CAER: Causal Action Effect Reweighting for World Model Training

2026-08-31

AI Summary

论文提出Causal Action Effect Reweighting方法，通过重新分配监督权重至受动作影响的token上，优化世界模型训练，提升视频生成质量。

Abstract

World models are becoming core infrastructure for embodied intelligence, with action-conditioned video generation providing controllable predictions of how scenes evolve after agent interventions. Yet existing models are commonly trained with space-time-uniform mean squared error, allowing abundant background tokens to dominate the gradient while sparse interaction dynamics remain under-optimized; such uniform fitting rewards reconstructing appearance rather than learning how actions change the world. We introduce Causal Action Effect Reweighting (CAER), a general training paradigm that redistributes supervision toward the tokens whose predicted future is causally affected by the action. CAER contrasts the model's own predictions with and without action conditioning to localize these tokens online, then normalizes the resulting effect map into a weight that preserves the total coefficient mass and changes only where it is spent. This online signal requires no external annotations or offline preprocessing, avoids additional data-processing time, and scales naturally with model and dataset size. Experiments across heterogeneous action-conditioned world-model tasks show that CAER converges to better solutions than uniform MSE training, with consistent improvements in the physical consistency, controllability, and visual quality of generated videos.

Problem

• world models
• action-conditioned video generation
• mean squared error
• interaction dynamics
• causal effect

Innovation

• Causal Action Effect Reweighting
• world model training
• action-conditioned video generation
• online localization