RECREG: A framework for regular safety properties in reinforcement learning
In brief
- RECREG introduces a theoretical framework for regular safety properties in RL.
- Proven safety and convergence guarantees in tabular environments.
- Extension to deep RL with promising empirical results.
RECREG is a new framework that enforces non-Markovian regular safety properties in reinforcement learning (RL) through recovery-based shielding. It provides theoretical guarantees and empirical results, including in tabular settings and an extension to deep RL.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error