ΔWAM: Distilling Action Tangent Fields into World Action Models

📄 arXiv: 2610.09734v1 📥 PDF

作者: Ke Wu, Hanwen Huang, Bo Gu, Kaizhao Zhang, Xiangting Meng, Yupeng Zheng, Zijun Xu, Jieru Zhao, Wenchao Ding

分类: cs.RO

发布日期: 2026-10-07

备注: 9 pages, 4 figures

💡 一句话要点

提出ΔWAM以提高机器人在复杂环境中的动作预测能力

🎯 匹配领域: 支柱二：RL算法与架构 (RL & Architecture) 支柱三：空间感知与语义 (Perception & Semantics)

关键词: 世界动作模型 动作切线场 动态预测 机器人策略 鲁棒性提升 去噪蒸馏 多模态学习

📋 核心要点

  1. 现有的世界动作模型在处理复杂环境时，往往受到外观和场景持久性的主导影响，导致动作监督效率低下。
  2. 本文提出动作切线场，通过局部泰勒展开重新定义世界监督，使其更关注与动作相关的动态变化。
  3. 在LIBERO-Plus、RoboTwin和RoboTwin2.0-Plus等数据集上，所提方法在多种环境扰动下表现出更强的鲁棒性，超越了预训练策略。

📝 摘要（中文）

世界动作模型（WAM）通过将稀疏的动作监督与密集的未来预测相结合来改善机器人策略。然而，许多可预测的未来主要受外观和场景持久性的影响，而非动作依赖的动态。基于这一观察，本文引入了动作切线场，通过局部泰勒展开重新定义世界监督，强调动作与未来动态变化之间的关系。实验表明，该方法在多个环境扰动下表现出更强的鲁棒性，并在没有大规模预训练的情况下超越了预训练策略的表现。

🔬 方法详解

问题定义：本文旨在解决现有世界动作模型在复杂环境中对动作监督的低效问题，尤其是其对外观和场景持久性的依赖。

核心思路：通过引入动作切线场，利用局部泰勒展开来重新定义世界监督，使其能够更有效地捕捉动作对未来动态的影响。

技术框架：整体架构包括动作条件的世界模型（ACWM），通过残差变分自编码器（Residual-VAE）空间表示未来动态，并探测动作变化与残差世界变化之间的局部对应关系。

关键创新：最重要的创新在于将世界监督的定义从单纯的外观预测转变为关注动作相关的动态变化，从而提高了监督的有效性。

关键设计：在模型设计中，采用了强大的动作条件世界模型，并通过多步VideoDiT去噪蒸馏为单步推理，以提高推理效率。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果表明，所提出的ΔWAM方法在LIBERO-Plus、RoboTwin和RoboTwin2.0-Plus数据集上，针对光照、背景、相机布局等多种环境扰动，表现出显著的鲁棒性，超越了多种基线策略，尤其是在没有大规模预训练的情况下，提升幅度明显。

🎯 应用场景

该研究的潜在应用领域包括机器人导航、自动驾驶、智能监控等场景，能够有效提升机器人在动态和复杂环境中的决策能力和适应性，具有重要的实际价值和未来影响。

📄 摘要（原文）

World Action Models (WAM) improve robot policies by augmenting sparse action supervision with dense future prediction. However, much of the predictable future is dominated by appearance and scene persistence rather than action-dependent dynamics. We observe that several recent WAM designs, including optical flow, motion-centric representations, and latent actions, can be understood from a common perspective in which world supervision becomes more efficient as it contains a higher proportion of action-relevant variation. Based on this insight, we introduce Action Tangent Fields, which reformulate world supervision through a local Taylor expansion of how actions induce changes in future dynamics. We represent future dynamics in Residual-VAE space, where the future latent remains recoverable from the current latent and its residual, and use a strong action-conditioned world model (ACWM) to probe the local correspondence between action variations and residual-world variations. This local first-order structure is distilled into the WAM to guide its denoising supervision toward dynamics that are more tightly coupled to action, rather than merely predictable from appearance. Across LIBERO-Plus, RoboTwin, and RoboTwin2.0-Plus, our method consistently improves robustness to lighting, background, camera, layout, and other environmental perturbations. Despite using no large-scale embodied pretraining, it achieves stronger robustness under several distribution shifts than pretrained policies. We further distill multi-step VideoDiT denoising into a single step for efficient inference. Our results suggest that effective WAM supervision should remain information-rich while concentrating its predictive capacity on the directions along which actions change the future.