Stanford HomeBody drops the VLA: GPT Astra directly orchestrates Unitree G1 humanoid skills
CyberRobooo · x · 2026-09-28
Stanford's HomeBody tests a different path to humanoid intelligence: a Unitree G1 explores an unfamiliar kitchen, builds persistent spatial memory and a Real2Sim digital twin, then uses GPT Astra to plan and compose skills like Navigate, Pick, Place and Open Drawer — no environment-specific training data or extra policy learning.
- Key architectural bet: standard stacks chain a System 2 VLM into a learned System 1 VLA; HomeBody removes the VLA and lets the VLM call a plug-and-play library of composable skills
- The VLM decides what to do, skills handle how; execution results feed back so it can revise failed steps
- Demos include tidying the kitchen and retrieving a remembered object from an underspecified request
The open question: as frontier models get better at spatial reasoning and planning, how much of the robot stack still needs custom building?
More from Embodied
- Edge AI chip startup SiMa.ai raises $150M Series C at $1.45B valuation — dauber · 2026-09-29
- Cheap DC motor with harmonic drive: outer gear rides on PTFE tubing — _Stocko_ · 2026-09-29
- C. Elegans Connectome With 302 Neurons Simulated in Real Time on a Smartwatch — bodyaz · 2026-09-29
- Cloudini 1.4 released: 20% better pointcloud compression and slightly faster — facontidavide · 2026-09-29
- LeCun Amplifies SF World Models Reading Club on Oct 10 Featuring AdaJEPA and NVIDIA Robotics — ylecun · 2026-09-29
- Embodied AI is repeating cobots' path: demos proven dead ends a decade ago — yongqianme · 2026-09-29