Hybrid Transfer Reinforcement Learning: Provable Sample Efficiency from Shifted-Dynamics Data

TL;DR

This paper introduces a hybrid transfer reinforcement learning framework that leverages shifted-dynamics offline data to improve sample efficiency, providing theoretical guarantees and demonstrating superior performance over standard online RL.

Contribution

The paper proposes HySRL, a novel transfer algorithm that uses prior knowledge of dynamics shift to achieve provably better sample complexity in target RL tasks.

Findings

HySRL outperforms baseline online RL in experiments.

Shifted-dynamics data alone does not reduce sample complexity without prior shift information.

Prior knowledge of dynamics shift enables problem-dependent sample efficiency.

Abstract

Online Reinforcement learning (RL) typically requires high-stakes online interaction data to learn a policy for a target task. This prompts interest in leveraging historical data to improve sample efficiency. The historical data may come from outdated or related source environments with different dynamics. It remains unclear how to effectively use such data in the target task to provably enhance learning and sample efficiency. To address this, we propose a hybrid transfer RL (HTRL) setting, where an agent learns in a target environment while accessing offline data from a source environment with shifted dynamics. We show that -- without information on the dynamics shift -- general shifted-dynamics data, even with subtle shifts, does not reduce sample complexity in the target environment. However, with prior information on the degree of the dynamics shift, we design HySRL, a transfer algorithm that achieves problem-dependent sample complexity and outperforms pure online RL. Finally, our experimental results demonstrate that HySRL surpasses state-of-the-art online RL baseline.