Robust Task Representations for Offline Meta-Reinforcement Learning via Contrastive Learning
About
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.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Meta-Reinforcement Learning | Hopper-Param (ID) | Average Return302 | 30 | |
| Meta-Reinforcement Learning | Cheetah-Vel-Sparse (OOD) | Average Return260 | 15 | |
| Meta-Reinforcement Learning | Cheetah-Vel | Average Return60 | 10 | |
| Meta-Reinforcement Learning | Walker-Param (ID) | Average Return119 | 10 | |
| Offline Meta-Reinforcement Learning | Point-Robot sampled 10 unseen (test) | Average Return-7.8 | 10 | |
| Meta-Reinforcement Learning | Walker-Param | Average Return127 | 10 | |
| Meta-Reinforcement Learning | Point-Robot Sparse | Average Return38 | 10 | |
| Meta-Reinforcement Learning | Walker Param-Sparse | Average Return78 | 10 | |
| Offline Meta-Reinforcement Learning | Walker-Rand-Params sampled 10 unseen (test) | Average Return312.5 | 10 | |
| Meta-Reinforcement Learning | Cheetah-Vel ID | Average Return214 | 10 |