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

Scaling Up Occupancy-centric Driving Scene Generation: Dataset and Method

arXiv 25.10 2025 53.8 method, benchmark, application

TLDR

Introduces Nuplan-Occ dataset and a unified framework for generating semantic occupancy, multi-view videos, and LiDAR point clouds for autonomous driving.

Reasoning

Strengths include a large-scale dataset and novel techniques for multi-modal generation; weaknesses are that the paper does not explicitly address world models or reinforcement learning, limiting relevance to those keywords.

Read-first score

Read-first score 53.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 17.

Methodology quality 25%
100

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

Recency 8%
86.7

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

Reproducibility 25%
46

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

Topical relevance 42%
24.3

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 639.

Keyword Scores

world dynamics prediction
6
video world model
4
generative world model
3
world model
2
world simulator
2
interactive world model
0
model-based reinforcement learning world model
0

Deep Analysis

Innovations

  • Nuplan-Occ: the largest semantic occupancy dataset to date, constructed from the Nuplan benchmark
  • Unified framework that jointly synthesizes high-quality semantic occupancy, multi-view videos, and LiDAR point clouds
  • Spatio-temporal disentangled architecture for high-fidelity spatial expansion and temporal forecasting of 4D dynamic occupancy
  • Gaussian splatting-based sparse point map rendering strategy to enhance multi-view video generation
  • Sensor-aware embedding strategy that explicitly models LiDAR sensor properties for realistic multi-LiDAR simulation

Methodology

The authors curate Nuplan-Occ, a large-scale semantic occupancy dataset derived from the Nuplan benchmark. They develop a unified framework with a spatio-temporal disentangled architecture to jointly generate 4D dynamic occupancy, multi-view videos, and LiDAR point clouds. Two novel techniques are introduced: Gaussian splatting-based sparse point map rendering for multi-view video generation, and sensor-aware embedding for realistic multi-LiDAR simulation.

Key Results

Extensive experiments demonstrate that the method achieves superior generation fidelity and scalability compared to existing approaches, and validates its practical value in downstream tasks.

Limitations

  • The method still depends on annotated occupancy data for training, albeit using a newly curated large-scale dataset
  • Generalization to other driving datasets beyond Nuplan-Occ is not evaluated

Resources

Paper

Tags