GenePlan: Evolving Better Generalized PDDL Plans using Large Language Models

TL;DR

GenePlan introduces a novel LLM-assisted evolutionary framework that generates interpretable, domain-generalized PDDL planners, achieving near state-of-the-art performance and rapid solution times across multiple benchmark domains.

Contribution

It presents a new approach combining large language models with evolutionary algorithms to automatically generate generalized PDDL planners.

Findings

Achieved an average SAT score of 0.91 across benchmarks.

Outperformed LLM-based baselines like chain-of-thought prompting.

Generated planners solve new instances quickly and cost-effectively.

Abstract

We present GenePlan (GENeralized Evolutionary Planner), a novel framework that leverages large language model (LLM) assisted evolutionary algorithms to generate domain-dependent generalized planners for classical planning tasks described in PDDL. By casting generalized planning as an optimization problem, GenePlan iteratively evolves interpretable Python planners that minimize plan length across diverse problem instances. In empirical evaluation across six existing benchmark domains and two new domains, GenePlan achieved an average SAT score of 0.91, closely matching the performance of the state-of-the-art planners (SAT score 0.93), and significantly outperforming other LLM-based baselines such as chain-of-thought (CoT) prompting (average SAT score 0.64). The generated planners solve new instances rapidly (average 0.49 seconds per task) and at low cost (average $1.82 per domain using GPT-4o).