Humanoid robots seem to be popping up in every YouTube channel, news report, and movie. But is the humanoid revolution imminent, or is it just a party trick?
We wanted to find out, so Soon’s Dagsen Love picked up a $53,000 Unitree humanoid robot from America’s first humanoid robot store to see how easy it is for a non-technical person to program and train one. To help him on his robo-journey, he spoke with a team of geniuses building robot operating systems.
With just 12 hours, can someone train a robot to do their bidding? Or will it end up as a 77-pound paperweight?
00:00 Humanoids are everywhere, how hard are they to use? 00:57 Meeting a robot operating system developer 02:12 What can a Unitree humanoid robot do straight out of the box? 03:25 Visiting America’s first humanoid robot store 04:11 Unpacking the robot 04:50 Taking the robot for a run 05:22 Programming the robot 06:31 Calling in engineering experts 07:15 Speaking Chinese with a robot 07:27 The history of humanoid robots 08:40 A breakthrough 09:09 Robot vs obstacle course
The video explores the current reality of humanoid robots, contrasting viral social media clips with the actual difficulty of operating them. The creator attempts to transform a Unitree G1 robot from a basic machine into an autonomous agent by installing the Dimensional operating system, highlighting the gap between hardware availability and software usability.
The media makes it seem like humanoid robots are inevitable.
Mr. Beast hosted his own Robot Olympics.
There's a robot in Jackass now.
They're sorting mail.
And they're even beating human marathon records in China.
But what is it like to actually pick up a $53,000 humanoid robot from a robo dealer,
turn it on, and try to have it do your bidding?
How hard are these things to use?
That's why I'm going to meet a team
building an operating system to give these robots life.
If LM hallucinates, you may be able to do some data.
If a robot hallucinates, you kill someone.
Then I'm going to pick up my robot,
a Unitrix G1 EDU Plus U2 from San Francisco's most illustrious robo dealer.
So this will prepare you when you become a mobster in the future.
And finally, I'm going to experiment with this robot
to see if 12 hours of hacking gets me my own robot enforcer
or simply a 77-pound paperweight.
Let's see how this goes.
So we are here today with Stas Pometre, the founder of Dimensional.
And Stas, would you like to explain what Dimensional is?
We're basically the operating system for robotics.
This is like our main robot testing facility.
What is this? Who makes this?
This is Agibot.
The robot has been switched to the stable state.
And you just control this from your smartphone.
Wow.
Holy.
That's it.
Yeah, this is why you want it to be not human-operated.
So what is your vision for Dimensional?
I think the eventuality and the inevitability here is that
robots will be in every home, every office.
There will be as many robots as personal computers.
We're seeing this incredible adoption of robots.
All of them need all this tooling that does not exist today.
We now have thousands of developers and hundreds of companies
using Dimensional to give their agents physical bodies.
Wow.
If you were to compare this to the computing era,
what era of Microsoft or Windows would you say we're in?
We're in like 1982, 1983.
There was a gap at the time between the first hobbyist platforms
and the first real mass market platforms.
We're like about to ride up this hockey stick.
I'm about to turn my robot on for the very first time.
Let's get it.
I have a Unitree for the next 24 hours.
What is the coolest thing I can do with it?
Yeah, that's a good question.
Can this robot be my butler?
What kind of use cases?
Say like I have a tray over here of animals.
Can this robot dance?
Yeah.
That is a policy you have to train like with RL.
Could you define what a policy is, please?
Yeah, a policy makes robots do things.
Could a Unitree robot walk across a balance beam?
Yeah, if you train a policy to do so.
Can this robot do push-ups?
No.
If you train a policy, you could.
If you train a policy.
I imagine this will be an answer to the rest of our questions.
Yes, actually.
Can this robot do a jumping jack?
It could, if you train a policy to do so.
Nice.
Can this robot complete an obstacle course?
Uh, yeah.
Yeah, I can.
Can this robot fight?
Yeah.
Yeah, out of the box it could fight?
Not out of the box.
These Unitree robots, out of the box,
how much can you actually do out of the box?
They cannot yet respond to national language commands.
They're not building any memory or understanding of your physical space.
They don't come with navigation out of the box.
They don't come with any hosted telegraph features out of the box.
We're in the process, obviously, of making everything, all of our modules,
all of our agent tech logic, incredibly foolproof,
so that anyone can program these things out of the box.
Despite all of these insane robot clips online,
it turns out these things can't do much straight out of the box.
So, obviously, I had to try my hand at setting one of these doohickeys up myself.
So I'm here in Lower Pack Heights,
and we're about to go meet with this robot dealer named Six.
We rented a robot from him.
We're going to load dimensional on it,
and we're going to see what these humanoids are really capable of doing.
Dude, this is exciting stuff.
Well, welcome to the first humanoid store in the United States,
or what will become the first humanoid store in the United States, the Rec Shop.
Wow, and what was the inspiration behind the store?
So we've been doing a humanoid robot fight club.
Oh my god!
We realized that we needed a public space that we could do these fights regularly,
and it kind of evolved into a store.
Oh, and how is the build-out been coming along?
Well, so we just signed the lease a week and a half ago,
so you ready to see your boy?
I am so excited.
So you're going to have to name him.
Yo.
So this is the Unitree G1.
You kind of want to have two people help you out with this.
Holy, I'm too weak for this guy.
The vibe coding class is going to have to get used to picking up robots.
