The skill was mastered thanks to the new multimodal Helix system. With its help, the bot is able to simultaneously perceive speech, react to its surroundings, and perform tasks such as folding towels.

To implement Helix, the startup had to put two brains into the robot. The first is a fast LLM with 80 million parameters that directly controls movements, and the second is a slow model with 7 billion parameters that is responsible for speech, vision, and conditional thinking