From Pixels to States: Rethinking Interactive World Models as Game Engines

arXiv 2026 46.8 method

TLDR

Rethinks interactive world models as game engines, analyzing four dimensions and introducing a large-scale gameplay dataset.

Reasoning

Strengths: Structured analysis using game engine loop and a large-scale dataset. Weaknesses: Lacks empirical evaluation results; more of a position paper.

Read-first score

Read-first score 46.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 52.

Recency 6%

100

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

Topical relevance 29%

74.3

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=analysis,dataset,result

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=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: 461.

Keyword Scores

interactive world model

world model

generative world model

video world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Reinterprets interactive world models through the lens of a recurrent action-state-observation loop, mirroring traditional game engines.
A taxonomy of existing approaches along four dimensions: player action control, game state dynamics, state-observation persistence, and real-time interactive generation.
A scalable data engine that collects over 90 hours of gameplay from Black Myth: Wukong with frame-aligned player actions, ground-truth game states, and visual observations, enriched with structured and semantic annotations.

Methodology

The paper proposes a conceptual framework based on the game engine's action-state-observation loop, and systematically examines existing interactive world models along four critical dimensions, grouping them into representative families and discussing their strengths and trade-offs. It also develops a data collection pipeline for the game Black Myth: Wukong to produce a rich, synchronized dataset of gameplay data.

Key Results

The analysis yields a structured taxonomy of interactive world model approaches, and the data engine provides a dataset of over 90 hours of gameplay with frame-aligned actions, ground-truth game states, and visual observations, along with structured and semantic annotations.

Resources

Paper