A humanoid robot recently showed up for sale online.
It’s called the Unitree R1. It stands roughly 4 feet tall, weighs around 55 pounds and has 26 joints.
Image: Unitree
The Unitree R1 can walk, balance and recover from a fall. In demo videos, it runs and even does basic acrobatics.
And you can order one for as little as $4,900.
Meanwhile, Kia just confirmed it’s preparing to deploy Boston Dynamics’ Atlas humanoid robot inside one of its U.S. factories, expanding on similar plans from parent company Hyundai, which is expected to begin using Atlas at its Georgia plant in 2028.
Image: Boston Dynamics
The robots will start with repetitive and higher-risk tasks on the production line.
Tesla is also pushing into this space with its Optimus robot, with plans to produce thousands of units for internal use before expanding availability.
And in China, Chery’s robotics subsidiary, AiMOGA, has started selling its own humanoid robot online for about $41,000, with deliveries scheduled this year.
Image: Chery
It’s pretty incredible to think that in 2026 you can compare prices on humanoid robots the way you compare prices on a car.
It’s starting to look like a real market.
But there are still a few kinks that need to be worked out before these machines can be trusted on the job…
Or in your home.
Robots Everywhere
Unitree shipped more than 5,000 humanoid robots last year. That’s far more than most competitors, which are still operating in the dozens or low hundreds.
But based on forecasts for the humanoid robot market, you can soon expect that number to grow exponentially.
Goldman Sachs sees it reaching roughly $38 billion by 2035. It also expects the cost of humanoid robots to fall by roughly 40% over the next decade, which is a big reason broader deployment will become possible.
Other estimates run significantly higher depending on how fast the market expands into logistics, retail and the home.
The hardware is getting cheaper, and demand is starting to build. Coming out of CES, I was convinced the timeline for robots had moved up.
But there’s a problem.
You see, humanoid robots still can’t reliably handle objects in unpredictable environments.
Walking used to be their major roadblock, but it isn’t anymore. Today’s humanoids can run, jump and even do backflips. They can balance, recover and navigate well enough to be useful in even some difficult environments.
That means they carry a fixed part from one station to another.
But give them a slightly unfamiliar object, change the angle, the lighting or the position, and things can fall apart quickly. Something as simple as picking up a loose cable or aligning a connector can still cause problems.
The issue is coordination.
We still haven’t given robots the ability to connect perception, decision-making and movement in real time, without needing a perfectly staged environment.
Because it’s harder than it sounds.
AI systems have made rapid progress in language and vision because they were trained on massive datasets pulled from the internet. Robotics doesn’t have an equivalent dataset for physical interaction. There’s no large-scale library of touch, force and movement that covers the variability of the real world.
So robots learn slowly, task by task, environment by environment.
That’s why some systems still rely on teleoperation for complex work. When Figure AI introduced its humanoid robot, early demonstrations showed human operators stepping in behind the scenes to guide tasks. 1X Technologies has taken a similar approach with its Neo robot, using remote human control to handle situations where autonomy breaks down.
It’s also why reliability drops when robots move outside of tightly controlled workflows.
And it explains the disconnect in the market right now.
On one level, humanoid robots are making real progress. Prices are coming down and manufacturers are preparing to deploy them at scale. The global market for humanoid robotics is expected to explode over the next decade.
But the hardest part is still being built.
The ability for a robot to walk into an unfamiliar environment, understand what it’s seeing and act correctly without retraining is not solved.
Until it is, these machines will be most useful in places where the world can be simplified for them. Structured environments like factories and warehouses where variability is limited and tasks can be tightly defined.
That’s enough to support a market.
But it’s not enough to support the broader vision people have in mind when they think about humanoid robots moving through everyday life.
Here’s My Take
Humanoid robots are making progress fast.
You can now buy one for a few thousand dollars, and some companies are preparing to deploy them at scale.
This represents the early stage of adoption. But there are still many miles to go on this journey.
Right now, humanoid robots are mostly confined to factories and warehouses because those environments remove a lot of uncertainty.
But the real opportunity here is much bigger.
For humanoid robots to move beyond those environments, they need to handle the kind of variability we humans deal with every day without thinking about it.
But how do you teach a machine to understand and act in the physical world as easily as we do?
In my next issue, I’ll show you one very promising way Big Tech is trying to answer that question.
But in the meantime, I want to hear from you. Would you welcome a humanoid robot into your home?
Shoot me an email to [email protected] with your opinion on the pending robot revolution.
We won’t reveal your full name in the event we publish a response.
So let me know what you think!
Regards,
Ian KingChief Strategist, Banyan Hill Publishing
