companies

Physical Intelligence

Develops the pi-zero family of foundation models that let many different robot bodies learn new physical tasks from demonstration.

Physical Intelligence, styled as pi or Pi, is a San Francisco robotics-AI lab founded in 2024 by a group of researchers and engineers from Google DeepMind, Stanford, and UC Berkeley, including Karol Hausman, Sergey Levine, Chelsea Finn, and Brian Ichter. Rather than building its own robot hardware, Pi focuses on foundation models meant to run across many different robot bodies, arms, mobile manipulators, humanoids, the same way a large language model runs across many text tasks. Its flagship line is pi-zero and its successor pi-zero-point-five, vision-language-action models trained on broad demonstration data that output low-level motor commands directly rather than symbolic plans. Public demos have shown a single pi-zero policy folding laundry, bussing tables, assembling boxes, and bagging groceries across different hardware, an unusually broad range for one policy in this field. Pi open-sourced both models through its openpi repository, a move that stood out in a sector where most frontier models stay closed. The company has raised roughly $1.1 billion across seed, Series A, and Series B rounds, reaching a valuation near $5.6 billion, with backers including OpenAI, Thrive Capital, Lux Capital, and Jeff Bezos.

Founded

2024

Headquarters

San Francisco, United States

Sector

lab