GaussianDWM: 3D Gaussian Driving World Model for Unified Scene Understanding and Multi-Modal Generation

arXiv 25.12 2025 52.8 method, application

TLDR

Proposes GaussianDWM, a 3D Gaussian driving world model for unified scene understanding and multi-modal generation, evaluated on nuScenes and NuInteract.

Reasoning

Strengths include novel 3D Gaussian representation for text-scene alignment, task-aware sampling, and dual-condition generation. Weaknesses: limited to driving domain, abstract cut off, no baseline comparisons mentioned.

Read-first score

Read-first score 52.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 32.

Recency 6%

86.7

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

Reproducibility 18%

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

Citation impact 18%

59.4

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.59381572

Methodology quality 18%

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

Topical relevance 29%

45.7

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

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 499.

Keyword Scores

world model

generative world model

video world model

world simulator

world dynamics prediction

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Unified DWM framework based on 3D Gaussian scene representation enabling both 3D scene understanding and multi-modal generation
Early modality alignment by embedding linguistic features into each Gaussian primitive
Task-aware language-guided sampling strategy to remove redundant 3D Gaussians and inject compact 3D tokens into LLM
Dual-condition multi-modal generation model with high-level language condition and low-level image condition

Methodology

The proposed GaussianDWM framework uses 3D Gaussian scene representation with embedded linguistic features for early modality alignment. It employs a task-aware language-guided sampling strategy to select compact 3D tokens for LLM, and a dual-condition generation model combining high-level language and low-level image conditions for multi-modal generation.

Key Results

The method achieves state-of-the-art performance on nuScenes and NuInteract datasets.

Resources

Paper Code