Enhancing Offline Model-Based RL via Active Model Selection: A Bayesian Optimization Perspective

TL;DR

This paper introduces BOMS, a Bayesian optimization-based active model selection framework that significantly improves offline model-based reinforcement learning by efficiently choosing the best dynamics model with minimal online interaction.

Contribution

The paper proposes a novel Bayesian optimization approach for model selection in offline MBRL, utilizing a new model-induced kernel for probabilistic inference and demonstrating improved performance with limited online interaction.

Findings

BOMS outperforms baseline methods in various RL tasks.

Achieves significant improvements with only 1-2.5% online interaction.

The model-induced kernel enables efficient and accurate model selection.

Abstract

Offline model-based reinforcement learning (MBRL) serves as a competitive framework that can learn well-performing policies solely from pre-collected data with the help of learned dynamics models. To fully unleash the power of offline MBRL, model selection plays a pivotal role in determining the dynamics model utilized for downstream policy learning. However, offline MBRL conventionally relies on validation or off-policy evaluation, which are rather inaccurate due to the inherent distribution shift in offline RL. To tackle this, we propose BOMS, an active model selection framework that enhances model selection in offline MBRL with only a small online interaction budget, through the lens of Bayesian optimization (BO). Specifically, we recast model selection as BO and enable probabilistic inference in BOMS by proposing a novel model-induced kernel, which is theoretically grounded and computationally efficient. Through extensive experiments, we show that BOMS improves over the baseline methods with a small amount of online interaction comparable to only $1\%$-$2.5\%$ of offline training data on various RL tasks.