Physical Intelligence co-founder building general-purpose robot models
Chelsea Finn
Profile
Chelsea Finn is an assistant professor at Stanford (CS and EE) and co-founder of Physical Intelligence — the robotics foundation-model startup she launched in 2024 with her former PhD advisor Sergey Levine, Karol Hausman, and Lachy Groom. Her Stanford IRIS lab studies how robots can learn general-purpose skills from experience rather than hand-coded behaviors. If you care about robots that actually adapt to new environments — not demos on stage, but machines that fold laundry in a kitchen they’ve never seen — she’s one of the handful of researchers whose work matters most.
She’s best known for MAML (Model-Agnostic Meta-Learning), the algorithm she introduced in 2017 as a PhD student at Berkeley under Levine and Pieter Abbeel. MAML is one of the most cited ML papers of the last decade: a clean idea — train a model so that a few gradient steps on a new task yield good performance — that works across classification, regression, and reinforcement learning. It reframed “learning to learn” from a curiosity into a practical tool, and it’s still the default baseline anyone compares against when they publish new meta-learning work.
More recently, her group has been at the center of the shift toward robot foundation models. She co-authored π0, Physical Intelligence’s vision-language-action flow model that generalizes across dozens of dexterous manipulation tasks, and she helped lead the ALOHA and Mobile ALOHA projects — low-cost teleoperation rigs that made high-quality bimanual imitation learning accessible outside of big-lab budgets. The combination matters: cheap hardware for data collection plus a generalist policy trained across embodiments is what a real “GPT moment for robotics” is going to look like, if it comes.
Her honors include the Presidential Early Career Award for Scientists and Engineers (PECASE, 2025), a Sloan Research Fellowship (2023), the IEEE RAS Early Academic Career Award (2022), and the ACM Doctoral Dissertation Award. For developers trying to reason about where robotics is heading, Finn is a clean signal: rigorous academic output, shipping code on GitHub, and a commercial bet on generalist policies that she’s willing to put her name on.
Key Articles & Papers
Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks π0: A Vision-Language-Action Flow Model for General Robot Control Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (ALOHA) End-to-End Training of Deep Visuomotor Policies Learning to Learn with Gradients (PhD dissertation) IRIS Lab — Chelsea Finn, StanfordVideos
YouTube
Spotify Podcasts
