arXiv cs.AI· Junha Chun, Youngjoon Jeong, Taesup Kim·· 13 小时前AI 评分31
表征先验如何塑造世界动作模型的泛化能力
How Representational Priors Shape Generalization in World Action Models
AI 导读
研究揭示世界动作模型（WAMs）的泛化差异源于其生成骨干网络继承的表征几何。作者提出“干扰正交性”指标，衡量特征对分布外视觉变化的正交程度，并在多种图像与视频骨干上验证：训练前正交性越高，WAM 训练后对视觉变化的鲁棒性越强，可前瞻性预测策略泛化表现。鲁棒 WAM 还表现出干扰抑制，即相对压低视觉变化引起的特征位移并生成稳健动作。
正文 · 原文
Abstract:World action models (WAMs) are shown to substantially diverge under generalization performance based on their generative backbones. We investigate this gap through the representational geometry that WAMs inherit from their backbones and how policy training reorganizes it. Our central hypothesis is that a useful representational prior would structure geometry that separates nuisance information from the in-distribution manifold in advance of training as a policy. We quantify this with nuisance orthogonality; orthogonality of the feature under visual variation outside the empirical in-distribution manifold. Across various image and video backbones, we show that a higher orthogonality from the representational prior is associated with stronger robustness to visual variations after WAM training, enabling policy generalization measurable prospectively for world models. In addition, analyzing the resulting robust WAMs reveals nuisance suppression; relative suppression of feature displacement caused by such variations and robust action generation of the policy. Together, these findings provide a geometric perspective on how the representational prior of the world model shapes the generalization performance of WAMs.
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2610.29168 [cs.RO] |
| (or arXiv:2610.29168v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.29168 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Junha Chun [view email]
[v1]
Wed, 21 Oct 2026 03:59:35 UTC (1,812 KB)
来源：arXiv cs.AI · arxiv.org