1X launches world model enabling NEO robot to learn tasks by watching videos
1X said its new world model puts it a step closer to a future where robots can teach themselves to do any task a human can do. | Source: 1X Technologies
1X Technologies AS last week announced its latest 1X World Model. The company said the AI update for NEO enables the humanoid robot to turn any request into an AI capability on demand, using a video model grounded in real-world physics.
“After years of developing our World Model and making NEO’s design as close to human as possible, NEO can now learn from internet-scale video and apply that knowledge directly to the physical world,” said Bernt Børnich, founder and CEO of 1X. “With the ability to transform any prompt into new actions—even without prior examples—this marks the starting point of NEO’s ability to teach itself to master nearly anything you could think to ask.”
With this update, 1X said NEO can use video data fine-tuned on robot data to perform AI tasks, even with objects and environments it has never encountered before.
The Palo Alto, Calif.-based company said it has designed NEO for household use. The humanoid is available through 1X’s early access program for $20,000, which includes priority delivery in 2026. A subscription model will also be available for $499/month.
NEO can do old tasks in unfamiliar places, plus learn new ones
With this update, users can give NEO a simple voice or text prompt, and the robot uses what it’s looking at to generate visualizations of future actions. A built-in inverse dynamics model then translates these into precise movements for NEO to complete the request, 1X explained.
“With the 1X World Model, you can turn any prompt into a fully autonomous robot action — even with tasks and objects NEO’s never seen before,” said Daniel Ho, an AI researcher at 1X.
Demonstrations in 1X’s latest video (below) demonstrate NEO’s ability to generalize beyond training data. For simple prompts like packing a lunch box, it visualizes and executes, even with unfamiliar objects.
The robot can also handle completely novel tasks, such as operating a toilet seat, opening a sliding door, ironing a shirt, brushing a human’s hair, and more, without any prior examples in its dataset, said 1X. This highlights the transfer of broad human knowledge through the World Model, 1X said.
1X says robot can learn on its own
Where traditional AI models for humanoid robots have relied on data collected by human operators, the 1X World Model enables NEO to collect its own data and autonomously master new capabilities. This opens the door for robots to eventually teach themselves anything, 1X claimed.
Where improvement in AI capabilities for humanoids has been bottlenecked by the speed in which robot data can be collected by human operators, 1X said its World Model doesn’t only self improve from NEO collecting it’s own data. It also benefits from the improvement of video models, given that the world model uses a video model at its core.
Traditional models have historically struggled with changes in lighting, clutter, or chaos that are commonplace in the home. The 1X World Model applies human-like understanding to navigate extreme variability, maintaining composure amid rapid environmental shifts, according to the company.
The post 1X launches world model enabling NEO robot to learn tasks by watching videos appeared first on The Robot Report.