Context Shift Reduction for Offline Meta-Reinforcement Learning

TL;DR

This paper introduces CSRO, a novel method for offline meta-reinforcement learning that effectively reduces context shift caused by distribution discrepancies, thereby enhancing generalization to unseen tasks using only offline datasets.

Contribution

The paper proposes a new approach called CSRO that minimizes policy influence on context representations during training and testing, addressing the context shift problem in OMRL.

Findings

CSRO significantly reduces context shift in OMRL.

CSRO improves generalization performance across various domains.

Experimental results outperform previous methods in challenging tasks.

Abstract

Offline meta-reinforcement learning (OMRL) utilizes pre-collected offline datasets to enhance the agent's generalization ability on unseen tasks. However, the context shift problem arises due to the distribution discrepancy between the contexts used for training (from the behavior policy) and testing (from the exploration policy). The context shift problem leads to incorrect task inference and further deteriorates the generalization ability of the meta-policy. Existing OMRL methods either overlook this problem or attempt to mitigate it with additional information. In this paper, we propose a novel approach called Context Shift Reduction for OMRL (CSRO) to address the context shift problem with only offline datasets. The key insight of CSRO is to minimize the influence of policy in context during both the meta-training and meta-test phases. During meta-training, we design a max-min mutual information representation learning mechanism to diminish the impact of the behavior policy on task representation. In the meta-test phase, we introduce the non-prior context collection strategy to reduce the effect of the exploration policy. Experimental results demonstrate that CSRO significantly reduces the context shift and improves the generalization ability, surpassing previous methods across various challenging domains.