Chelsea Finn

Chelsea Finn earned her PhD at Berkeley under Sergey Levine and Pieter Abbeel, where her 2017 MAML paper — model-agnostic meta-learning, a method for teaching networks to learn new tasks from a handful of examples — became one of the most-cited algorithms in machine learning. As a Stanford professor she built the IRIS lab around robots that learn from experience, co-leading large-scale efforts like RT-2 and the ALOHA manipulation systems with Google DeepMind collaborators.

A foundation model for the physical world

In March 2024 Finn co-founded Physical Intelligence with Karol Hausman, Sergey Levine, Brian Ichter, and Lachy Groom, aiming to do for robotics what large language models did for text: one generalist model, any robot, any task. The startup released its first robot policy model, π0, that October — demonstrating laundry folding and table bussing across different robot bodies — and raised $400 million at a $2.4 billion valuation from backers including Jeff Bezos and OpenAI, followed by a further $600 million round in late 2025.

Finn is frank that the field’s bottleneck is data — there is no internet of robot motions to scrape — and that generalist robots remain a long-term project. Asked about competition, she has said she worries less about rivals than that “no one will solve the problem.” Her bet is that diverse robot data plus scale will crack physical intelligence the way text data cracked language.