1X has launched a new World Model Lab aimed at speeding the development of humanoid robots that can operate with far less human supervision. The company says the lab will focus on building smarter models for its Neo robot by training on richer, more varied data drawn from robots operating in the real world.
The effort builds on 1X’s earlier World Model, introduced in January, which used video-based training to help Neo translate prompts into actions, including tasks involving objects it had not previously encountered. With manufacturing ramping up and more robots entering production, the company says it now has access to a much larger stream of data that can be used to improve its systems.
1X chief executive Bernt Børnich said the company does not believe fine-tuning alone will lead to artificial general intelligence. Instead, he argues that major AI advances come from training on better data from the start, rather than layering robot-specific learning on top of models built for other domains.
Sam Sinha, who has been hired to lead the new lab as head of world models, made a similar case. A former founding researcher at video-generation startup Luma AI, Sinha said robotics has often been treated as an afterthought in AI development. In his view, humanoid systems need to be trained on their most important signals from the beginning, rather than being adapted later with a limited set of robot demonstrations.
The company says the training mix will include live visual streams, joint and force readings, pressure data from Neo’s hands, simulation, and large collections of human video. 1X is betting that a robot designed to resemble the human body more closely will be able to learn more effectively from those human sources.
That thinking explains several design choices in Neo, including tendon-driven movement rather than gears and a hand with 22 actuated degrees of freedom. The company says the goal is to close the gap between human behavior in video and the robot’s own physical capabilities.
For 1X, the lab is not just an AI initiative. It is tied directly to the company’s hardware strategy. Børnich said the robot and the model cannot be developed separately because the physical design affects what the AI can learn and do. He argued that the company’s integrated approach to manufacturing is necessary to generate the data needed to improve the models.
1X manufactures its Neo robots at its facility in Hayward, California, and says that producing at scale helps create a feedback loop. More robots in the field means more data, which helps train better models, which in turn should make the robots more capable and commercially useful.
Sinha described that loop as a potential competitive advantage, calling it a data moat. He compared the strategy to software products that were shipped before they were perfect so they could gather user interactions and improve over time.
1X is entering a crowded field. Other companies including Physical Intelligence, Google DeepMind, NVIDIA, Figure, Apptronik, Unitree and Agibot are all developing robot foundation models or humanoid systems of their own. Some Chinese manufacturers already have significant production scale, which gives them a head start in collecting training data.
Even so, 1X says its advantage lies in combining a human-like body with close integration between hardware and AI. Børnich said the company can move from major design changes to a robot coming off the production line in about four weeks, which he described as a fast iteration cycle for the sector.
The company also says it plans to begin shipping 20,000 preordered humanoid robots this year. It expects the new lab to produce early results before the end of 2026, with additional improvements arriving through software updates after deployment.
Børnich said Neo should deliver something useful with full autonomy by the end of the year, while more advanced capabilities may arrive later. The company is positioning the lab and the robot together as part of a longer-term push toward machines that can handle a wider range of human tasks with increasing independence.