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

ADriver-I: A General World Model for Autonomous Driving

arXiv 23.9 2023 73.3 method, application

TLDR

ADriver-I is a general world model for autonomous driving using interleaved vision-action pairs, MLLM, and diffusion to predict control signals and future frames.

Reasoning

The paper introduces a novel world model that unifies perception and control via vision-action pairs, demonstrating strong empirical results on nuScenes and private datasets. However, the abstract lacks details on limitations, baseline comparisons, and ablation studies, and the reliance on private datasets may hinder reproducibility.

Read-first score

Read-first score 73.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 56.

Methodology quality 25%
100

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

Topical relevance 42%
80

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

Recency 8%
65.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2023

Reproducibility 25%
38

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

Field roles

Methodology anchor

Rank sensitivity

Stability: volatile; rank range: 82.

Keyword Scores

world model
10
generative world model
9
world dynamics prediction
9
interactive world model
8
video world model
8
world simulator
7
model-based reinforcement learning world model
5

Deep Analysis

Innovations

  • Introduction of interleaved vision-action pair to unify visual features and control signals
  • Construction of a general world model combining MLLM and diffusion model for autonomous driving
  • Autoregressive prediction of control signals and future frames, enabling driving in a self-created world

Methodology

ADriver-I uses interleaved vision-action pairs as input to a multimodal large language model (MLLM) and diffusion model. It autoregressively predicts the control signal for the current frame, then conditions on historical vision-action pairs and the generated control to predict future frames. This process repeats infinitely, allowing the model to drive in a world it generates itself.

Key Results

Experiments on nuScenes and a large-scale private dataset show ADriver-I achieves impressive performance compared to several constructed baselines.

Limitations

  • No quantitative results or specific metrics reported in the abstract
  • Evaluation relies on a private dataset, limiting reproducibility
  • Comparison only to constructed baselines, not to state-of-the-art modular or end-to-end systems
  • Potential error accumulation in the autoregressive loop over long horizons

Resources

Paper

Tags