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

SimWorld: A Unified Benchmark for Simulator-Conditioned Scene Generation via World Model

arXiv 25.3 2025 67.1 benchmark, system, application

TLDR

Proposes SimWorld, a benchmark combining simulation engine and world model for controllable scene generation to improve autonomous driving perception models.

Reasoning

The paper presents a novel data generation pipeline integrating simulation and world models, with real-world validation. However, it lacks comparison to existing methods and focuses narrowly on autonomous driving.

Read-first score

Read-first score 67.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 27.

Methodology quality 25%
90

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

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Reproducibility 25%
85

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

Topical relevance 42%
38.6

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

Field roles

FrontierMethodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 564.

Keyword Scores

world model
10
generative world model
9
world simulator
7
video world model
1
interactive world model
0
world dynamics prediction
0
model-based reinforcement learning world model
0

Deep Analysis

Innovations

  • Proposes a simulator-conditioned scene generation engine that combines a simulation system with a world model for controllable data generation.
  • Introduces a unified benchmark with proportionally constructed virtual and real data to evaluate world models in real-world scenes.
  • Novel data generation pipeline that leverages the scene simulation capabilities of a simulation engine and the robust data generation capabilities of a world model.

Methodology

The method builds a simulation system consistent with real-world scenes to collect simulation data and labels, which serve as conditions for a world model to generate images. It combines the simulation engine's scene simulation with the world model's data generation, and provides a benchmark mixing virtual and real data for evaluation.

Key Results

Quantitative results show that images generated by the proposed pipeline significantly improve the performance of downstream perception models. The generative performance was explored specifically in urban autonomous driving scenarios.

Limitations

  • Evaluation is limited to urban autonomous driving scenarios, leaving other driving environments or application domains unexplored.
  • The approach depends on the simulation system's consistency with real-world scenes; any domain gap between simulation and reality may affect the quality and usefulness of generated data.

Tags