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

LaGen: Towards Autoregressive LiDAR Scene Generation

arXiv 25.11 2025 40.5 method, system, application

TLDR

LaGen is the first autoregressive LiDAR scene generation framework for long-horizon interactive generation using bounding box conditions.

Reasoning

The paper introduces a novel autoregressive approach for LiDAR scene generation, addressing limitations of prior single-frame or deterministic methods. Strengths include long-horizon interactive capability and strong empirical results on nuScenes. Weaknesses are the focus solely on LiDAR modality and lack of generalization discussion.

Read-first score

Read-first score 40.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 39.

Recency 6%
86.7

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

Methodology quality 18%
70

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

Topical relevance 29%
55.7

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=dataset

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: 250.

Keyword Scores

generative world model
9
world model
8
interactive world model
7
world simulator
6
world dynamics prediction
6
video world model
2
model-based reinforcement learning world model
1

Deep Analysis

Innovations

  • First framework capable of frame-by-frame autoregressive generation of long-horizon LiDAR scenes
  • Scene decoupling estimation module to enhance interactive generation capability for object-level content
  • Noise modulation module to mitigate error accumulation during long-horizon generation
  • Utilizing bounding box information as conditions for generating high-fidelity 4D scene point clouds
  • Construction of a protocol based on nuScenes for evaluating long-horizon LiDAR scene generation

Methodology

LaGen is an autoregressive model that takes a single-frame LiDAR input and uses bounding box information as conditions to generate subsequent frames. It incorporates a scene decoupling estimation module for object-level interactive generation and a noise modulation module to reduce error accumulation. The model is evaluated on the nuScenes dataset, comparing against state-of-the-art LiDAR generation and prediction models.

Key Results

LaGen outperforms state-of-the-art LiDAR generation and prediction models, especially on the later frames of long-horizon generation.

Resources

Paper

Tags