An Optimisation Framework for Unsupervised Environment Design
This work addresses robustness for reinforcement learning agents in high-risk settings, offering incremental improvements with better theoretical foundations.
The paper tackles the problem of improving reinforcement learning agent robustness through unsupervised environment design (UED), providing a new optimization framework with stronger theoretical guarantees and demonstrating empirical outperformance over prior methods in various environments.
For reinforcement learning agents to be deployed in high-risk settings, they must achieve a high level of robustness to unfamiliar scenarios. One method for improving robustness is unsupervised environment design (UED), a suite of methods aiming to maximise an agent's generalisability across configurations of an environment. In this work, we study UED from an optimisation perspective, providing stronger theoretical guarantees for practical settings than prior work. Whereas previous methods relied on guarantees if they reach convergence, our framework employs a nonconvex-strongly-concave objective for which we provide a provably convergent algorithm in the zero-sum setting. We empirically verify the efficacy of our method, outperforming prior methods in a number of environments with varying difficulties.