GenePlan: Evolving Better Generalized PDDL Plans using Large Language Models

Researchers present GenePlan, a framework that uses large language models with evolutionary algorithms to generate domain-specific planners for classical planning tasks in PDDL. The system achieved a 0.91 SAT score across eight benchmark domains, nearly matching state-of-the-art performance while significantly outperforming other LLM-based approaches.

• →GenePlan combines LLMs with evolutionary algorithms to create interpretable Python planners for classical planning problems.
• →The framework achieved 0.91 SAT score compared to 0.93 for state-of-the-art planners and 0.64 for chain-of-thought prompting.
• →Generated planners solve new instances rapidly at 0.49 seconds per task on average.
• →The approach costs only $1.82 per domain using GPT-4o, making it cost-effective for planning applications.
• →GenePlan was tested across six existing and two new benchmark domains, demonstrating broad applicability.