First return, then explore

arxiv:2004.12919

Original source · Paper JSON

Reinforcement learning · 2020-04-27

Cites (2) · Cited by (4)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2
3. Part 3
4. Part 4
5. Part 5

Contents

• First return, then explore
• Abstract
• The Go-Explore family of algorithms
• Learning to play Atari when returning without a policy
• A hard-exploration robotics environment
• Policy-based Go-Explore
• Conclusion
• Supplementary Information
• 0.10 Detachment and derailment in reinforcement learning
• 0.11 Derailment in robotics
• 0.12 Policy-based Go-Explore and Stochastic Environments
• 0.13 Policy-based Go-Explore vs. DTSIL
• 0.14 No-ops and sticky actions
• 0.15 PPO and SIL
• 0.16 The backward algorithm
• 0.17 Go-Explore vs. Agent57
• 0.18 ALE issues
• 0.19 Infrastructure