Researchers plug GPT-6 Astra directly into a robot and let it clean up an unfamiliar kitchen
No training layer. Stanford and Caltech's HomeBody system lets GPT-6 Astra call directly into robot skills—and it worked in an unfamiliar kitchen without task-specific tuning.
Why it matters
Direct language-model control of embodied AI, bypassing traditional learned control policies, demonstrates a simpler path to generalizable robot autonomy. This matters for practitioners building agentic robotics: it suggests end-to-end LLM-to-action can scale without custom layers per task.
The key facts
5 to knowStanford and Caltech research; GPT-6 Astra powers HomeBody system
Robot performed independent kitchen tidying in unfamiliar environment
Skipped specialized control layer; LLM calls modular skills (grasping, navigation) directly
No task-specific training required; generalizes across domains
Published in The Decoder; date Sep 27, 2026
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Researchers from Stanford and Caltech had a humanoid robot powered by GPT-6 Astra independently tidy up an unfamiliar kitchen. Their HomeBody system skips a specially trained control layer, letting the language model call directly into modular skills like grasping and navigating.