Drive&Gen: Co-Evaluating End-to-End Driving and Video Generation Models

IROS 25 2025 51.1 benchmark, application

TLDR

Proposes DriveGen to co-evaluate end-to-end driving and video generation models using statistical measures and synthetic data.

Reasoning

Strengths include novel statistical measures for evaluating video realism via driving models and using controllable generation to study distribution gaps. Weaknesses are limited methodological details and lack of quantitative results in the abstract.

Read-first score

Read-first score 51.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 36.

Recency 8%

86.7

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

Methodology quality 25%

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

Topical relevance 42%

51.4

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=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 404.

Keyword Scores

generative world model

world model

video world model

world simulator

world dynamics prediction

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Novel statistical measures leveraging end-to-end driving models to evaluate the realism of generated videos
Using controllability of video generation models to investigate distribution gaps affecting end-to-end planner performance
Demonstrating that synthetic data from video generation models offers a cost-effective alternative to real-world data for improving end-to-end model generalization beyond existing Operational Design Domains

Methodology

The paper proposes statistical measures that use end-to-end driving models to evaluate the realism of generated videos. It exploits the controllability of video generation models to conduct targeted experiments investigating distribution gaps that affect end-to-end planner performance. Finally, it uses synthetic data produced by the video generation model to improve end-to-end model generalization.

Key Results

Synthetic data produced by the video generation model effectively improves end-to-end model generalization beyond existing Operational Design Domains, facilitating the expansion of autonomous vehicle services into new operational contexts.

Resources

Paper