

Chelsea Finn: Building Robots That Can Do Anything
40 snips Jul 22, 2025
Chelsea Finn, an Assistant Professor at Stanford and co-founder of Physical Intelligence, discusses her journey in robotics from MIT to leading innovative projects today. She explains how her team is developing robots capable of adaptive learning in messy real-world settings. Topics include their groundbreaking work on robotic grasping, overcoming challenges in task generalization, and the use of large datasets to teach robots physical common sense. Finn shares insights on the surprising moments in data and the pivotal advancements that make intelligent robotic assistance possible.
AI Snips
Chapters
Transcript
Episode notes
General Purpose Robots With Foundation Models
- Developing general purpose robots is crucial to avoid building a new company for each application.
- Foundation models like in language are promising for versatile robotic tasks.
Scale Is Necessary But Insufficient
- Scale is necessary but not sufficient for effective robotic models.
- Diverse, relevant data is more important than massive scale alone.
Training Robot Laundry Folding
- Started folding single-brand shirts successfully before tackling harder crumpled laundry.
- Pre-training then fine-tuning on curated high-quality data unlocked consistent folding ability.