π0 is a general-purpose robotic foundation model designed to enable AI systems to gain physical intelligence through embodied training, allowing them to perform various tasks similar to large language models and chatbot assistants. π0 acquires physical intelligence through hands-on experience on robots, capable of directly outputting low-level motor commands to control multiple types of robots, and can be fine-tuned for specific application scenarios. The development of π0 represents a significant advancement in the application of artificial intelligence in the physical world, offering the most capable and dexterous general-purpose robotic policies to date by integrating large-scale multitasking and multi-robot data collection with novel network architectures.