Stanford's HomeBody lets VLM-driven Unitree G1 tidy rooms autonomously
rsasaki0109 · x · 2026-10-12
Stanford and Caltech researchers present HomeBody, a system that gives frontier VLMs persistent spatial memory and a composable humanoid skill library for long-horizon tasks.
- Key idea: instead of the usual System 2 VLM → learned System 1 VLA → System 0 controller chain, a plug-and-play VLM directly orchestrates reusable motor skills, with execution feedback letting it revise decisions.
- Results: a Unitree G1 guided by GPT Astra tidied up across an unseen kitchen and retrieved a remembered object from an underspecified request — no environment-specific training data or extra policy learning.
- Authors: Gio Huh (Caltech), Cayden Gu, Takara E. Truong, C. Karen Liu, Guy Tevet (Stanford).
Related event: Stanford's HomeBody Gives VLM-Powered Humanoids Spatial Memory(2 posts)→
More from Embodied
- ARK analyst predicts Cybercab hits 10% of Austin ridehail before locals notice, 70% before the world does — skorusARK · 2026-10-12
- DeepSeek reportedly leads $400M round for Chinese robotics AI startup — AIFlow_ML · 2026-10-12
- Hugging Face team launches a friendly RL competition for its Microduck robot — Thom_Wolf · 2026-10-12
- NTU's MAMHOI factorizes scene-aware human-object interaction via affordances — nanyang-technological-university-singapore · 2026-10-12
- Dev used Claude Code agents to turn Quake (1996) into a first-class VR experience — SuperV1234 · 2026-10-12
- MIT's Draft tool lets engineers design and tweak robots by adjusting parameters — Jasonio · 2026-10-12