Robust Task Representations for Offline Meta-Reinforcement Learning via Contrastive Learning
This addresses a key bottleneck in offline meta-reinforcement learning for improving robustness to distribution mismatches, representing an incremental advance with specific gains.
The paper tackles the problem of offline meta-reinforcement learning, where existing methods fail to distinguish between task and behavior policy factors, leading to unstable task representations; they propose a contrastive learning framework that achieves robust task representations, showing advantages in generalization to out-of-distribution behavior policies on various benchmarks.
We study offline meta-reinforcement learning, a practical reinforcement learning paradigm that learns from offline data to adapt to new tasks. The distribution of offline data is determined jointly by the behavior policy and the task. Existing offline meta-reinforcement learning algorithms cannot distinguish these factors, making task representations unstable to the change of behavior policies. To address this problem, we propose a contrastive learning framework for task representations that are robust to the distribution mismatch of behavior policies in training and test. We design a bi-level encoder structure, use mutual information maximization to formalize task representation learning, derive a contrastive learning objective, and introduce several approaches to approximate the true distribution of negative pairs. Experiments on a variety of offline meta-reinforcement learning benchmarks demonstrate the advantages of our method over prior methods, especially on the generalization to out-of-distribution behavior policies. The code is available at https://github.com/PKU-AI-Edge/CORRO.