Physical Intelligence, a hot robotics startup, says its new robot brain can figure out tasks it was never taught | TechCrunch

"Once it crosses that threshold where it goes from only doing exactly the stuff that you collect the data for to actually remixing things in new ways, the capabilities are going up more than linearly with the amount of data."

"The core claim in the paper is compositional generalization - the ability to combine skills learned in different contexts to solve problems the model has never encountered."

Physical Intelligence's new model, π0.7, allows robots to perform tasks they were not explicitly trained for, showcasing compositional generalization. This model represents a step towards a general-purpose robot brain that can learn from plain language instructions. The findings suggest a potential inflection point in robotic AI, similar to advancements seen in large language models. The model's ability to remix learned skills indicates a more favorable scaling property, leading to increased capabilities beyond linear growth with data. A notable demonstration involved an air fryer, which the model had not encountered during training.

Read at TechCrunch