Leveraging LLMs for Generating Document-Informed Hierarchical Planning Models: A Proposal

Morgan Fine-Morris, Vincent Hsiao, Leslie N Smith, Laura M. Hiatt, Mak Roberts

Keywords: Knowledge Representation/Engineering, Hierarchical Planning Model Generation

TL;DR: A conceptual paper presenting an LLM-driven pipeline for generating hierarchical planning models and a plan for evalution.

Abstract: Hierarchical planning facilitates faster planning using domain-specific decomposition methods, which require considerable knowledge engineering; automated generation of methods is an open problem. We propose a domain-independent LLM (Large Language Model) pipeline that uses Natural Language background information to generate methods. We describe a potential evaluation scheme for testing its performance against a classical planning baseline as well as an initial case study.

Submission Number: 34