Chelsea Finn

Chelsea Finn has shaped modern robot learning around a demanding premise: machines should acquire new behaviour from limited experience and remain steerable in unfamiliar settings. Her recent work extends that principle from general-purpose manipulation to autonomous surgical research through collaborative, empirically tested systems.

Chelsea Finn is an associate professor of computer science and electrical engineering at Stanford and a co-founder of Physical Intelligence. Her research asks how embodied systems can learn broadly useful skills, adapt from sparse feedback and act safely outside a narrow script. The intellectual foundation is model-agnostic meta-learning, or MAML, which Finn introduced as first author in 2017 to prepare models for rapid adaptation to new tasks. In 2025 and 2026, that agenda moved into larger collaborative systems. She co-authored π0.7, a generalist vision-language-action model conditioned on language and additional multimodal context; the team reported zero-shot transfer across robot embodiments and multi-stage performance in unseen environments. She also contributed to FAST, an action-tokenisation method trained on one million robot trajectories that the paper reported could reduce training time by as much as fivefold when paired with π0. In surgical robotics, Finn co-authored SRT-H, which completed autonomous ex-vivo cholecystectomy trials on eight gallbladders, and SutureBot, whose 1,890-demonstration dataset and goal-conditioned method improved insertion-point accuracy by 59% to 74% over the study baseline. Each result is a team achievement under controlled conditions; Finn’s distinction is the consistent learning framework connecting them.