LiDARCrafter: Dynamic 4D World Modeling from LiDAR Sequences

arXiv 25.8 2025 63.1 method, system, application

TLDR

LiDARCrafter generates and edits dynamic 4D LiDAR sequences from natural language using a tri-branch diffusion network, achieving state-of-the-art performance on nuScenes.

Reasoning

The paper presents a novel framework for LiDAR-based world modeling with strong controllability and temporal coherence, supported by a comprehensive benchmark. However, it is limited to LiDAR modality and evaluated only on nuScenes, lacking comparison to video-based methods.

Read-first score

Read-first score 63.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Methodology quality 25%

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

Recency 8%

86.7

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

Topical relevance 42%

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 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 333.

Keyword Scores

world model

generative world model

world dynamics prediction

world simulator

video world model

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Unified framework for 4D LiDAR generation and editing from free-form natural language inputs
Parsing instructions into ego-centric scene graphs to condition a tri-branch diffusion network
Tri-branch diffusion network generating object structures, motion trajectories, and geometry
Autoregressive module for temporally coherent 4D LiDAR sequences with smooth transitions
Comprehensive benchmark with diverse metrics spanning scene-, object-, and sequence-level aspects

Methodology

LiDARCrafter uses a tri-branch diffusion network conditioned on ego-centric scene graphs parsed from natural language instructions to generate object structures, motion trajectories, and geometry. An autoregressive module ensures temporal coherence across 4D LiDAR sequences. The model is evaluated on the nuScenes dataset using a newly established benchmark with scene-, object-, and sequence-level metrics.

Key Results

LiDARCrafter achieves state-of-the-art performance in fidelity, controllability, and temporal consistency across all levels on the nuScenes dataset.

Resources

Paper Project