Stanford and Caltech researchers have built HomeBody, a system that lets a humanoid robot clean an unfamiliar kitchen with no trained control layer between the AI and the robot’s motors.The system connects OpenAI’s GPT-6 Astra directly to a Unitree G1 robot. Instead of a learned intermediary, the language model calls a library of skills such as grasping, navigating and opening drawers.First the robot explores the room. It builds a digital twin of the space inside Nvidia’s Isaac Sim and logs object locations into a spatial memory.

That lets it find items even after they leave its camera view.In tests, the robot gathered coffee bags, threw out spoiled milk and juice cartons, and retrieved medicine from a closed drawer it had to locate from memory. The code has been published on GitHub.The team reports limits: the model’s latency slows the action loop, finger servos overheated during long runs, and running a large model in the loop is costly. It is a research demo, not a product.