If you saw a Post-it note stuck on the wall in your office, you'd probably pick it up and read it, almost instinctively.
When Bernt Børnich asked Neo, his humanoid robot, to look at a Post-it note on the wall, that's exactly what happened. Neo hadn't been trained explicitly for such a task, nor was there a bespoke program for peeling paper off paint, holding it at the right distance and reading it aloud.
"There's a moment where you see something like that, and your brain starts interpolating forward," Børnich told me. "You realize where this is going."
Don't miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.
The tech world is on a journey toward robots that look, move and behave like humans that fit into everyday environments with us, and Børnich sees momentum building. He's the CEO of 1X, the California-based company that built Neo and is working on a lot more robots just like it.
Neo had a splashy introduction last fall, where it was billed as a general-purpose home robot, a human helper rather than a factory machine. It's early adopter technology, with a $20,000 price tag and delivery to customers promised loosely for sometime this year. Early demos relied heavily on people operating Neo by remote control.
Now, with the advent of what 1X calls its World Model -- a computer program that acts as a virtual simulator for the environment around a robot -- Børnich says the company is finally showing why it built Neo the way it did. Not as a robot trained to do a fixed list of chores, but as one designed to understand the world well enough to figure things out on its own.
The moment may be well-timed. At CES earlier this month, robots were all the rage. Eye-opening demos included LG's AI-enabled humanoid home robot that could bake and grab beverages, Boston Dynamics' agile, industrious Atlas humanoid (which won a CNET Best of CES award), robot vacuums with legs that can climb stairs, and bartending robots that can make drinks. Companies like Tesla, Figure, Agility Robotics and Sanctuary are all chasing versions of the same idea: humanoid robots that can move beyond narrow, preprogrammed tasks.
How Neo learns about the world around it
Folding laundry is on the checklist for what Neo's expected to do around the house.
Neo, which stands about 5 feet, 6 inches tall, walks at a sedate pace, taking short, deliberate steps. In videos shared by 1X, when Neo reaches for objects, its movements are slow and careful. It picks things up with a light touch, adjusting its grip rather than clamping down.
That's intentional. Neo is designed to be physically similar to a human, not just in shape, but in how it interacts with the world.
"The world was designed for humans," Børnich said. "And our World Model is trained on human interactions. Neo's similarity to the human form is what lets that knowledge transfer."
This is the core bet behind 1X's approach. Instead of relying on massive amounts of custom robot training data, the company designed a robot that's expected to learn from the world around it. Neo has five-fingered hands. It opens doors in the way a person would. It reaches, bends and balances using motions that look familiar because they are.
Neo's training data is publicly available in internet videos showing humans doing everyday things. Not perfectly staged demos, but messy, real-world interactions like opening cabinets, folding clothes and handling fragile objects. According to 1X, its World Model is based on roughly a million hours of general video, then adapted using hundreds of hours of first-person footage, including video captured from Neo's own cameras, along with smaller amounts of data collected from the robot operating in real environments.
Customers concerned about privacy will be able to choose whether the video captured by a Neo in their home is used to improve 1X's models. In the long term, the company hopes opting out won't meaningfully degrade the experience. In the near term, however, Børnich says that users who decline data sharing should expect more limited capabilities, because the system is still learning how to generalize across homes and situations.
What the World Model does
Give Neo a prompt and it'll use 1X's World Model to figure how how to get the job done.
World models aren't unique to 1X. In AI research, the term generally refers to systems that learn how a real-world environment changes over time so they can predict outcomes. It's a 3D understanding, in contrast to the language-driven models behind ChatGPT and Gemini. If you ask Neo to pick up a glass, it simulates the motion in advance, planning for a favorable outcome -- in this case, not breaking the glass.
"With this update, Neo plans its actions by simulating what might happen in the real world," Børnich said. "It can anticipate applying too much force or knocking something over and adjust its behavior to avoid accidents rather than reacting after the fact."
Børnich offers an analogy. A person can walk up to a washing machine they've never seen before and still figure out how to open it. They look for hinges, handles or locks, and if one approach fails, they try another. That's not memorization. It's reasoning about how objects tend to work.
Neo, he says, is starting to show that kind of behavior. It doesn't always succeed on the first attempt, and it's often slower than a person. Much slower, in fact -- taking minutes instead of seconds to conduct tasks. But it can approach unfamiliar situations with a plan rather than just freezing or failing outright.
One important limitation is memory. Neo can remember the layout of a home or where rooms are located, Børnich said, since that kind of spatial context is handled as part of normal operation. But what's still unresolved is richer long-term memory that blends sight, sound and touch, the kind that would let Neo reliably recall past conversations or personal preferences over time. That deeper, experience-based memory remains an active area of research.
Neo's limitations
1X is aiming to get Neo into customer homes sometime in 2026.
When the Wall Street Journal's Joanna Stern visited 1X late last year, Neo did almost nothing on its own. It attempted precisely two tasks autonomously. Every other movement was teleoperated by a human wearing a VR headset.
That appears to be changing.
In a video announcing the World Model earlier this month, everything shown was done autonomously, according to 1X. Neo puts toast in a toaster, removes an air fryer basket, opens a sliding glass door, waters plants, wipes tables, packs a lunch box, irons and steams shirts, plunges a toilet, rolls out dough, opens bags of chips, stacks blocks, organizes fruit and even makes a heart shape with its hands. None of those actions were pre-scripted routines. And none involved a human behind the curtain, according to the company.
"The right way to think about it is you can ask it to do anything, and Neo will generate a new ability," Børnich said. "It will attempt it."
In his own home, Børnich says he uses Neo autonomously to wipe counters, open the front door for guests, put dishes away, water plants and tidy living spaces. That doesn't mean Neo works flawlessly or quickly -- tasks may take minutes rather than seconds. But it does mean autonomy is no longer theoretical.
Teleoperation isn't going away, though. When Neo runs into cases it can't yet solve on its own, a remote human operator will step in using a virtual-reality headset and controllers, effectively seeing through the robot's eyes and guiding its movements.
Compared to the cost of training and running large AI models, Børnich says, teleoperation is a small operational expense, and one that should shrink as autonomy improves. The company wouldn't disclose the number of teleoperators on the team.
When will customers get their Neos?
Neo has a soft fabric covering and a minimalist esthetic.
Those who want to try out Neo in their own homes can put down a $200 deposit, and Børnich says more than 10,000 units have already been reserved. Those first customers won't be getting a finished consumer appliance.
"People really understand this is early," Børnich said. "They want to be part of the journey."
Børnich says he's "increasingly confident" that 1X can deliver a robot that's fully autonomous out of the box at some point this year. But that's a goal, not a guarantee.
"We've earned an enormous amount of trust from a lot of people," he said. "We take that very seriously."