The Embodied World Model Based on LLM with Visual Information and Prediction-Oriented Prompts

TL;DR

This paper explores enhancing embodied AI by leveraging visual data and prompt engineering with large language models, demonstrating improved world modeling and autonomous exploration capabilities.

Contribution

It introduces methods to utilize visual data and prompts to improve LLM-based embodied AI's performance as a world model.

Findings

LLMs can extract essential information from visual data

Prompt design enhances LLM's world modeling ability

Utilizing visual data improves embodied AI performance

Abstract

In recent years, as machine learning, particularly for vision and language understanding, has been improved, research in embedded AI has also evolved. VOYAGER is a well-known LLM-based embodied AI that enables autonomous exploration in the Minecraft world, but it has issues such as underutilization of visual data and insufficient functionality as a world model. In this research, the possibility of utilizing visual data and the function of LLM as a world model were investigated with the aim of improving the performance of embodied AI. The experimental results revealed that LLM can extract necessary information from visual data, and the utilization of the information improves its performance as a world model. It was also suggested that devised prompts could bring out the LLM's function as a world model.