aiqbench analysis
This research addresses a critical failure mode in latent world models where frozen representations cause manipulated objects to appear static or teleport rather than move naturally. The authors propose Decode-augmented rollout training (DART), which restores temporal motion structure by refining the flow with decode-path supervision while keeping the core representation frozen. This method significantly improves prediction quality and highlights that standard pixel-level evaluation metrics can inadvertently reward these flawed, static predictions.
Frozen Flows Forget: Diagnosing and Restoring Lost Motion in a Latent-flow World Model
arXiv:2609.28414v1 Announce Type: cross Abstract: Latent world models that integrate a flow in a frozen self supervised latent space train stably and cheaply, yet silently lose the property manipulation depends on most: motion. The pretrained flow never moves the manipulated object; retraining it with latent-only losses only trades stillness for teleport-like motion. We trace the failure to the training signal, not the representation: anchor-sparse, latent-only supervision never says where along the horizon change belongs. Decode-augmented rollout training (DART) repairs this while keeping the representation frozen, retraining only the flow with decode-path supervision. DART outperforms its latent only parent on the full protocol, restores the temporal structure of motion, and re-couples predicted motion to the scene; at larger scale it further improves prediction quality, closing nearly half the remaining gap to an oracle-informed interpolation reference. Finally, we report an unexpected finding about evaluation: pixel error alone rewards frozen predictions.
Abstract:Latent world models that integrate a flow in a frozen self supervised latent space train stably and cheaply, yet silently lose the property manipulation depends on most: motion. The pretrained flow never moves the manipulated object; retraining it with latent-only losses only trades stillness for teleport-like motion. We trace the failure to the training signal, not the representation: anchor-sparse, latent-only supervision never says where along the horizon change belongs. Decode-augmented rollout training (DART) repairs this while keeping the representation frozen, retraining only the flow with decode-path supervision. DART outperforms its latent only parent on the full protocol, restores the temporal structure of motion, and re-couples predicted motion to the scene; at larger scale it further improves prediction quality, closing nearly half the remaining gap to an oracle-informed interpolation reference. Finally, we report an unexpected finding about evaluation: pixel error alone rewards frozen predictions.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Robotics (cs.RO)
arXiv:2609.28414 [cs.CV]
(or arXiv:2609.28414v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.28414 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 245 words
Related coverage
MoRAE: Flow-Friendly Self-Supervised Latents for Text-to-Motion Generation
arXiv · Aug 3, 2026
EverAnimate: Minute-Scale Human Animation via Latent Flow Restoration
arXiv · Sep 29, 2026
2026.9: There's room on this bus
Home Assistant · Sep 2, 2026
PEARS: Physical-Prior-Guided Efficient Adaptation via Failure Reasoning and Diffusion Steering for Tactile Manipulation
arXiv · Oct 7, 2026