LLM-Based Offline Learning for Embodied Agents via Consistency-Guided Reward Ensemble

TL;DR

This paper introduces CoREN, a framework that uses an ensemble of consistent, domain-grounded rewards derived from LLMs to improve offline reinforcement learning for embodied agents, outperforming other methods.

Contribution

The paper presents a novel consistency-guided reward ensemble method (CoREN) that enhances offline RL for embodied agents by grounding LLM-generated rewards to specific environments.

Findings

CoREN significantly outperforms other offline RL agents on VirtualHome.

CoREN achieves comparable performance to large LLM-based agents with fewer parameters.

The framework effectively grounds LLM estimates to target environments.

Abstract

Employing large language models (LLMs) to enable embodied agents has become popular, yet it presents several limitations in practice. In this work, rather than using LLMs directly as agents, we explore their use as tools for embodied agent learning. Specifically, to train separate agents via offline reinforcement learning (RL), an LLM is used to provide dense reward feedback on individual actions in training datasets. In doing so, we present a consistency-guided reward ensemble framework (CoREN), designed for tackling difficulties in grounding LLM-generated estimates to the target environment domain. The framework employs an adaptive ensemble of spatio-temporally consistent rewards to derive domain-grounded rewards in the training datasets, thus enabling effective offline learning of embodied agents in different environment domains. Experiments with the VirtualHome benchmark demonstrate that CoREN significantly outperforms other offline RL agents, and it also achieves comparable performance to state-of-the-art LLM-based agents with 8B parameters, despite CoREN having only 117M parameters for the agent policy network and using LLMs only for training.