arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新，休息一下吧，周末愉快~~
arXiv 2602.22010cs.ROcs.CV

世界引导：在条件空间中进行世界建模以生成动作

World Guidance: World Modeling in Condition Space for Action Generation

The University of Hong Kong（香港大学）

Yue Su, Sijin Chen, Haixin Shi, Mingyu Liu, Zhengshen Zhang, Ningyuan Huang, Weiheng Zhong, Zhengbang Zhu, Yuxiao Liu, Xihui Liu

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

WoG通过在条件空间中建模未来观测，提升视觉-语言-动作模型的动作生成能力与泛化性能。

摘要：

利用未来观测建模来促进动作生成为增强视觉-语言-动作（VLA）模型的能力提供了一个有前途的途径。然而，现有方法难以在保持高效、可预测的未来表示和保留足够的细粒度信息以指导精确动作生成之间取得平衡。为了解决这一限制，我们提出了WoG（World Guidance），一个框架，通过将未来观测注入动作推理管道中，将其映射为紧凑的条件。然后VLA被训练同时预测这些压缩的条件和未来动作，从而在条件空间中实现有效的世界建模以进行动作推理。我们证明，建模和预测此条件空间不仅有助于细粒度的动作生成，还表现出优越的泛化能力。此外，它能够有效学习大量的人类操作视频。在模拟和现实环境中的广泛实验验证了我们的方法在基于未来预测的方法上显著优于现有方法。项目页面可在：https://selen-suyue.github.io/WoGNet/上找到。

英文摘要：

Leveraging future observation modeling to facilitate action generation presents a promising avenue for enhancing the capabilities of Vision-Language-Action (VLA) models. However, existing approaches struggle to strike a balance between maintaining efficient, predictable future representations and preserving sufficient fine-grained information to guide precise action generation. To address this limitation, we propose WoG (World Guidance), a framework that maps future observations into compact conditions by injecting them into the action inference pipeline. The VLA is then trained to simultaneously predict these compressed conditions alongside future actions, thereby achieving effective world modeling within the condition space for action inference. We demonstrate that modeling and predicting this condition space not only facilitates fine-grained action generation but also exhibits superior generalization capabilities. Moreover, it learns effectively from substantial human manipulation videos. Extensive experiments across both simulation and real-world environments validate that our method significantly outperforms existing methods based on future prediction. Project page is available at: https://selen-suyue.github.io/WoGNet/

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

↑