[论文] M-Plan: Comprehensive Planning For LLM Agents with Memory Arrangement

论文概要

研究领域: ML
作者: Ziyang Liu, Jiakang Yuan, Jiayuan Fan, Tao Chen
发布时间: 2026-10-22
arXiv: 2610.27032

中文摘要

基于大语言模型（LLM）的智能体最近在单轮任务中表现出色，但在长时程任务中仍然困难，尤其是在没有先验训练的环境中。单LLM的一个关键局限是处理长上下文的能力不足，这源于持续决策和推理能力的薄弱。虽然配备长期记忆库的现有多智能体系统（MAS）可以部分弥补这一差距，但它们大多依赖知识复用或摘要，这种方法往往受限于直接检索而没有任何针对当前任务的适配，且检索到的记忆可能包含大量待处理的历史噪声。相比之下，以任务特定指导为特征的MAS往往缺乏专门为指导设计的长期记忆。为应对这些挑战，我们提出M-Plan，一个集成长期和短期记忆的闭环MAS框架。通过结合全局-局部和过去-当前指导，M-Plan在保持即插即用设计的同时增强了智能体在多轮任务上的表现。在任务执行过程中，智能体接收任务特定的里程碑和洞察，仅在出现困惑时获得局部提示，共同构成全局-局部指导。这些指导的生成和利用也构建了短期记忆。任务完成后，当前任务的相关信息在被发送到长期记忆存储以供未来检索之前会被优化或重组。在包括ALFWorld、Webshop和PDDL在内的多轮任务上的大量实验表明，M-Plan在准确性和效率上均优于多种基线方法。

原文摘要

Large language model (LLM)-based agents have recently demonstrated outstanding performance in single-turn tasks, yet they continue to struggle with long-horizon tasks, particularly in environments without prior training. A key limitation of single LLMs is their difficulty in processing long contexts, which stems from weaknesses in continuous decision-making and reasoning capabilities. Although existing multi-agent systems (MAS) equipped with long-term memory bases can partially address this gap, they mostly rely on knowledge reuse or summarization, an approach that is often constrained by directly retrieval without any adaptation for the current task, while the retrieved memory may contain much history noises waiting to be preprocessed. In contrast, MAS characterized by task-specific guida...

自动采集于 2026-10-22

#论文 #arXiv #ML #小凯