AD Research Hub
DashboardPeoplePapersWorkshopsDatasetsResearch Map
Login
⌘K
AD Research Hub — Anomaly Detection in Computer Vision
← Back to People

Chelsea Finn

Stanford University

Frontier Research Maph-index: 5452 citations
HomepageSemantic Scholar
Frontier Research Map

Featured Work

Foundation Models for Robotics

official keynote page — 2024

Why Now

She belongs in the younger frontier layer because she has been one of the clearest voices on how large pretrained models meet embodiment, adaptation, and control.

Key Ideas

  • -Foundation models for robotics are only valuable if they adapt online to changing tasks, embodiments, and environments.
  • -The frontier challenge is not only pretraining but turning broad priors into reliable control policies.
  • -Generalization in robotics needs mechanisms for fast adaptation, not just larger datasets.

Open Questions

  • ?What parts of a robotics stack should be foundation models versus fast task-specific adapters?
  • ?How do we combine broad pretrained priors with safety and sample efficiency in the real world?
  • ?Can one model family handle language, perception, action, and adaptation without becoming too brittle?
Younger Agenda-Setters and Adjacent ML Thinkershigh confidence
Cross-References

Themes

roboticsfoundation modelsadaptation

Sign in to access this content

Sign in