MindDrive: An All-in-One Framework Bridging World Models and Vision-Language Model for End-to-End Autonomous Driving

arXiv 25.12 2025 62.7 method, system

TLDR

MindDrive integrates a world action model for future scene prediction and a VLM for multi-objective evaluation, achieving SOTA on NAVSIM benchmarks.

Reasoning

The paper presents a novel framework combining world models and vision-language models for end-to-end autonomous driving, with clear strengths in integrating trajectory generation and reasoning. However, the abstract lacks details on the world model architecture and training, and the benchmarks may not fully reflect real-world complexity.

Read-first score

Read-first score 62.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 47.

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=benchmark,evaluation,experiment,metric

Topical relevance 42%

67.1

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

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 168.

Keyword Scores

world model

generative world model

world dynamics prediction

world simulator

interactive world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Integration of world models and vision-language models for end-to-end autonomous driving
Future-aware Trajectory Generator (FaTG) based on a World Action Model (WaM) for ego-conditioned 'what-if' simulations
VLM-oriented Evaluator (VLoE) leveraging large vision-language model for multi-objective evaluation across safety, comfort, and efficiency

Methodology

MindDrive proposes a structured reasoning paradigm of 'context simulation - candidate generation - multi-objective trade-off'. It uses a Future-aware Trajectory Generator (FaTG) based on a World Action Model (WaM) to predict potential future scenes and generate foresighted trajectory candidates, then employs a VLM-oriented Evaluator (VLoE) to perform multi-objective evaluations across safety, comfort, and efficiency dimensions. The framework is evaluated on the NAVSIM-v1 and NAVSIM-v2 benchmarks.

Key Results

MindDrive achieves state-of-the-art performance across multi-dimensional driving metrics on NAVSIM-v1 and NAVSIM-v2 benchmarks, significantly enhancing safety, compliance, and generalization.

Resources

Paper