MultiWorld: Scalable Multi-Agent Multi-View Video World Models

arXiv 2026 50.2 method

TLDR

MultiWorld scales video world models to multi-agent multi-view scenarios with accurate control and consistency.

Reasoning

The paper addresses the important gap of multi-agent modeling in video world models, introducing a scalable framework with multi-agent control and multi-view consistency. However, it lacks real-world physical validation, and the experiments appear limited to simulated environments.

Read-first score

Read-first score 50.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 61.

Recency 6%

100

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

Topical relevance 29%

87.1

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

Methodology quality 18%

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

Reproducibility 18%

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

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 640.

Keyword Scores

world model

video world model

world simulator

generative world model

interactive world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Multi-Agent Condition Module for precise multi-agent controllability
Global State Encoder to ensure coherent observations across different views
Flexible scaling of agent and view counts with parallel synthesis for efficiency

Methodology

MultiWorld is an action-conditioned video generation model that takes historical frames and current actions of multiple agents as input to predict future frames. It incorporates a Multi-Agent Condition Module to inject agent-specific actions and a Global State Encoder to enforce multi-view consistency. The model synthesizes views in parallel and is evaluated on multi-player game and multi-robot manipulation tasks.

Key Results

MultiWorld outperforms baselines in video fidelity, action-following ability, and multi-view consistency on multi-player game and multi-robot manipulation tasks.

Resources

Paper