Factored Latent Action World Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Wang, Zizhao, Shi, Chang, Hu, Jiaheng, Rohling, Kevin, Martín-Martín, Roberto, Zhang, Amy, Stone, Peter
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866917529663832064
author Wang, Zizhao
Shi, Chang
Hu, Jiaheng
Rohling, Kevin
Martín-Martín, Roberto
Zhang, Amy
Stone, Peter
author_facet Wang, Zizhao
Shi, Chang
Hu, Jiaheng
Rohling, Kevin
Martín-Martín, Roberto
Zhang, Amy
Stone, Peter
contents Learning latent actions from action-free video has emerged as a powerful paradigm for scaling up controllable world model learning. Latent actions provide a natural interface for users to iteratively generate and manipulate videos. However, most existing approaches rely on monolithic inverse and forward dynamics models that learn a single latent action to control the entire scene, and therefore struggle in complex environments where multiple entities act simultaneously. This paper introduces Factored Latent Action Model (FLAM), a factored dynamics framework that decomposes the scene into independent factors, each inferring its own latent action and predicting its own next-step factor value. This factorized structure enables more accurate modeling of complex multi-entity dynamics and improves video generation quality in action-free video settings compared to monolithic models. Based on experiments on both simulation and real-world multi-entity datasets, we find that FLAM outperforms prior work in prediction accuracy and representation quality, and facilitates downstream policy learning, demonstrating the benefits of factorized latent action models.
format Preprint
id arxiv_https___arxiv_org_abs_2602_16229
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Factored Latent Action World Models
Wang, Zizhao
Shi, Chang
Hu, Jiaheng
Rohling, Kevin
Martín-Martín, Roberto
Zhang, Amy
Stone, Peter
Machine Learning
Learning latent actions from action-free video has emerged as a powerful paradigm for scaling up controllable world model learning. Latent actions provide a natural interface for users to iteratively generate and manipulate videos. However, most existing approaches rely on monolithic inverse and forward dynamics models that learn a single latent action to control the entire scene, and therefore struggle in complex environments where multiple entities act simultaneously. This paper introduces Factored Latent Action Model (FLAM), a factored dynamics framework that decomposes the scene into independent factors, each inferring its own latent action and predicting its own next-step factor value. This factorized structure enables more accurate modeling of complex multi-entity dynamics and improves video generation quality in action-free video settings compared to monolithic models. Based on experiments on both simulation and real-world multi-entity datasets, we find that FLAM outperforms prior work in prediction accuracy and representation quality, and facilitates downstream policy learning, demonstrating the benefits of factorized latent action models.
title Factored Latent Action World Models
topic Machine Learning
url https://arxiv.org/abs/2602.16229