[论文] Motion-guided Contrastive Regularization For Latent World Models

论文概要

研究领域: CV 作者: Yiwen Chen, Qiqi Hou, Jisoo Jeong, Yinhao Zhu, Huaizu Jiang, Amir Said 发布时间: 2026-11-04 arXiv: 2611.06901

中文摘要

JEPA 式世界模型学习动作条件的潜在动态而不重建观测，但仅靠分布正则化不能保证其表征保留控制所需信息。我们发现静态坍缩这一失败模式：编码器保留了特定回合的静态内容，却抑制了操作相关的状态变化。静态内容在时间上可预测但跨回合多样，即使使用 SIGReg 等正则化器，这种失败仍可能持续。OGBench-Cube 上的受控研究表明，引入随机背景大幅降低规划性能，而底层操作任务未变。为解决此问题，我们提出运动引导的对比正则化：将预测的潜在转移与同一回合候选观测的转移比较，降低静态背景身份在区分动作结果方面的有用性。光流相似度提供软对比目标，使运动相似的候选获得相似监督。结合潜在预测和 SIGReg，目标鼓励表征保留回合内状态区分和相关转移间的相似性。方法在 OGBench-Cube（随机背景）上将规划成功率从 LeWM 的 52% 提升至 78%，LIBERO-Object 上从 14% 提升至 54%，同时在其他四个视觉控制基准上保持相当性能。

原文摘要

JEPA-style world models learn action-conditioned latent dynamics without reconstructing observations, but distributional regularization alone does not ensure that their representations preserve information needed for control. We identify static collapse, a failure mode in which the encoder preserves episode-specific static content while suppressing manipulation-relevant state changes. Because static content is temporally predictable yet diverse across episodes, this failure can persist despite regularizers such as SIGReg. A controlled study on OGBench-Cube shows that introducing randomized backgrounds substantially degrades planning performance while leaving the underlying manipulation task unchanged. To address this problem, we propose motion-guided contrastive regularization. Our approac...

#论文 #arXiv #CV #小凯