文章背景与核心概要
近年来,将几何深度学习与物理先验相结合成为了人工智能和机器人学领域的前沿热点。本文介绍了一种名为“场即知晓”(The Field Knows)的全新连续度量场框架,它仅通过单一的因果对比损失,就能学习并编码跨越不同维度的复杂几何结构。
该研究的核心技术在于将场景映射到固定的对称矩阵基底系数中,通过李代数指数映射构建出黎曼或洛伦兹度量。令人瞩目的是,无论是处理平面及机械臂构型空间中的机器人避障测地线,还是模拟洛伦兹时空中黑洞的事件视界,该框架均能展现出极强的零样本泛化能力和统一的架构适应性,证明了“场懂几何,几何懂物理”的深刻统一性。
场即知晓:从导航到黑洞的跨维度几何学
arXiv: 2608.07566
作者: Chenghao Xu
提交时间: 2026年8月3日
研究领域: 人工智能 (cs.AI); 机器人学 (cs.RO)
摘要 (Summary)
《场即知晓》引入了一种新颖的连续度量场框架,该框架能够学习并编码跨越不同维度的复杂几何结构。通过利用单一的因果对比损失,该模型将场景映射到一个固定的对称矩阵基底中,随后通过指数映射将其转换为黎曼度量或洛伦兹度量。
该研究表明,该框架不仅仅是记忆构型,而是捕捉到了可迁移的几何结构。其实用多功能性通过解决各种问题得到了展示——从机器人导航中的避障,到黑洞样事件视界的自发生成——所有这些都使用相同的架构和训练协议。
The Field Knows introduces a novel continuous metric field framework capable of learning and encoding complex geometric structures across varying dimensions. By utilizing a single causal contrastive loss, the model maps scenes into a fixed symmetric matrix basis, which is then exponentiated into Riemannian or Lorentzian metrics.
The research demonstrates that this framework does not merely memorize configurations but captures transferable geometric structures. Its versatility is showcased through its ability to solve diverse problems—ranging from obstacle-avoidance in robotic navigation to the spontaneous generation of black-hole-like event horizons—using the same architecture and training protocol.
摘要详情 (Abstract)
我们引入了一种通过单一因果对比损失训练的连续度量场框架。该框架将场景编码为固定对称矩阵基底的系数,将其组合成一个李代数元素,并将结果指数映射为黎曼度量或洛伦兹度量。
在各个维度中,该场发现了全谱系的几何结构:从跨越平面及机械臂构型空间的机器人导航避障测地线,到洛伦兹时空中黑洞的事件视界。广泛的零样本泛化研究表明,该场捕捉的是可迁移的几何结构,而不是死记硬背特定的构型。在黑洞设置中,因果损失自发地演化出具有正确洛伦兹特征的真正黑洞般结构。相同的损失、相同的架构以及相同的训练协议,在各个维度上产生了全方位的几何现象。场知晓几何,而几何知晓物理。
We introduce a continuous metric field framework trained by a single causal contrastive loss. The framework encodes a scene into coefficients of a fixed symmetric matrix basis, assembles them into a Lie algebra element, and exponentiates the result to a Riemannian or Lorentzian metric.
Across dimensions, this field discovers the full spectrum of geometric structures: from obstacle-avoiding geodesics in robot navigation across planar and manipulator configuration spaces, to event horizons of black holes in Lorentzian spacetime. Extensive zero-shot generalization studies demonstrate that the field captures transferable geometric structure rather than memorizing specific configurations. In the black hole setting, the causal loss spontaneously evolves genuine black-hole-like structures with the correct Lorentzian signature. The same loss, the same architecture, and the same training protocol produce the full range of geometric phenomena across dimensions. The field knows geometry, and geometry knows physics.
访问与资源 (Access & Resources)
- 查看 PDF: arXiv:2608.07566
- HTML 版本: 实验性 HTML
- TeX 源码: 源码
- DOI: 10.48550/arXiv.2608.07566
- View PDF: arXiv:2608.07566
- HTML Version: Experimental HTML
- TeX Source: Source Code
- DOI: 10.48550/arXiv.2608.07566
引用与元数据 (Citation & Metadata)
- 引用方式: arXiv:2608.07566 [cs.AI]
- 文献计量工具:
- Cite as: arXiv:2608.07566 [cs.AI]
- Bibliographic Tools: