知识模型提示提升大型语言模型在规划任务中的性能

Large Language Models (LLMs) show promise in planning tasks, but their performance is often limited by their understanding of environmental dynamics and planning rules. This paper introduces Knowledge Model Prompting (KMP), a novel method that enhances LLM planning capabilities by explicitly providing task-relevant knowledge models within the prompt. KMP decomposes the knowledge model into state representation, action definitions, and transition functions, integrating them as structured text. Experimental results demonstrate that KMP significantly boosts LLM performance across various planning benchmarks, including classical, state-based, and multi-agent planning. Compared to traditional Chain-of-Thought (CoT) prompting, KMP not only improves planning success rates but also enhances interpretability and robustness. This research highlights the critical role of explicit knowledge representation in improving LLM's complex reasoning abilities, offering a new paradigm for future LLM applications in planning.