LLM-Evolved Domain-Independent Heuristics for Symbolic AI Planning

2026-05-28 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

Researchers have developed LLM-Evolved Domain-Independent Heuristics for symbolic AI planning, which surpass hand-engineered state-of-the-art solutions. This approach uses evolutionary search, where a large language model mutates parent heuristics written in C++. Candidate heuristics are stored in a MAP-Elites archive, keyed on informedness and speed, with fitness scores calculated by blending coverage and solving time. On unseen testing domains, the best evolved heuristic solves more tasks than even the strongest baseline, with the full heuristic suite spanning the Pareto frontier of the informedness-speed tradeoff. The study also found that seeding evolution from the trivial blind heuristic outperforms seeding from the strong FF heuristic, and LLM reasoning effort significantly affects compilation success rates. These evolved programs are plain C++, allowing them to function as drop-in replacements in existing planners while retaining underlying search guarantees.

Key takeaway

For AI Scientists developing symbolic AI planning systems, you should investigate integrating LLM-evolved domain-independent heuristics. This research demonstrates that such heuristics can exceed hand-engineered state-of-the-art performance, solving more tasks on unseen domains. Your teams can utilize these C++-based drop-in replacements to enhance planner capabilities, inheriting existing soundness and completeness guarantees. Consider exploring evolutionary search with LLMs, potentially starting with simple heuristic seeds, to discover novel and highly effective planning solutions.

Key insights

LLMs can evolve domain-independent heuristics for symbolic AI planning that outperform human-designed ones.

Principles

• Evolutionary search enhances LLM-generated code.
• Domain-independent heuristics are LLM-evolvable.
• Trivial heuristic seeding can outperform strong baselines.

Method

LLMs mutate C++ heuristics; evolutionary search optimizes via a MAP-Elites archive, blending coverage and solving time for fitness.

In practice

• Integrate evolved C++ heuristics into existing planners.
• Explore trivial heuristics for evolutionary seeding.
• Benchmark heuristics on informedness-speed tradeoff.

Topics

• Large Language Models
• Symbolic AI Planning
• Heuristic Search
• Evolutionary Algorithms
• MAP-Elites
• C++ Heuristics

Best for: Research Scientist, AI Scientist

Related on AIssential

• Hierarchical Task Network Planning with LLM-Generated Heuristics — LLMs can generate effective search heuristics for HTN planning, improving efficiency.
• Stop Saying “It’s Just Next Token Prediction” (You Sound Like a 2023 Tutorial) — Modern AI's shift from imitation to optimization, via RLHF, DPO, and GRPO, enables reasoning beyond mere next token prediction.
• Large Language Model-Driven Full-Component Evolution of Adaptive Large Neighborhood Search — LLM-driven evolution can automate and enhance complex metaheuristic algorithm design by decoupling and optimizing core components.
• Google Goes to EXTREMES: In-Context Symbolic AI — LLMs can perform symbolic reasoning with high accuracy when rigorously constrained by in-context PDDL definitions.
• AIエージェントが最適化プログラミングコンテストで初優勝 — AI agents can now win complex, real-time optimization contests by scaling LLM inference and self-learning.
• Evolution Is Back: A New Way to Fine‑Tune LLMs — Evolution Strategies offer a black-box, gradient-free alternative for LLM fine-tuning, particularly effective with sparse rewards.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.