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

Agentic Game Development as a Verifiable Trajectory Data Engine for Scaling World Models

arXiv 2026 43.9 method

TLDR

Proposes RLHEV, using game development with engine and human verification as a verifiable trajectory data engine to scale world models, avoiding fuzzy proxies.

Reasoning

The paper presents a compelling conceptual argument for using game engines as reward environments for world model training. However, it lacks experimental validation, benchmarks, or datasets, making it a position/proposal paper rather than an empirical study.

Read-first score

Read-first score 43.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 41.

Recency 6%
100

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

Methodology quality 18%
80

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

Topical relevance 29%
58.6

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

Reproducibility 18%
38

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

Citation impact 18%
0

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

Citation velocity 12%
0

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: 450.

Keyword Scores

world model
10
generative world model
7
world simulator
6
interactive world model
5
world dynamics prediction
5
video world model
4
model-based reinforcement learning world model
4

Deep Analysis

Innovations

  • Argues that scaling world models requires a recursive data engine with grounded reward signals, not just more crawled video and compute.
  • Positions game development as a missing reward environment for spatial world models, where game-engine scenes are executable world specifications enabling dense verification (collision, physics, navigability, bounded playability).
  • Proposes Reinforcement Learning with Human-Engine Verification (RLHEV), a post-training paradigm combining dense engine signals with implicit human acceptance feedback from development.

Methodology

The paper proposes RLHEV as a conceptual post-training paradigm for spatial world models. It uses agentic game development trajectories as data, with a game engine providing dense executable checks such as collision, physics, navigability, and bounded playability, while implicit human acceptance during development provides a global verification signal.

Key Results

The title and abstract do not report experimental results or quantitative metrics; they present a proposed paradigm and conceptual motivation.

Limitations

  • No explicit limitations are stated in the title or abstract.

Tags