Just put it here.
Yep, just hook this in.
We're about to battery up our very first robot.
Yep, we're in.
I'm about to turn my robot on for the very first time that I'm going to name him.
Let's get it.
Hold it.
There you go.
Wow.
And you can hear that fan get started.
Arms are getting a bit heavier, it seems.
I'll give you a step-by-step on how to control him for the first time.
Yes.
We're going to activate something called running mode.
Carabiners are off.
There you go.
And now it's like an RC car.
Just go for it.
Wow.
We are controlling the robot for the first time.
I have the RC controller.
This is my little buddy, and his name is Arthur.
That's his name, like King Arthur.
Just run him as fast as you want.
I'm coming for you.
The most impressive thing about this is how much it's designed to balance itself.
Wow.
Let's go back to the developer zone.
Nice, Arthur.
Controlling my very own humanoid robot is pretty cash money,
but I really want to push the limits of what this robot can do agentically with agents.
So I dropped out of high school at 15.
I don't really know anything.
I have codex, I have clawed code.
How difficult do you think this is going to be?
15?
Of course.
If you vibe code a robot incorrectly, it's not going to just do something slightly wrong.
It could rip its own leg off.
So you have to download an app called Unitree Explorer.
This allows you to interface with the robot.
We can have you connect into dev mode, straight into the computer, straight into the robot.
We're in.
Okay.
Time for vibe coding.
We fired up the robot.
We SSH'd in.
We're troubleshooting some errors.
I'm working with my AI agents on how we can get this robot running agentic.
Just do everything.
Let's see how much AI can help you.
Let's see how much AI does now.
All right.
So we got dimensional running.
Look right here.
Look at how beautiful this is.
I've never installed this before.
So this is a...
Apply network optimization.
Why not?
Programming a robot is legit hard.
Especially if you have zero coding or robotics experience.
So I call in some backup.
This failure right here, system dependency's failed.
It's good that I'm here because all the problems are my fault.
So I can be here to take the blame.
Now I kind of expected to hit one more error here.
Yeah, there we go.
So have you done robotics stuff before?
No.
None at all?
Not at all.
Is it going to work?
Oh, this is interesting.
Errors.
Permission denied.
We are currently around six hours in.
And dimensional is almost fully installed.
I've never had this happen.
This is proving to be quite a bit more difficult than I expected.
As Six Born does.
Wow.
Oh, check that out.
What's going on over here?
Wow.
Thankfully, Six was giving some private demos of his robots.
Which served as a nice break from all this darn vibe coding.
I asked, do you prefer English or Chinese?
And he says Chinese.
If it feels like humanoid robots have suddenly taken over your feed, you're not wrong.
But the story actually starts more than 50 years ago.
If you want to trace today's humanoid robots back to their roots,
it starts in Japan in 1973 with a robot called WABOT-1.
Fast forward 27 years and Honda unveils ASIMO.
For a while, this was the robot.
It convinced an entire generation we all have robot assistants by now.
Then Boston Dynamics entered the chat.
Their robot, ATLAS, went from a clunky research project to doing parkour,
backflips, and things that honestly look like CGI.
But the biggest breakthrough was AI.
Because for decades, the challenge was not just getting robots to move,
it was getting them to understand the world around them.
Now, companies like Tesla, Figure, and Unitry are combining advanced robotics
with powerful AI models to build humanoids that can learn, adapt, and perform real-world tasks.
After more than 50 years of research,
humanoid robots are finally leaving the lab and entering the real world.
And this time, they might actually stick.
Hopefully, you'll get the most running.
You're very close right now.
You've also got an achievement, which is finding an error I've never seen before.
Wow!
I just really want an agent to control my robot.
After more than 10 hours trying to run the dimensional operating system on Arthur...
It's working.
It's working!
Yo!
Wow!
Look at their visualization right there.
Live data right here, as you can see.
So it detects obstacles.
You can tell it where to go.
And then as you put stuff in its way, it'll make a new path to find some way to get there.
And then later, that's the foundation for doing agentic stuff.
There we go.
Now it's dead.
Well, maybe we'll...
Oh, yo, yo, yo!
Oh, wait, wait, the cable.
We want to go over here.
We want to take the robot on the obstacle course now.
We got it!
Right now, I'm just routing him through the obstacle course manually so that he has a map of it.
So he can complete it autonomously on his next route.
This might be a little high for him.
And as you can see on my screen, he's mapping.
We have radar showing the entire map of the territory.
I can't believe we got this working, to be honest.
Like, what?
Now that Arthur has made it through the obstacle course under my control...
No one can defeat us.
It's time to let him go through it on his own.
This is all agentic.
Yo, what?
Oh my god.
He's a little confused.
All autonomous.
This is incredible.
I have no idea what I'm doing, by the way.
So I spent around 12 hours with this Unitry G1, my little buddy named Arthur.
I love having a robot.
I think that if I had $53,000 lying around, I'd probably buy one and f*** around with it.
I was able to get Dimensional up and running and get him agentically walking.
I accomplished my goals.
It was a lot harder than I expected.
A lot of random glitches and bugs.
Stick him back to his workshop.
All of the videos of the robots doing cool things on social media,
you must consider the myriad hours real people spent working
to make those robots do those specific things.
Full autonomy is something that is coming soon.
Arthur, bid you farewell, my friend.
May we meet again.
Sorry, Brian.