analiza aiqbench
This arXiv paper proposes Flow Equivariant World Modeling, which keeps a latent memory that shifts and transforms with self-motion and inferred object motion so that out-of-view regions stay aligned over time. The authors report gains over diffusion, memory-augmented, and recurrent world models on 2D and 3D partially observed video benchmarks, though the evidence comes from a preprint and its generality beyond those benchmarks remains to be tested. For ML researchers, the notable idea is building time-parameterized symmetries directly into the memory architecture rather than relying on learning them from data.
Flow Equivariant World Models: Structured Memory for Dynamic Environments
arXiv:2601.01075v3 Announce Type: replace Abstract: Embodied systems experience the world as 'a symphony of flows': a combination of many continuous streams of sensory input coupled to self-motion, interwoven with the dynamics of external objects. These sensory streams and the underlying dynamics of the world obey smooth, time-parameterized symmetries which existing world models ignore. Without a memory that respects this structure, partial observability presents a major obstacle to existing methods: each observation reveals only a fraction of the world, while unobserved regions continue to evolve. In this work, we introduce Flow Equivariant World Modeling, a framework that leverages time-parameterized symmetries within a latent memory for stable and accurate dynamics prediction over long horizons. The latent memory shifts and transforms equivariantly with self-motion and inferred external object motion, keeping information about out-of-view regions aligned as time progresses. We demonstrate the advantage of this framework over state-of-the-art diffusion, memory-augmented, and recurrent world model architectures on 2D and 3D partially observed video world modeling benchmarks. More broadly, our results suggest that predictive representations become more powerful when they are organized in line with the temporal and dynamical structure of the world they model. Project page: https://flowequivariantworldmodels.github.io/
Abstract:Embodied systems experience the world as 'a symphony of flows': a combination of many continuous streams of sensory input coupled to self-motion, interwoven with the dynamics of external objects. These sensory streams and the underlying dynamics of the world obey smooth, time-parameterized symmetries which existing world models ignore. Without a memory that respects this structure, partial observability presents a major obstacle to existing methods: each observation reveals only a fraction of the world, while unobserved regions continue to evolve. In this work, we introduce Flow Equivariant World Modeling, a framework that leverages time-parameterized symmetries within a latent memory for stable and accurate dynamics prediction over long horizons. The latent memory shifts and transforms equivariantly with self-motion and inferred external object motion, keeping information about out-of-view regions aligned as time progresses. We demonstrate the advantage of this framework over state-of-the-art diffusion, memory-augmented, and recurrent world model architectures on 2D and 3D partially observed video world modeling benchmarks. More broadly, our results suggest that predictive representations become more powerful when they are organized in line with the temporal and dynamical structure of the world they model. Project page: this https URL
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2601.01075 [cs.LG]
(or arXiv:2601.01075v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2601.01075 Focus to learn more arXiv-issued DOI via DataCite
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.
Text integral sub licența deschisă a publicației · 290 cuvinte
Subiecte conexe
Understanding the limits of latent world models
arXiv · 29 oct. 2026
Polarizable atomic multipoles for learning long-range electrostatics
arXiv · 2 sept. 2026
Learning Transferable Skills using Goal-Conditioned Bisimulation
arXiv · 2 oct. 2026
Limit Theorems for Disordered Quantum Trajectories
arXiv · 1 oct. 2026