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

Video World Models with Long-term Spatial Memory

arXiv 25.6 2025 61.8 method

TLDR

Introduces geometry-grounded long-term spatial memory to improve consistency in video world models.

Reasoning

Strengths include a novel memory mechanism inspired by human cognition and custom datasets for evaluation. Weaknesses are the narrow focus on video world models without broader real-world validation or discussion of limitations.

Read-first score

Read-first score 61.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 43.

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%
70

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

Topical relevance 42%
61.4

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 25%
46

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 166.

Keyword Scores

video world model
10
world model
9
generative world model
8
interactive world model
6
world simulator
5
world dynamics prediction
3
model-based reinforcement learning world model
2

Deep Analysis

Innovations

  • Geometry-grounded long-term spatial memory for video world models
  • Mechanisms to store and retrieve information from long-term spatial memory
  • Custom datasets for training and evaluating world models with explicitly stored 3D memory mechanisms

Methodology

The framework introduces a geometry-grounded long-term spatial memory with mechanisms to store and retrieve information. Custom datasets are curated to train and evaluate world models that incorporate explicit 3D memory mechanisms.

Key Results

Evaluations show improved quality, consistency, and context length compared to relevant baselines.

Resources

PaperProject

Tags