WorldAuditBench：多模态智能体交互式3D世界审计基准

AI 摘要

研究者提出 WorldAuditBench，一个用于交互式 3D 世界审计的基准，包含基于 Unreal Engine 5 和 Three.js 构建的 13 个环境中的 213 个异常任务，覆盖五类异常。他们用两种审计范式评估五个前沿模型：先由 VLA 探索再由 VLM 识别异常，以及端到端 VLM 智能体直接以视觉推理指导动作选择。模型成功率在 6.6% 到 42.3% 之间，远低于人类 83.4% 的表现，表明多模态智能体在耦合动作与视觉推理方面仍有明显局限。

正文节选

WorldAuditBench: Interactive 3D World Auditing with Multimodal Agents Abstract As interactive 3D worlds are increasingly used to study intelligent behavior, it becomes important to develop efficient pipelines for identifying anomalies in these simulated environments, such as floating objects, traversable walls, or objects inconsistent with the surrounding scene. Multimodal AI systems, including vision-language models (VLMs) and vision-language-action models (VLAs), have shown potential for autom

发布时间：—

抓取时间：2026-10-01 10:53

来源机构：Hugging Face