arXiv cs.AI· Jingxuan Wei, Jiahao Fu, Fuqiang Wang, Xiaoming Wu, Yuwen Liu, Kaiwen Wei, Ming Yang, Junnan Zhu·· 3 小时前AI 评分37
MAGAR 与 MAGAR-Bench：面向 GUI 3D 机械装配的多模态智能体基准与最小返工纠错框架
MAGAR: Multimodal Agents for GUI-Based 3D Mechanical Assembly and Minimal Rework
AI 导读
研究者提出 MAGAR-Bench 基准，含来自 26 个装配族的 4,000 个多模态样本，区分装配决策、原子动作与最小返工三类任务，并给出基于冻结策略微调的选择性纠错框架 MAGAR。在留出的装配族记录状态上，MAGAR 将任务均衡总分 Overall 从 37.50 提升到 41.39，三项任务均有提升；209 次干预中 70 次纠正错误、22 次引入新错误。
正文 · 原文
Abstract:A GUI command can be correct at the interface level yet still produce an incorrect assembly. An agent must identify the relevant components, respect their structural relationships, and determine which parts should move during recovery. We introduce MAGAR-Bench, a benchmark of 4,000 multimodal samples from 26 assembly families that separates Assembly Decision, Atomic Action, and Minimal Rework. For rework, the objective counts distinct moved parts within the validated recovery candidate set for each state, rather than the number of actions. We also introduce MAGAR, a selective correction framework built on a frozen policy obtained by supervised fine-tuning. Its key design matches correction granularity to output dependencies. Structured candidates revise assembly fields and rework scope, while complete action candidates keep native GUI primitives coupled with their parameters. A learned selector retains the original prediction unless a candidate passes a task-specific intervention threshold. On recorded states from held-out assembly families, MAGAR improves Overall, the task-balanced aggregate, from 37.50 to 41.39, with gains on all three tasks. Across 209 interventions, 70 correct an error and 22 introduce one. These results support task-specific correction while highlighting the remaining tradeoff between fixing errors and preserving reliable predictions.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.02472 [cs.AI] |
| (or arXiv:2611.02472v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.02472 arXiv-issued DOI via DataCite |
Submission history
From: Jiahao Fu [view email]
[v1]
Mon, 2 Nov 2026 03:40:49 UTC (5,051 KB)
[v2]
Tue, 10 Nov 2026 09:01:56 UTC (5,058 KB)
来源：arXiv cs.AI · arxiv.org