⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

Factored Latent Action World Models

arXiv 26.2 2026 54.5 method

TLDR

A factored latent action model that decomposes scenes into independent factors for better multi-entity dynamics and video generation.

Reasoning

The paper introduces a novel factorization approach for latent action world models, showing clear improvements over monolithic models on both simulation and real-world datasets. Strengths include a principled decomposition and empirical validation; weaknesses are limited discussion of scalability to very large scenes and potential inference complexity.

Read-first score

Read-first score 54.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 53.

Recency 6%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 18%
90

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,dataset,experiment,metric

Topical relevance 29%
75.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%
38

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset

Citation impact 18%
21.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.21646031

Citation velocity 12%
0

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 599.

Keyword Scores

world model
9
video world model
9
generative world model
8
world dynamics prediction
8
interactive world model
7
model-based reinforcement learning world model
7
world simulator
5

Deep Analysis

Innovations

  • Factored Latent Action Model (FLAM) that decomposes the scene into independent factors, each with its own latent action and next-step prediction
  • Factorized inverse and forward dynamics models for multi-entity action-free video
  • Improved modeling of complex multi-entity dynamics compared to monolithic latent action models

Methodology

FLAM is a factored dynamics framework that decomposes the scene into independent factors, each inferring its own latent action via an inverse model and predicting its next-step factor value via a forward model. It is trained on action-free video data and evaluated on simulation and real-world multi-entity datasets, with comparisons to monolithic baselines using prediction accuracy, representation quality, and downstream policy learning metrics.

Key Results

FLAM outperforms prior monolithic models in prediction accuracy and representation quality on multi-entity datasets, and facilitates improved downstream policy learning.

Limitations

  • The independence assumption between factors may not capture complex inter-entity interactions in highly coupled scenes
  • No explicit limitations are stated in the abstract; the above is inferred from the factorized structure

Resources

Paper

Tags