Hierarchical Task Network Planning with LLM-Generated Heuristics

2026-05-08

AI Summary

This work addresses the long-standing limitation in Hierarchical Task Network (HTN) planning caused by the scarcity of efficient and informative heuristic functions. It introduces large language models (LLMs) into HTN planning for the first time, leveraging domain-specific prompts to guide LLMs in generating high-quality heuristic estimates that substantially enhance planning efficiency. The approach is evaluated using the Pytrich planner across six standard HTN benchmark domains, comparing nine distinct LLMs against established baselines such as TDG and LMCount. Experimental results demonstrate that LLM-derived heuristics achieve coverage nearly on par with optimal planners and reduce search overhead significantly in 83% of shared problem instances, effectively overcoming the bottleneck imposed by insufficient heuristic guidance in traditional HTN planning.

Abstract

HTN planning is a variation of classical planning where, instead of searching for a linear sequence of actions, an algorithm decomposes higher-level tasks using a method library until only executable actions remain. On one hand, this allows one to introduce domain knowledge that can speed up the search for a solution through the method library. On the other hand, it creates challenges that go beyond those of classical state-space search. While recent research produced a number of heuristics and novel algorithms that speed up HTN planning, these heuristics are not yet as informative as those available in classical planning algorithms. We investigate whether large language models (LLMs) can generate effective search heuristics for HTN planning, extending the methodology of Corrêa, Pereira, and Seipp (2025) from classical to hierarchical planning. Using the Pytrich planner on six standard total-order HTN benchmark domains, we evaluate heuristics generated by nine LLMs under domain-specific prompting and compare them against the TDG and LMCount domain-independent baselines and the PANDA planner. Our results show that LLM-generated heuristics nearly match the coverage of the best available HTN planner, while substantially reducing search effort on 83% of shared problems.

Problem

• HTN planning
• heuristics
• large language models
• search efficiency
• hierarchical planning

Innovation

• Hierarchical Task Network (HTN) Planning
• Large Language Models (LLMs)
• Heuristic Generation
• Automated Planning
• Domain-Specific Prompting