Internet Videos Help Robots Learn Human Movement

Scientists have created a new way to teach robots how to move by using millions of internet videos showing human actions. The breakthrough system, called Humanoid-X, contains 20 million robot poses matched with written descriptions of movements.

This video-based training method aims to make robots better at copying human actions in real-world situations.

The research team has also developed Universal Humanoid-1 (UH-1), a model that lets robots understand and follow voice commands for movement.

"The scalability offered by internet videos allows us to capture the vast nuances of human movement, which is critical for training humanoids capable of operating in dynamic and unpredictable settings," said a researcher working on the project.

The system could improve how robots work in healthcare and emergency response situations.

Researchers now face challenges about data quality and ethical use of public videos in their training system.

The technology marks a shift from older methods that relied on pre-programmed movements and limited datasets.

This development could change how robots interact with humans in daily tasks, though researchers continue to work on improving the system's reliability.