Zero-shot Model-based Reinforcement Learning using Large Language Models

TL;DR

This paper explores leveraging large language models for zero-shot model-based reinforcement learning in continuous environments, introducing a novel method called Disentangled In-Context Learning to improve dynamics prediction and uncertainty estimation.

Contribution

It proposes Disentangled In-Context Learning (DICL) to enable LLMs to predict dynamics in continuous MDPs, addressing multivariate data and control signal challenges, with theoretical and experimental validation.

Findings

DICL improves dynamics prediction accuracy.

The approach yields well-calibrated uncertainty estimates.

Demonstrated effectiveness in policy evaluation and data-augmented RL.

Abstract

The emerging zero-shot capabilities of Large Language Models (LLMs) have led to their applications in areas extending well beyond natural language processing tasks. In reinforcement learning, while LLMs have been extensively used in text-based environments, their integration with continuous state spaces remains understudied. In this paper, we investigate how pre-trained LLMs can be leveraged to predict in context the dynamics of continuous Markov decision processes. We identify handling multivariate data and incorporating the control signal as key challenges that limit the potential of LLMs' deployment in this setup and propose Disentangled In-Context Learning (DICL) to address them. We present proof-of-concept applications in two reinforcement learning settings: model-based policy evaluation and data-augmented off-policy reinforcement learning, supported by theoretical analysis of the proposed methods. Our experiments further demonstrate that our approach produces well-calibrated uncertainty estimates. We release the code at https://github.com/abenechehab/dicl.