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

VL-SAFE: Vision-Language Guided Safety-Aware Reinforcement Learning with World Models for Autonomous Driving

arXiv 25.5 2025 62.9 method, application

TLDR

Proposes VLM-SAFE, an offline safe RL framework integrating vision-language models with world models for semantic safety-aware autonomous driving.

Reasoning

Strengths include novel integration of VLM for semantic safety signals with world model for imagination, addressing sample efficiency and safety. Weaknesses: abstract lacks details on experimental validation and real-world performance; reliance on offline data may limit generalization.

Read-first score

Read-first score 62.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 8%
86.7

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

Methodology quality 25%
80

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

Topical relevance 42%
62.9

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%
38

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 136.

Keyword Scores

world model
10
world dynamics prediction
9
model-based reinforcement learning world model
9
generative world model
6
world simulator
5
video world model
3
interactive world model
2

Deep Analysis

Innovations

  • Integration of a vision-language model (VLM) to provide semantic safety signals grounded in scene understanding for autonomous driving RL
  • Use of a learned world model to imagine future trajectories from observed context, enabling offline reasoning without real-world interaction
  • VLM-based safety evaluation of imagined futures to explicitly couple future anticipation with semantic risk assessment
  • Unified closed-loop framework (observe-imagine-evaluate-act) for safer and more efficient offline policy learning

Methodology

VL-SAFE is an offline safe RL framework that follows a human cognitive loop of observe-imagine-evaluate-act. It uses offline driving data, a VLM to provide semantic safety signals, a world model to predict future trajectories, and VLM-based safety evaluation of those imagined futures to guide actor-critic policy optimization.

Key Results

In simulation experiments, VL-SAFE achieves improved safety, stronger robustness under traffic-density shift, and a better safety-performance trade-off compared to representative baselines.

Tags