Research topic
Background: Offline reinforcement learning and distribution shift, Conservative learning and offline-to-online improvement. Evaluation limits and transfer uncertainty remain explicit.
Independent educational guide. Not affiliated with or endorsed by the universities, professors or laboratories described here. This collection reflects the material currently mapped on Socratic Learn.
Berkeley RAIL research studies robot learning from demonstrations, reward, prior datasets, and interaction: policies, reinforcement learning, manipulation, and generalization. The independent Levine primer also covers his co-founder work at Physical Intelligence, a separate company.
7 lessons ยท ~37 minutes
Explore the labRecommended through the concepts linked to this topic. Each lesson belongs to an existing learning series.
How Robots Learn
How can robots learn without millions of dangerous mistakes?Explain why fixed-data learning cannot test an unfamiliar action.
Supports: Offline reinforcement learning
2020
Levine S, Kumar A, Tucker G, Fu J
Why this matters: Berkeley / RAIL research. Explains fixed datasets and the fundamental distribution-shift problem.
2020
Kumar A, Zhou A, Tucker G, Levine S
Why this matters: Berkeley / RAIL research. Introduces conservative values for poorly supported actions.