CAER: Causal Action Effect Reweighting for World Model Training
cs.AI
Submitted: 2026-08-31
Updated: 2026-08-31
Comments: 14 pages, 8 figures. Project page: https://manifoldai-research.github.io/CAER/
Project page: https://manifoldai-research.github.io/CAER
License: http://creativecommons.org/licenses/by/4.0/
The gist: World models are becoming core infrastructure for embodied intelligence, with action-conditioned video generation providing controllable predictions of how scenes evolve after agent interventions.
Terminology
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.
Sources
- Wan: Open and Advanced Large-Scale Video Generative Models
- HunyuanVideo: A Systematic Framework For Large Video Generative Models
- Cosmos-Transfer1: Conditional World Generation with Adaptive Multimodal Control
- Cosmos World Foundation Model Platform for Physical AI
- Worldscape-MoE: A Unified Mixture-of-Experts World Model for Scalable Heterogeneous Action Control
- WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation
- WorldScape Policy 2.0: Empowering Steerable World Action Modeling with Reasoning-Augmented Memory and In-Context Learning
- iWorld-Bench: A Benchmark for Interactive World Models with a Unified Action Generation Framework
- Classifier-Free Diffusion Guidance
- LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning
- RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation
- Stereo Magnification: Learning View Synthesis using Multiplane Images
- PoseAnything: Universal Pose-guided Video Generation with Part-aware Temporal Coherence
- CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer
- Matrix-game 2.0: An open-source, real-time, and streaming interactive world model
- Matrix-Game: Interactive World Foundation Model
- Matrix-Game 3.0: Real-Time and Streaming Interactive World Model with Long-Horizon Memory
- WorldPlay: Towards Long-Term Geometric Consistency for Real-Time Interactive World Modeling
- HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds
- World Simulation with Video Foundation Models for Physical AI
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems
- VSAL: A Vision Solver with Adaptive Layouts for Graph Property Detection