Generalization in Reinforcement Learning by Soft Data Augmentation
This work tackles the problem of improving generalization and training stability for vision-based RL methods, which is crucial for deploying RL in real-world scenarios.
This paper addresses the challenge of generalization in Reinforcement Learning (RL) where extensive data augmentation can hinder optimization. They propose SODA, a method that disentangles data augmentation from policy learning by maximizing mutual information between augmented and non-augmented latent representations, leading to significant improvements in sample efficiency, generalization, and training stability.
Extensive efforts have been made to improve the generalization ability of Reinforcement Learning (RL) methods via domain randomization and data augmentation. However, as more factors of variation are introduced during training, optimization becomes increasingly challenging, and empirically may result in lower sample efficiency and unstable training. Instead of learning policies directly from augmented data, we propose SOft Data Augmentation (SODA), a method that decouples augmentation from policy learning. Specifically, SODA imposes a soft constraint on the encoder that aims to maximize the mutual information between latent representations of augmented and non-augmented data, while the RL optimization process uses strictly non-augmented data. Empirical evaluations are performed on diverse tasks from DeepMind Control suite as well as a robotic manipulation task, and we find SODA to significantly advance sample efficiency, generalization, and stability in training over state-of-the-art vision-based RL methods.