Chelsea Finn
Influential PI at Stanford IRIS Lab with foundational contributions to meta-learning, imitation learning, and VLA development.
“Two years ago, I founded a company called Physical Intelligence. And we're really interested in how we can basically develop any robot or allow any robot to do any task in the real world.”
Source→“In real-world experiments, we design three tasks covering single-arm placement, precision bimanual stacking, and long-horizon sequential coordination, where XS-VLA improves average task success from 21.7% to 65.0%”
Source→“Mobile ALOHA (Fu, Zhao, and Finn 2024) platform... In real-world experiments, we design three tasks covering single-arm placement, precision bimanual stacking, and long-horizon sequential coordination”
Source→“Chelsea Finn...Referenced as co-author of Ctrl-World (Guo et al. 2026)...Her Ctrl-World model was one of the three external world models evaluated”
Source→“Finn is one of the most influential researchers in robot learning, with foundational contributions to meta-learning (MAML), imitation learning, and now VLA development (co-author on π0).”
Source→“FRS is an inference-time mechanism to unlock that latent knowledge without retraining the base model.”
Source→“Authors: Andy Tang, William Chen, Andrew Wagenmaker, Chelsea Finn, Sergey Levine (Stanford + UC Berkeley). Date: June 2025. arXiv: 2606.13675.”
Source→“CHORUS: Decentralized Multi-Embodiment Collaboration with One VLA Policy — Stanford University | arXiv:2606.12352 | June 2026”
Source→“Co-author listed on CHORUS; also cited as co-author on OpenVLA, Mobile ALOHA, and π0.5 backbone (References)”
Source→“Chelsea Finn (RT-1, RT-2, ALOHA)... These are the intellectual ancestors of the approach, and their frameworks constitute the building blocks RDGen assembles.”
Source→“Senior author; her group's focus on generalization and sample efficiency is directly reflected in the paper's core thesis: 'The ability to efficiently and reliably learn new tasks has been a foundational challenge in robotics'.”
Source→“Stanford researchers have cracked a critical bottleneck in physical AI deployment: how to take a pretrained robot foundation model and push it to 100% task reliability in under 20 minutes of real robot time.”
Source→“Stanford / OpenVLA Team (Kim et al., 2024), Academic origin of OpenVLA, the open-source VLA baseline. Referenced as part of the broader VLA landscape being addressed.”
Source→“He was seeing over and over papers coming from Chelsea Finn and Sergey Levin's lab.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.