Stanford and Caltech Plug GPT-6 Astra Directly Into a Robot to Tidy Unfamiliar Kitchens

The Decoder · rss · 2026-09-27

Researchers from Stanford and Caltech present HomeBody, a system that let a humanoid robot powered by GPT-6 Astra independently tidy up an unfamiliar kitchen.

The key design choice: HomeBody skips a specially trained control layer entirely, letting the language model call modular skills like grasping and navigation directly. This removes scene-specific fine-tuning and gives the robot out-of-the-box generalization to unseen environments.

The demo adds evidence for the 'frontier LLM as the robot brain' approach, potentially reducing embodied AI's dependence on large robot-specific training datasets.

Original post →

More from Embodied

Embodied channel →