Multiplayer Interactive World Models with Representation Autoencoders

arXiv 2026 49 method

TLDR

First multiplayer world model using latent diffusion, trained on 10k hours of Rocket League, generates stable 4-player matches in real time.

Reasoning

Strengths include novel multiplayer conditioning, large-scale training, and stable long-horizon rollouts. Weaknesses are domain specificity to Rocket League and lack of explicit model-based RL integration or generalization evidence.

Read-first score

Read-first score 49, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

Recency 6%

100

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

Topical relevance 29%

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

Methodology quality 18%

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

Reproducibility 18%

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

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

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 534.

Keyword Scores

world model

generative world model

interactive world model

world simulator

video world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

First multiplayer world model for highly dynamic environments with complex physical interactions
Conditioning on action streams of multiple agents with attribution of scene changes to the correct player
Stable long-horizon rollouts from short-clip training, maintaining distributional quality for minutes to hours
Real-time generation of four-player matches using a 5B-parameter latent diffusion model

Methodology

A 5-billion-parameter latent diffusion model trained on 10,000 hours of Rocket League gameplay from publicly available bots, generating four-player matches at 20 fps on a single Nvidia B200 GPU. The study systematically investigates video codec, generative objective, multiplayer conditioning scheme, and scaling behavior.

Key Results

Rollouts stay stable far beyond the training horizon, with distributional quality holding steady for at least five minutes and observed rollouts continuing for hours without collapse. Targeted evaluations probe physical understanding beyond visual appearance.

Limitations

Persistent failure modes remain despite scaling model and data
Evaluation limited to Rocket League environment

Resources

Paper