[论文] PhysiCalWorld: In-Context Physics Calibration for World Models

论文概要

研究领域: CV 作者: Yunhao Luo, Wentao He, Yinpei Dai, Hyelin Nam, Hongze Fu, Joyce Chai, Nima Fazeli, Jeong Joon Park 发布时间: 2026-10-21 arXiv: 2610.30125

中文摘要

现有世界模型通常通过当前视觉帧或狭窄的历史窗口来感知场景。然而，这些观察可能无法揭示决定转移动力学的物理属性，例如物体的质量分布、关节阻尼或表面摩擦。如果不对世界模型校准这些物理因素，预测将趋向于回归经验先验，并在新测试场景中产生严重漂移。为此我们提出 PhysiCalWorld，通过上下文提示（in-context prompting）将基于动作条件的世界模型校准到实例特定的动力学，无需微调或解析物理建模。我们的方法将测试场景中收集的探索性交互片段编码为一个紧凑的物理描述符 token，使后续预测立足于场景的实际物理行为。PhysiCalWorld 仅使用潜在预测损失进行端到端训练，无需物理标签。推理时，描述符 token 只需提取一次，即可在多次 rollout 中重复利用，计算开销可忽略。我们在涵盖多样物理变化的四个环境上评估了 PhysiCalWorld，结果表明它显著提高了 rollout 预测准确率和视觉规划成功率。学习到的描述符还形成了具有物理结构的潜在空间，并能泛化到训练期间未见过的物理范围。

原文摘要

Existing world models typically perceive a scene through current visual frame or a narrow history window. However, these observations may not reveal the physics properties that determine transition dynamics, such as an object mass distribution, joint damping, or surface friction. Without calibrating these physics factors for world models, predictions tend to regress toward an empirical prior and induce severe drift in novel test scenes. To this end, we propose PhysiCalWorld, which calibrates action-conditioned world models to the instance-specific dynamics via in-context prompting, without fine-tuning or analytical physics modeling. Our method encodes an exploratory interaction clip collected within the test scene into a compact physics descriptor token, which grounds subsequent prediction...

#论文 #arXiv #CV #小凯