GeoDrive: 3D Geometry-Informed Driving World Model with Precise Action Control

arXiv 25.5 2025 74 method

TLDR

GeoDrive integrates 3D geometry into driving world models for precise action control, improving spatial awareness and scene modeling.

Reasoning

The paper introduces a novel 3D geometry-informed driving world model with action control, showing strong results in spatial awareness and accuracy. However, the abstract lacks explicit mention of real-world datasets or benchmarks, and limitations such as occlusion handling are not fully addressed.

Read-first score

Read-first score 74, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 53.

Recency 8%

86.7

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

Reproducibility 25%

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

Topical relevance 42%

75.7

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

Methodology quality 25%

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

Field roles

FrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 48.

Keyword Scores

world model

interactive world model

world simulator

world dynamics prediction

generative world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Explicit integration of robust 3D geometry conditions into driving world models to enhance spatial understanding and action controllability
Dynamic editing module during training that edits vehicle positions to improve renderings for dynamic modeling
Interactive scene editing capabilities including object editing and object trajectory control

Methodology

GeoDrive first extracts a 3D representation from the input frame, then obtains a 2D rendering based on a user-specified ego-car trajectory. A dynamic editing module is introduced during training to enhance the renderings by editing the positions of vehicles, enabling dynamic modeling.

Key Results

The method significantly outperforms existing models in both action accuracy and 3D spatial awareness, and can generalize to novel trajectories while offering interactive scene editing capabilities.

Resources

Paper Code