Complex LLM Planning via Automated Heuristics Discovery

TL;DR

This paper introduces AutoHD, a novel method enabling LLMs to generate and refine heuristic functions for complex planning, significantly improving accuracy without additional training.

Contribution

AutoHD allows LLMs to explicitly generate and evolve heuristics for better planning, eliminating the need for external verifiers or fine-tuning.

Findings

Nearly doubled accuracy on some benchmarks

Significant improvements over multiple baselines

Heuristics provide interpretability and insights

Abstract

We consider enhancing large language models (LLMs) for complex planning tasks. While existing methods allow LLMs to explore intermediate steps to make plans, they either depend on unreliable self-verification or external verifiers to evaluate these steps, which demand significant data and computations. Here, we propose automated heuristics discovery (AutoHD), a novel approach that enables LLMs to explicitly generate heuristic functions to guide inference-time search, allowing accurate evaluation of intermediate states. These heuristic functions are further refined through a heuristic evolution process, improving their robustness and effectiveness. Our proposed method requires no additional model training or fine-tuning, and the explicit definition of heuristic functions generated by the LLMs provides interpretability and insights into the reasoning process. Extensive experiments across diverse benchmarks demonstrate significant gains over multiple baselines, including nearly twice the accuracy on some datasets, establishing our approach as a reliable and interpretable solution for complex planning tasks.