📄 中文摘要
定性数值规划（QNP）作为广义规划（GP）的重要抽象模型，旨在同时计算解决多个实例的通用计划。近期研究表明，大型语言模型（LLMs）能够作为广义规划者。本研究探讨了LLMs能否作为QNP抽象生成器来解决GP问题，并提出了一种自动调试方法以修正抽象错误。研究中提出了一种提示协议：将GP领域和训练任务输入LLMs，促使其生成抽象特征，并将初始状态、动作集和目标进一步抽象为QNP问题。同时，设计了一种自动调试方法来检测抽象错误，引导LLMs修复抽象。实验结果表明该方法的有效性。
基于预训练大型语言模型的广义规划抽象生成
📄 English Summary
Abstraction Generation for Generalized Planning with Pretrained Large Language Models
Qualitative Numerical Planning (QNP) serves as a crucial abstraction model for Generalized Planning (GP), which aims to compute general plans that can solve multiple instances simultaneously. Recent studies have shown that Large Language Models (LLMs) can function as generalized planners. This research investigates the potential of LLMs to act as QNP abstraction generators for GP problems and presents an automated debugging method to correct abstraction errors. A prompt protocol is proposed, where a GP domain and training tasks are input into LLMs, prompting them to generate abstract features and further abstract the initial state, action set, and goal into QNP problems. An automated debugging method is designed to detect abstraction errors, guiding LLMs to rectify these abstractions. Experimental results demonstrate the effectiveness of this approach.