Stabilizing Deep Q-Learning with ConvNets and Vision Transformers under Data Augmentation
This addresses a key problem for RL practitioners by enabling more stable and efficient use of data augmentation to improve generalization in visual RL tasks, though it is incremental as it builds on existing algorithms.
The paper tackled the instability and reduced sample efficiency caused by data augmentation in off-policy reinforcement learning (RL) with visual observations, identifying high-variance Q-targets as the root cause and proposing a stabilization technique that improves stability and achieves competitive generalization results, such as scaling to Vision Transformers and matching state-of-the-art methods in unseen visual environments.
While agents trained by Reinforcement Learning (RL) can solve increasingly challenging tasks directly from visual observations, generalizing learned skills to novel environments remains very challenging. Extensive use of data augmentation is a promising technique for improving generalization in RL, but it is often found to decrease sample efficiency and can even lead to divergence. In this paper, we investigate causes of instability when using data augmentation in common off-policy RL algorithms. We identify two problems, both rooted in high-variance Q-targets. Based on our findings, we propose a simple yet effective technique for stabilizing this class of algorithms under augmentation. We perform extensive empirical evaluation of image-based RL using both ConvNets and Vision Transformers (ViT) on a family of benchmarks based on DeepMind Control Suite, as well as in robotic manipulation tasks. Our method greatly improves stability and sample efficiency of ConvNets under augmentation, and achieves generalization results competitive with state-of-the-art methods for image-based RL in environments with unseen visuals. We further show that our method scales to RL with ViT-based architectures, and that data augmentation may be especially important in this setting.