ChatHTN: Interleaving Approximate (LLM) and Symbolic HTN Planning

2025-05-17

AI Summary

Symbolic Hierarchical Task Network (HTN) planners lack sufficient solvability for open-domain complex tasks, while large language models (LLMs) offer flexibility but lack formal correctness guarantees. Method: We propose ChatHTN—a novel framework that dynamically and rigorously interleaves approximate LLM-based reasoning (using ChatGPT) with symbolic HTN planning. At critical decomposition points, ChatGPT generates candidate subtasks; each decomposition step is then formally verified against HTN semantic constraints to ensure soundness. The result is an end-to-end hierarchical plan provably correct with respect to the domain model and goal specification. Contribution/Results: Implemented in open-source Python, all generated plans are mathematically proven to achieve their goals. Experiments demonstrate that ChatHTN significantly extends the applicability of HTN planning to open-domain tasks, substantially improving solvability and practicality of complex tasks without compromising formal reliability.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Planning with Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

We introduce ChatHTN, a Hierarchical Task Network (HTN) planner that combines symbolic HTN planning techniques with queries to ChatGPT to approximate solutions in the form of task decompositions. The resulting hierarchies interleave task decompositions generated by symbolic HTN planning with those generated by ChatGPT. Despite the approximate nature of the results generates by ChatGPT, ChatHTN is provably sound; any plan it generates correctly achieves the input tasks. We demonstrate this property with an open-source implementation of our system.

Problem

• Combining symbolic HTN planning with ChatGPT queries
• Generating interleaved task decompositions for HTN planning
• Ensuring provably sound plans despite approximate solutions

Innovation

• Combines symbolic HTN with ChatGPT queries
• Interleaves symbolic and LLM-generated task decompositions
• Ensures provably sound plans despite approximations