Over the past eight months we’ve developed a general-purpose robotic foundation model, which we call π0. We hope this will be a first step toward our long-term goal of developing artificial physical intelligence. Like large language models (LLMs), our model is trained on broad and diverse data and can follow various text instructions. Unlike LLMs, it acquires physical intelligence by training on embodied experience from robots, learning to directly output low-level motor commands via a novel architecture. It can control a variety of different robots, and can either be prompted to carry out the desired task, or fine-tuned to specialize it to challenging application scenarios.
