Essay #90
I'm elbow-deep in robots, and just about to get my first one moving.
I did what I had to. I vibe-coded it's behavior using a game engine.
I described a motion in plain English, let the model rough out the joint commands, ran it... aaaaand watched a virtual robot collapse like a folding chair.
Frustrated, but not stopped. I fixed it. Ran it again. And again. Eventually a little virtual humanoid robot lifted its foot and stepped.
I sat back and thought: am I doing robotics right now? Hm. I'm doing kinetics, animation, and behavior trees, on a programmable body. I'm speaking it to life?
Hold that. First, some context.
Somewhere right now a consultant is explaining that you shouldn't buy a robot. You should subscribe to one.
Robots as a Service. Or ..."RaaS." It which sounds like something you'd mutter after stubbing your toe. It joins a proud tradition: software as a service, infrastructure as a service, your car's heated seats as a service. Tech spent twenty years learning that anything can be rented instead of sold, because a subscription sells one thing forever. And now, with AI staring down rising compute costs and models that lose money on every query, the robot companies have looked at all of it and landed on the same monthly invoice, this time with arms.
I get it. If you run a warehouse, a rented robot is one you don't have to buy, fix, or defend in a budget meeting. It's a flat fee, like a streaming service that can occasionally drop a pallet on your foot. The market is already worth tens of billions and expected to more than double by 2030. But the whole model rests on one assumption: that robots stay expensive and mysterious.
If my midnight experiment is any indication, neither will hold. My bet: training a robot will be about as hard as launching Photoshop.
We will all be able to do it.
While a lot of American robotics companies were shooting demo videos, Chinese companies were shipping. Last year they built roughly nine out of every ten humanoid robots on Earth. Unitree alone moved more than five thousand, and its cheapest humanoid now starts around $4,900 (basically the cost of a used Honda.) Then a few of them did the genuinely wild thing: they gave away the hard part, open-sourcing the models that turn a spoken instruction into movement. The mysterious piece, which is the one that justified the subscription, is becoming a free download.
We watched this in the AI business over the last two years. When the smart part becomes free, the value slides to the edge, to what you can teach it to do.
Which is the corner I've been playing in. Booster Robotics, the company whose robot I was coaching at midnight, was started by Tsinghua engineers with a school-first, open-source attitude. Open SDK, open training tools, parts you're allowed to understand. Their K1 runs under five grand.
And my favorite fact in all of robotics right now: at RoboCup last summer, a student team from Leipzig ported the soccer code they'd written for an older robot onto Booster K1s and won the world championship 11 to nothing. They didn't have better machines than the labs they beat. They had access, and each other.
College students. Open code. A world title. I feel like this is the way personal computing was a scene in 1978, and it answers the RaaS question for me:
assemble the body, and let the community build the soul.
Now back to that step, because I haven't been able to stop thinking about it.
When personal computers arrived they were engineering objects. Command lines, beige boxes, machines built by engineers for engineers. What made them personal was the day designers showed up and asked a different question. Not "does it run," but "how should this thing and a human get along?" Interface design and UX weren't jobs, and then suddenly they were the difference between a computer and something people loved.
A robot is a computer that stands in the room with you. Every UX question comes back in three dimensions, with your body in the loop. And my midnight step convinced me these questions get answered the way I answered mine: as craft decisions about motion. How fast a hand reaches tells you whether the machine is confident or malfunctioning. A pause before a turn is the difference between "excuse me" and getting loomed over. Whether a robot about to move shows you first, that little anticipatory lean, decides whether you relax around it or never do.
None of that is wiring. It's staging, timing, character. It's the syllabus of every animation and theater program in the country, pointed at a new kind of computing. The engineers will make robots that work. Someone else has to make robots that read. That's the job, and almost nobody is training for it, and my students, who are watching a third of the traditional design postings evaporate, are sitting on exactly the skills it requires without knowing it.
It's a new computing paradigm.
It's midnight again as I finish this, and the sim is open in the other window.
The next thing I want to teach it is a dance routine. Can it improvise by randomizing actions, or do I chain actions together to build personality? What should his head be doing? Where are the limits of a platform like this?
Somebody is going to figure it out on a five-thousand-dollar open-source robot in a dorm room or a garage, the same way a generation of kids figured out what software should feel like before anyone thought to call it UX.
The tools are cheap.
The code is open. And right now, the seat next to me is empty.
Though, right now I gotta go to bed.
The craft is just beginning. Make something move.
Make it Happen.
Nye Warburton is a creative technologist and educator from Savannah, Georgia. He improvises essays aloud and captures the best takes into an Obsidian Vault. From there, he performs agentic alchemy on the writing datasets using a variety of AI models.
Some of last years best essays are available in a printed edition at
Learning from Zero - On starting over again
Engines are the Next Frontier - AI + Game Engines will be nuts.
Community OS - On Communities developing their own operating system.
Nye
Comments