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

StateSpaceDiffuser: Bringing Long Context to Diffusion World Models

arXiv 25.5 2025 68.8 method

TLDR

StateSpaceDiffuser integrates state-space models into diffusion world models to maintain long-term context and temporal coherence in visual prediction.

Reasoning

The paper addresses a clear limitation of diffusion world models (lack of long-term memory) with a novel integration of state-space representations. Strengths include strong empirical results showing an order-of-magnitude improvement in coherent rollout length. Weaknesses: the evaluation is limited to simulated environments (2D maze and 3D environment), and the abstract does not discuss real-world deployment or generalization.

Read-first score

Read-first score 68.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

Recency 8%
86.7

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

Methodology quality 25%
80

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

Topical relevance 42%
77.1

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%
38

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 100.

Keyword Scores

world model
10
generative world model
8
interactive world model
8
world dynamics prediction
8
world simulator
7
video world model
7
model-based reinforcement learning world model
6

Deep Analysis

Innovations

  • Integrating state-space model features into diffusion world models to restore long-term memory while preserving high-fidelity synthesis
  • Developing an evaluation protocol that probes a model's ability to reinstantiate seen content in extended rollouts to measure temporal consistency
  • Demonstrating order-of-magnitude improvement in maintaining coherent visual context over extended rollouts compared to a diffusion-only baseline

Methodology

StateSpaceDiffuser combines a diffusion model with features from a state-space model that represents the entire interaction history, enabling long-context tasks. The model is evaluated using a custom protocol for temporal consistency in extended rollouts, with experiments in a 2D maze navigation and a complex 3D environment against a strong diffusion-only baseline.

Key Results

StateSpaceDiffuser significantly outperforms the diffusion-only baseline, maintaining coherent visual context for an order of magnitude more steps and delivering consistent views in both 2D and 3D environments.

Resources

Paper

Tags