RobustRLlib: A Unified Library and Benchmark for Robust Reinforcement Learning Algorithms
Robust reinforcement learning (RL) is critical for reliable real-world deployment, where policies may fail under diverse shifts and disruptions. Existing robust RL algorithms offer a rich set of mechanisms and principles, yet two major barriers limit their broader potential: limited algorithm reusability and an unclear empirical capability boundary. We introduce RobustRLlib, an algorithm-centric…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
