A Recipe for Unbounded Data Augmentation in Visual Reinforcement Learning

TL;DR

This paper introduces SADA, a generalized data augmentation recipe for visual reinforcement learning that enhances training stability and generalization across diverse augmentation types, extending beyond photometric methods.

Contribution

Proposes SADA, a new data augmentation approach that broadens the types of augmentations effective for visual RL, improving stability and generalization.

Findings

SADA significantly improves training stability in visual RL.

SADA enhances generalization across various tasks and augmentations.

The method outperforms prior approaches like SVEA on benchmark tasks.

Abstract

Q-learning algorithms are appealing for real-world applications due to their data-efficiency, but they are very prone to overfitting and training instabilities when trained from visual observations. Prior work, namely SVEA, finds that selective application of data augmentation can improve the visual generalization of RL agents without destabilizing training. We revisit its recipe for data augmentation, and find an assumption that limits its effectiveness to augmentations of a photometric nature. Addressing these limitations, we propose a generalized recipe, SADA, that works with wider varieties of augmentations. We benchmark its effectiveness on DMC-GB2 - our proposed extension of the popular DMControl Generalization Benchmark - as well as tasks from Meta-World and the Distracting Control Suite, and find that our method, SADA, greatly improves training stability and generalization of RL agents across a diverse set of augmentations. For visualizations, code and benchmark: see https://aalmuzairee.github.io/SADA/