Figure AI has unveiled its novel dual-system AI architecture, Helix, which enhances humanoid robots by enabling them to understand natural language commands and manipulate unfamiliar objects without prior training. This architecture blends high-level reasoning with real-time motor controls, facilitating spontaneous learning and adaptation for robots in everyday tasks. The system features a high-parameter vision-language model that processes speech and visuals at a slower rate paired with a rapid visuomotor policy for instantaneous movement. This innovation allows robots to collaborate efficiently, as demonstrated when two robots successfully stored grocery items despite lacking prior knowledge about the items’ function. Figure AI has begun integrating this technology commercially and has obtained significant funding to expand its operation, projecting a path towards deploying 100,000 robots in the coming years.

Source 🔗