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

Cross-View World Models

arXiv 26.2 2026 69.7 method

TLDR

Introduces Cross-View World Models that predict future states from different viewpoints using multi-view consistency, enabling planning from any perspective.

Reasoning

The paper presents a novel cross-view prediction objective that enforces geometric regularization, leading to view-invariant 3D representations. Its strength lies in the innovative use of multi-view consistency for spatial grounding, but it is limited to a simulated game environment (Aimlabs) without real-world validation.

Read-first score

Read-first score 69.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 57.

Recency 8%
100

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

Topical relevance 42%
81.4

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

Methodology quality 25%
80

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

Reproducibility 25%
30

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 135.

Keyword Scores

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

Deep Analysis

Innovations

  • Cross-view prediction objective: given a sequence of frames from one viewpoint, predict the future state from the same or a different viewpoint after an action is taken.
  • Cross-view consistency as geometric regularization to learn view-invariant representations of the environment's 3D structure.
  • Parallel imagination streams across viewpoints, enabling planning in whichever frame of reference best suits the task while executing from the egocentric view.

Methodology

The model is trained with a cross-view prediction objective using synchronized multi-view gameplay data from Aimlabs, an aim-training platform that provides precisely aligned multi-camera recordings with high-frequency action labels. The model learns to predict future states from the same or different viewpoints, enforcing cross-view consistency to regularize the learning of spatially grounded representations.

Key Results

Multi-view consistency provides a strong learning signal for spatially grounded representations, and the resulting model enables agents to plan from different viewpoints while executing from the egocentric view.

Limitations

  • Requires synchronized multi-view data, which may not be available in many real-world or single-view settings.
  • Demonstrated only on the Aimlabs platform; generalization to other environments or tasks is not evaluated.
  • The potential for perspective-taking in multi-agent settings is stated with uncertainty ('may offer'), indicating limited validation.

Resources

Paper

Tags