Humanoid robot advances remain far from everyday use

High-profile predictions from Elon Musk, Marc Andreessen, and Jensen Huang have cast humanoid robots as near-term general-purpose workers, but many roboticists remain skeptical. Techniques that power chatbots do not translate cleanly to robots operating in homes, workplaces, and other messy physical settings, where small failures can undermine an entire task.

Robot control systems have moved from hand-coded rules toward vision-language-action models, including Google DeepMind’s Gemini Robotics and Physical Intelligence’s π0 series. These systems can interpret visual inputs and commands, then guide robot movement, but they remain limited to tasks close to their training data. In robotics, 70% success rates are still considered failures because real-world tasks demand far greater reliability.

World models, which are trained to predict physical outcomes, are being positioned as a possible next step. Fei-Fei Li’s World Labs and Yann LeCun’s AMI Labs each raised $1 billion, but researchers including Yann LeCun and Agility Robotics’ Jonathan Hurst argue that simply adding more data will not solve general robotic capability.

China leads current humanoid robot production, shipping nearly 90% of the 15,000 humanoid robots sold in 2025. Hurst estimates useful autonomous home robots are still roughly a decade away.