Physical Intelligence releases π0, a generalist vision-language-action model

Physical Intelligence/π model family

Physical Intelligence released π0, a generalist robot policy pairing a vision-language model with a flow-matching action module. Trained on data from 8 different robots, it performs dexterous tasks such as laundry folding, table bussing and box assembly, and can be fine-tuned for new skills.

• π0 trained on diverse robot platform data; handles dexterous manipulation across multiple robot types.
• Demonstrated tasks: laundry folding, table clearing, box assembly, and other household chores.
• Architecture: vision-language model + flow-matching action module for continuous motor control.
• Fine-tunable foundation model enabling rapid task adaptation with limited additional data.