Improving Zero-shot Generalization in Offline Reinforcement Learning using Generalized Similarity Functions

TL;DR

This paper introduces Generalized Similarity Functions (GSF), a contrastive learning framework that enhances zero-shot generalization in offline reinforcement learning by better estimating observation similarities based on expected future behaviors.

Contribution

The paper proposes GSF, a theoretically-motivated contrastive learning approach that improves offline RL generalization by aggregating observations through generalized value functions.

Findings

GSF improves zero-shot generalization on offline Procgen benchmark.

GSF can recover existing self-supervised learning objectives.

Better observation similarity estimation enhances offline RL performance.

Abstract

Reinforcement learning (RL) agents are widely used for solving complex sequential decision making tasks, but still exhibit difficulty in generalizing to scenarios not seen during training. While prior online approaches demonstrated that using additional signals beyond the reward function can lead to better generalization capabilities in RL agents, i.e. using self-supervised learning (SSL), they struggle in the offline RL setting, i.e. learning from a static dataset. We show that performance of online algorithms for generalization in RL can be hindered in the offline setting due to poor estimation of similarity between observations. We propose a new theoretically-motivated framework called Generalized Similarity Functions (GSF), which uses contrastive learning to train an offline RL agent to aggregate observations based on the similarity of their expected future behavior, where we quantify this similarity using \emph{generalized value functions}. We show that GSF is general enough to recover existing SSL objectives while also improving zero-shot generalization performance on a complex offline RL benchmark, offline Procgen.