The Generalization Gap in Offline Reinforcement Learning

TL;DR

This paper evaluates the generalization capabilities of offline reinforcement learning methods compared to online methods, introduces a benchmark for testing generalization, and finds offline methods currently underperform in new environments, with data diversity being crucial.

Contribution

It introduces the first benchmark for offline RL generalization and provides empirical evidence that current offline algorithms struggle to generalize, emphasizing the importance of data diversity.

Findings

Offline algorithms perform worse than online RL on new environments.

Behavioral cloning outperforms other offline methods when trained on diverse data.

Increasing data diversity improves generalization across all offline algorithms.

Abstract

Despite recent progress in offline learning, these methods are still trained and tested on the same environment. In this paper, we compare the generalization abilities of widely used online and offline learning methods such as online reinforcement learning (RL), offline RL, sequence modeling, and behavioral cloning. Our experiments show that offline learning algorithms perform worse on new environments than online learning ones. We also introduce the first benchmark for evaluating generalization in offline learning, collecting datasets of varying sizes and skill-levels from Procgen (2D video games) and WebShop (e-commerce websites). The datasets contain trajectories for a limited number of game levels or natural language instructions and at test time, the agent has to generalize to new levels or instructions. Our experiments reveal that existing offline learning algorithms struggle to match the performance of online RL on both train and test environments. Behavioral cloning is a strong baseline, outperforming state-of-the-art offline RL and sequence modeling approaches when trained on data from multiple environments and tested on new ones. Finally, we find that increasing the diversity of the data, rather than its size, improves performance on new environments for all offline learning algorithms. Our study demonstrates the limited generalization of current offline learning algorithms highlighting the need for more research in this area.