arXiv cs.AI· Jintao Li, Zhiyong Cao, Baihua Xiao, Yu Bai, B\"orje F. Karlsson·· 11 小时前AI 评分34
EviSeek：以证据为中心的具身主动信息搜寻智能体框架
EviSeek: An Evidence-Centric Agentic Framework for Embodied Active Information Seeking
AI 导读
EviSeek 是一个免训练框架，将问题推导出的证据需求转化为显式获取目标，把观察条件下的动作与选择性保留的证据关联起来。在 3,436 道 ESI-Bench 问题上，EviSeek 使 Qwen3-VL-32B 的准确率较单模型智能体提升 7.1 个百分点，Gemini 3.1 Pro 提升 3.7 个百分点，同时提升视觉支撑加权准确率并降低平均 token 消耗。
正文 · 原文
Abstract:Active embodied question answering requires agents to acquire evidence through viewpoint changes and physical interaction. The central challenge is to avoid undirected interaction: taking plausible actions without specifying what evidence they should reveal for answering the question. Action plans and accumulated histories alone can leave the informational purpose of an interaction implicit. We introduce EviSeek, a training-free framework that makes question-derived evidence demands explicit acquisition objectives, linking observation-conditioned actions and selectively retained evidence to the information the agent seeks. On 3,436 ESI-Bench questions, EviSeek improves accuracy over the Single-model agent by 7.1 percentage points with Qwen3-VL-32B and 3.7 points with Gemini 3.1 Pro, while improving visual-support-weighted accuracy and reducing mean token usage. These findings support making the informational purpose of interaction explicit, with action selection and evidence retention organized around what the agent needs to learn to answer the question.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.06434 [cs.AI] |
| (or arXiv:2611.06434v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.06434 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jintao Li [view email]
[v1]
Wed, 4 Nov 2026 12:35:04 UTC (5,704 KB)
来源：arXiv cs.AI · arxiv.org