1X's NEO Home Robot Now Learns Tasks From Video — No Demonstrations Needed

1X Technologies unveiled a major update to its World Model AI system, enabling the NEO home robot to autonomously perform novel tasks from voice or text prompts without any prior teleoperated demonstrations. The system, called 1XWM, is a physics-grounded video diffusion model pretrained on internet-scale video data and fine-tuned on NEO's sensorimotor logs. By generating predicted video rollouts of future actions and using an inverse dynamics model to translate pixels into motor commands, NEO can visualize what it needs to do and execute accordingly — even for objects and tasks it has never encountered. 1X says this represents a paradigm shift from traditional vision-language-action models, leveraging NEO's human-like body design to close the sim-to-real gap. The company positions this as a critical step toward NEO collecting its own training data autonomously in home environments.