Goal Force：教视频模型完成物理条件化的目标
Goal Force: Teaching Video Models To Accomplish Physics-Conditioned Goals
把力向量与中间动力学作为条件信号注入视频生成模型，使其充当隐式神经物理模拟器；这一条件化范式可迁移到需要物理一致性的视频预测与机器人规划任务中。
原文依据 allows users to define goals via explicit force vectors and intermediate dynamics
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频生成领域的最新进展使得构建能够为机器人学与规划模拟潜在未来的「世界模型」成为可能。然而，为这类模型精确指定目标仍是一项挑战：文本指令往往过于抽象，难以刻画物理细节；而目标图像在动态任务中又常常无法指定。为解决这一问题，我们提出 Goal Force，一种新颖的框架，允许用户通过显式的力向量与中间动力学来定义目标，这与人类构想物理任务的方式相一致。我们在一个精心构建的合成因果基元数据集（如弹性碰撞与多米诺骨牌倒下）上训练视频生成模型，教会其随时间与空间传播力。尽管仅在简单物理数据上训练，我们的模型对复杂的真实世界场景展现出显著的零样本泛化能力，包括工具操作与多物体因果链。我们的结果表明，通过将视频生成建立在基本物理交互之上，模型能够涌现为隐式神经物理模拟器，从而在无需依赖外部引擎的情况下实现精确的、物理感知的规划。
英文原文摘要 1185 字符
Recent advancements in video generation have enabled the development of "world models" capable of simulating potential futures for robotics and planning. However, specifying precise goals for these models remains a challenge; text instructions are often too abstract to capture physical nuances, while target images are frequently infeasible to specify for dynamic tasks. To address this, we introduce Goal Force, a novel framework that allows users to define goals via explicit force vectors and intermediate dynamics, mirroring how humans conceptualize physical tasks. We train a video generation model on a curated dataset of synthetic causal primitives--such as elastic collisions and falling dominos--teaching it to propagate forces through time and space. Despite being trained on simple physics data, our model exhibits remarkable zero-shot generalization to complex, real-world scenarios, including tool manipulation and multi-object causal chains. Our results suggest that by grounding video generation in fundamental physical interactions, models can emerge as implicit neural physics simulators, enabling precise, physics-aware planning without reliance on external engines.