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

LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE

arXiv 25.9 2025 74 method

TLDR

LongScape combines intra-chunk diffusion and inter-chunk autoregression with action-guided chunking and Context-aware MoE for stable long-horizon video generation in embodied world models.

Reasoning

The paper presents a novel hybrid framework that addresses temporal inconsistency and visual drift in long-horizon video generation, with a clear methodological contribution (action-guided chunking and CMoE). However, the abstract lacks explicit mention of real-world benchmarks or datasets, and the evaluation details are not provided, limiting assessment of empirical rigor.

Read-first score

Read-first score 74, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 53.

Recency 8%
86.7

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

Reproducibility 25%
81

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

Topical relevance 42%
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

Methodology quality 25%
60

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

Field roles

FrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 48.

Keyword Scores

video world model
10
world model
9
generative world model
9
world dynamics prediction
8
interactive world model
7
world simulator
6
model-based reinforcement learning world model
4

Deep Analysis

Innovations

  • Hybrid framework combining intra-chunk diffusion denoising with inter-chunk autoregressive causal generation
  • Action-guided, variable-length chunking mechanism based on semantic context of robotic actions
  • Context-aware Mixture-of-Experts (CMoE) that adaptively activates specialized experts per chunk

Methodology

LongScape is a hybrid framework that adaptively combines intra-chunk diffusion denoising with inter-chunk autoregressive causal generation. It uses an action-guided, variable-length chunking mechanism that partitions video based on semantic context of robotic actions, and a Context-aware Mixture-of-Experts (CMoE) that adaptively activates specialized experts for each chunk during generation.

Key Results

Extensive experimental results demonstrate that the method achieves stable and consistent long-horizon generation over extended rollouts.

Tags