[论文] World Models' Last Exam in Physics
研究领域: CV 作者: Mingju Gao, Qingle Liu, Yuzhao Peng, Xinjie Lin, Ziming Qin, Zheng Jiang, Wenyi Li, Calvin Xiao, Youjie Zheng, Kaisen Yang, Qinhuai Na 发布时间: 2026-…
论文概要
研究领域: CV 作者: Mingju Gao, Qingle Liu, Yuzhao Peng, Xinjie Lin, Ziming Qin, Zheng Jiang, Wenyi Li, Calvin Xiao, Youjie Zheng, Kaisen Yang, Qinhuai Na 发布时间: 2026-10-06 arXiv: 2610.08791
中文摘要
视频世界模型可以生成视觉上令人信服但物理上不一致的序列,这引发了人们对其在具身AI系统中预测和规划可靠性的担忧。现有的评估通常依赖于基于模型的判断或参考视频,而直接的物理测试主要聚焦于力学。我们提出了「世界模型物理期末考」(World Models' Last Exam in Physics),这是一个基于测量的基准,用于评估视频世界模型中的物理一致性。该基准包含40个受控任务,涵盖力学、光学、流体、热学与相变、电磁学和表面张力。每个任务将初始图像和生成提示与预定义的物理标准配对,实现无需参考视频的可解释物理关系测试。其评估器结合了任务可观测性筛选和特定任务的定量物理测量。在8个视频生成模型的1280个视频上进行的实验揭示了持续的物理不一致性和跨任务的显著差异,最佳模型的总分为57.76/100。在具有已知物理关系的合成视频上的评估为测量模块在受控条件下的有效性提供了证据。评估器与人工判断的一致性也优于直接的视觉语言模型基线。
原文摘要
Video world models can produce visually convincing yet physically inconsistent sequences, raising concerns about their reliability for prediction and planning in embodied AI systems. Existing evaluations often rely on model-based judgments or reference videos, while direct physical tests largely focus on mechanics. We introduce World Models' Last Exam in Physics, a measurement-based benchmark for evaluating physical consistency in video world models. The benchmark comprises 40 controlled tasks spanning mechanics, optics, fluids, thermal and phase-change phenomena, electromagnetism, and surface tension. Each task pairs an initial image and a generation prompt with predefined physical criteria, enabling interpretable tests of observable physical relationships without requiring reference vide...
*自动采集于 2026-10-08*
#论文 #arXiv #CV #小凯