arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.00276cs.AI
收藏

基于 LLM 的规划器的定位与错误修正

Localizing and Correcting Errors for LLM-based Planners

Language Technologies Institute, Carnegie Mellon University, Pittsburgh, US(语言技术研究所,卡内基梅隆大学,匹兹堡,美国);Machine Learning Department, Carnegie Mellon University, Pittsburgh, USA(机器学习系,卡内基梅隆大学,匹兹堡,美国)

Aditya Kumar, William W. Cohen

AI总结:

本研究提出 L-ICL 方法,通过局部上下文学习修正 LLM 规划器的错误,显著提升规划任务的有效性。

摘要:

大型语言模型(LLMs)在数学和编程推理方面表现出强大的能力,但在符号经典规划任务上经常失败。我们的研究以及先前的工作表明,LLM 生成的计划经常违反其指令中给出的领域约束(例如,穿过墙壁)。为了解决这一失败,我们提出了通过局部上下文学习(L-ICL)示范迭代增强指令的方法:针对特定失败步骤的针对性修正。具体而言,L-ICL 会识别轨迹中的第一个约束违规,并注入一个最小的输入-输出示例,以提供正确行为的失败步骤。我们提出的 L-ICL 技术比显式指令或传统 ICL 更有效,后者会添加完整的解决问题轨迹。例如,在 8x8 的网格世界中,L-ICL 仅用 60 个训练示例就使有效计划的比例达到 89%,而最佳基线仅为 59%,提高了 30%。L-ICL 在其他领域(网格世界导航、迷宫、琐石板和积木世界)以及多个 LLM 架构上也表现出显著改进。

英文摘要:

Large language models (LLMs) have demonstrated strong reasoning capabilities on math and coding, but frequently fail on symbolic classical planning tasks. Our studies, as well as prior work, show that LLM-generated plans routinely violate domain constraints given in their instructions (e.g., walking through walls). To address this failure, we propose iteratively augmenting instructions with Localized In-Context Learning (L-ICL) demonstrations: targeted corrections for specific failing steps. Specifically, L-ICL identifies the first constraint violation in a trace and injects a minimal input-output example giving the correct behavior for the failing step. Our proposed technique of L-ICL is much effective than explicit instructions or traditional ICL, which adds complete problem-solving trajectories, and many other baselines. For example, on an 8x8 gridworld, L-ICL produces valid plans 89% of the time with only 60 training examples, compared to 59% for the best baseline, an increase of 30%. L-ICL also shows dramatic improvements in other domains (gridworld navigation, mazes, Sokoban, and BlocksWorld), and on several LLM architectures.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