Stanford and Caltech connect GPT Astra to a Unitree G1 to clean an unfamiliar kitchen

Stanford and Caltech built HomeBody so a Unitree G1 can navigate an unknown kitchen, tidy up, and pull items from drawers using GPT Astra without a separate trained control layer—the model calls grasping, navigation, and drawer skills directly. It scouts the room, builds a digital twin in Nvidia Isaac Sim, stores object locations, plans steps, and fixes mistakes. Limits include Astra latency, overheating finger servos, and high compute; the code is on GitHub.