Beckmann World Models: Direct Terminal-Map Learning for One-Step Action-Conditioned Prediction

查看原文 / View Original 编辑 / Edit

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.26903 (cs) [Submitted on 16 Oct 2026] Title: Beckmann World Models: Direct Terminal-Map Learning for One-Step Action-Conditioned Prediction Authors: Adam Lee, Shobhit Agarwal, June Pyo Suh.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.26903 (cs) [提交于 2026 年 10 月 16 日] 标题：贝克曼世界模型：用于单步动作条件预测的直接终端映射学习作者：Adam Lee, Shobhit Agarwal, June Pyo Suh。

Abstract: Action-conditioned world models support planning and policy evaluation by predicting future states. However, learning an accurate one-step predictor can require substantial training computation. We introduce Beckmann World Models (BWM), which learn a direct map from noise, visual history, and actions to future frames without a pretrained generative teacher.

摘要：动作条件世界模型通过预测未来状态来支持规划和策略评估。然而，学习一个准确的单步预测器可能需要大量的训练计算。我们引入了贝克曼世界模型（BWM），它无需预训练的生成式教师，即可学习从噪声、视觉历史和动作到未来帧的直接映射。

We adapt the terminal-map conservation law of Beckmann Transport Models to action-conditioned video prediction, training directly from noise-data pairs without conditional populations of generated samples required by DriftWorld. However, transport consistency on interpolated inputs does not ensure accurate prediction from the pure noise used at inference.

我们将贝克曼传输模型的终端映射守恒定律应用于动作条件视频预测，直接从噪声-数据对进行训练，而无需 DriftWorld 所需的生成样本条件群体。然而，插值输入上的传输一致性并不能确保推理时所用纯噪声的准确预测。

We address this gap with endpoint-aligned training that directly supervises the source prediction, emphasizes motion and spatial structure, and teaches recovery from generated histories. All components train the same map and retain one network call per future chunk.

我们通过端点对齐训练解决了这一差距，该训练直接监督源预测，强调运动和空间结构，并教授从生成的历史中进行恢复。所有组件训练同一个映射，并保持每个未来块进行一次网络调用。

On PushT and Robomimic Can, the final model reduces 64-frame prediction MSE by 68% and 57%, respectively, relative to same-backbone regression baseline. For policy evaluation, BWM achieves the lowest mean IoU error among the evaluated models, 0.0100, at 24.8 ms per four-frame PushT prediction.

在 PushT 和 Robomimic Can 上，最终模型相对于相同主干的回归基线，将 64 帧预测的均方误差（MSE）分别降低了 68% 和 57%。在策略评估方面，BWM 在评估模型中实现了最低的平均 IoU 误差（0.0100），每四帧 PushT 预测耗时 24.8 毫秒。

On Robomimic Can, the BWM stage trained with motion supervision and generated histories achieves 3.3 times higher throughput and 78% lower peak memory than local one-step DriftWorld. In a latent-space extension to RT-1 and Bridge-V2, equal-update controls show that feature supervision under generated histories improves perceptual prediction.

在 Robomimic Can 上，通过运动监督和生成历史训练的 BWM 阶段，其吞吐量比本地单步 DriftWorld 高出 3.3 倍，峰值内存降低了 78%。在针对 RT-1 和 Bridge-V2 的潜在空间扩展中，等量更新对照显示，生成历史下的特征监督改善了感知预测。

These findings show how terminal-map learning can combine efficient training with accurate one-step visual prediction.

这些发现表明，终端映射学习如何能够将高效的训练与准确的单步视觉预测相结合。