arXiv cs.CL· Alejandra Zambrano, Sara Vera Marjanovic, Imene Kerboua, Xing Han L\`u, Leila Kosseim·· 13 小时前AI 评分31
规划方式真的重要吗？LLM 网页智能体规划表示实证研究
Does The Way You Plan Matter? An Empirical Study of Planning Representations for LLM Web Agents
AI 导读
研究者提出 PlanAhead 静态规划器-执行器框架，在 WebArena 困难任务上系统比较顺序子目标、叙事、伪代码、清单 4 种规划表示，覆盖 OpenAI、阿里、Google 多模态 LLM 智能体。
正文 · 原文
Abstract:Despite recent advances, LLM-based web agents still struggle with limited exploration, omission of critical steps, and sensitivity to task constraints. Prior work suggests that many of these failures stem from weaknesses in planning, yet the impact of alternative natural language plan representations remains unexplored. To address this, we introduce PlanAhead, a static planner-executor framework that evaluates the impact of plan representation on agent performance. We first automatically categorize WebArena tasks into 3 difficulty levels, enabling consistent difficulty grading without human annotation. Then we systematically evaluate 4 different plan representations on the tasks categorized as hard: sequential subgoals, narrative, pseudocode, and checklist; across different families of multimodal LLM-powered agents(OpenAI, Alibaba, and Google). To account for stochastic variability, we introduce a novel decomposition of agent Success Rate (SR) into 2 interpretable evaluation metrics: Achievement Rate (AR) and Solved-Task Consistency (STC). Our results show that both, the plan formulation and the underlying LLM generating the plan, significantly influence web-agent robustness and task success.
| Comments: | Camera Ready version for REALM at EMNLP. Extended appendix, extended discussion section and change of main figures |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.29927 [cs.CL] |
| (or arXiv:2605.29927v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.29927 arXiv-issued DOI via DataCite |
Submission history
From: Alejandra Zambrano [view email]
[v1]
Thu, 28 May 2026 13:39:49 UTC (2,466 KB)
[v2]
Tue, 27 Oct 2026 10:44:13 UTC (3,439 KB)
来源：arXiv cs.CL · arxiv.org