Physical Intelligence Robot Makes Coffee for Journalist in San Francisco Warehouse Test
A Physical Intelligence robot prepared coffee and performed household tasks using a vision-language-action model. The San Francisco start-up is testing its system in simulated and real homes.
app.buzzsumo.comA journalist visiting Physical Intelligence’s San Francisco warehouse received a cup of coffee prepared entirely by one of the company’s robots. The same system has also learned to fold clothes, peel vegetables, and clean kitchens. Physical Intelligence was founded in 2024 and operates from a warehouse lined with gleaming steel panels.
It maintains two additional warehouses that contain mock supermarkets, bedrooms, and kitchens renovated weekly to expose its robots to varied settings. Employees were observed teaching the machines to fold shirts, place pillows on shelves, and cut bows on gift boxes. The company is also sending robots into occupied homes to test performance amid everyday clutter.
The robots rely on vision-language-action models that convert spoken instructions into sequences of physical actions. 7, cooked sweet potatoes in an air fryer after receiving step-by-step verbal guidance, despite never having used the appliance before.
Sergey Levine, a founder and University of California, Berkeley professor, said the company seeks an adaptable control system that works across many machines rather than a single robot design.
He noted that Physical Intelligence has operated for two years. Ingmar Posner of the University of Oxford said vision-language-action models represent the most direct transfer of large-language-model capabilities to robotics. He added that real-world deployment at commercial scale remains distant because users often test robots in unpredictable ways.
@NewScientist reported the demonstration and the company’s training approach.
