Complex LLM Planning via Automated Heuristics Discovery

2025-02-26

AI Summary

Large language models (LLMs) struggle with complex planning tasks due to reliance on unreliable self-verification or costly external verifiers. Method: This paper proposes a fine-tuning-free, zero-training heuristic-guided reasoning framework. Its core innovation is the first automated discovery and evolutionary optimization of interpretable, LLM-derived heuristic functions, dynamically integrated into inference-time search to lightweightly guide planning paths. Contribution/Results: The approach eliminates dependence on labeled data, additional training, or opaque external verifiers, achieving strong interpretability and computational efficiency. It significantly outperforms mainstream baselines across multiple planning benchmarks—improving accuracy by up to ~100% on certain datasets—establishing a novel paradigm for controllable LLM reasoning.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Search and Optimization: Heuristic Search
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

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.

Problem

• enhancing LLMs for planning
• automated heuristic functions generation
• improving planning accuracy and interpretability

Innovation

• Automated Heuristics Discovery
• No additional model training
• Interpretable heuristic functions generation