arXiv cs.AI· Yuesheng Xu, Hector Munoz-Avila·· 12 小时前AI 评分22
从 LLM 生成经验中课程引导获取分层任务知识
Curriculum-Guided Acquisition of Hierarchical Task Knowledge from LLM-Generated Experience
AI 导读
研究在 HTN 学习框架中比较三种分层任务知识获取方式：问题求解中的在线学习、人工设计课程，以及从经核验的 LLM 生成解题轨迹中自动生成课程。该方法用 LLM 求解小型训练问题，提取任务完成信息并转化为有序的状态-任务学习问题序列，用于 HTN 方法获取。在多个规划域上的结果显示，自动生成课程能恢复专家设计课程的大部分收益，同时大幅降低在线分层知识获取的 token 成本。
正文
Abstract:Hierarchical task knowledge enables efficient problem solving by allowing previously acquired skills to be reused across tasks and problem instances. In this paper, we investigate curriculum-guided acquisition of hierarchical task knowledge in an HTN learning framework. We compare three approaches: online learning during problem solving, human-designed curricula, and an automated curriculum generation procedure that derives learning experiences from verified LLM-generated solution traces. The proposed procedure solves small training problems with an LLM, extracts task accomplishment information, and transforms it into ordered sequences of state-task learning problems for HTN method acquisition. We evaluate the three approaches across multiple planning domains. The results demonstrate that automatically generated curricula recover much of the benefit of expert-designed curricula while substantially reducing the token cost of online hierarchical knowledge acquisition.
| Comments: | 15 pages, 2 figures, 1 table. Presented at the Thirteenth Annual Conference on Advances in Cognitive Systems (ACS 2026) |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.34677 [cs.AI] |
| (or arXiv:2610.34677v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.34677 arXiv-issued DOI via DataCite (pending registration) |
|
| Journal reference: | Xu, Y., and Munoz-Avila, H. (2026). Curriculum-Guided Acquisition of Hierarchical Task Knowledge from LLM-Generated Experience. Proceedings of the Thirteenth Annual Conference on Advances in Cognitive Systems (ACS 2026) |
Submission history
From: Hector Munoz-Avila [view email]
[v1]
Mon, 26 Oct 2026 01:16:29 UTC (430 KB)
来源：arXiv cs.AI · arxiv.org