世界之链：潜运动中的世界模型思维
Chain of World: World Model Thinking in Latent Motion
把预训练视频 VAE 显式分解出的结构/运动潜变量当作「潜运动链」来预测，替代逐帧重建未来画面，为视觉研究者提供了一种在潜空间做时序世界建模、并与动作序列联合自回归的紧凑方案，可直接迁移到机器人策略学习与视频预测任务。
原文依据 a pretrained video VAE serves as a latent motion extractor, explicitly factorizing video segments into structure and motion latents
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论视觉-语言-动作（VLA）模型在具身智能领域颇具前景，但它们往往忽视视觉动态背后的预测性与时序因果结构。世界模型式 VLA 通过预测未来帧来应对这一问题，却把容量浪费在重建冗余背景上。为克服这些局限，我们提出 CoWVLA，一种新的「世界之链」范式，将世界模型的时序推理与解耦的潜运动表征统一起来。首先，预训练视频变分自编码器充当潜运动提取器，将视频片段显式分解为结构潜变量与运动潜变量。随后，在预训练阶段，VLA 从一条指令和一个初始帧出发，推断出连续的潜运动链并预测该片段的终止帧。最后，在协同微调阶段，通过在统一的自回归解码器中联合建模稀疏关键帧与动作序列，使该潜动态与离散动作预测对齐。这一设计既保留了时序推理与世界知识带来的世界模型优势，又保持了潜动作的紧凑性与可解释性，从而实现高效的视觉运动学习。在机器人仿真基准上的大量实验表明，CoWVLA 优于现有的世界模型方法与潜动作方法，并达到适中的计算效率，凸显其作为更有效 VLA 预训练范式的潜力。项目网站：https://fx-hit.github.io/cowvla-io。
英文原文摘要 1473 字符
Vision-Language-Action (VLA) models are promising for embodied intelligence, yet they often overlook the predictive and temporal-causal structure underlying visual dynamics. World-model VLAs address this by predicting future frames, but waste capacity reconstructing redundant backgrounds. To overcome these limitations, we introduce CoWVLA, a new "Chain of World" paradigm that unifies world-model temporal reasoning with a disentangled latent motion representation. First, a pretrained video VAE serves as a latent motion extractor, explicitly factorizing video segments into structure and motion latents. Then, during pre-training, the VLA learns from an instruction and an initial frame to infer a continuous latent motion chain and predict the segment's terminal frame. Finally, during co-fine-tuning, this latent dynamic is aligned with discrete action prediction by jointly modeling sparse keyframes and action sequences in a unified autoregressive decoder. This design preserves the world-model benefits of temporal reasoning and world knowledge while retaining the compactness and interpretability of latent actions, enabling efficient visuomotor learning. Extensive experiments on robotic simulation benchmarks show that CoWVLA outperforms existing world-model and latent-action approaches and achieves moderate computational efficiency, highlighting its potential as a more effective VLA pretraining paradigm. Project website: https://fx-hit.github.io/cowvla-io.