The Safety Challenge of World Models for Embodied AI Agents: A Review

arXiv 25.10 2025 57.2 survey

TLDR

A review of world models for embodied AI focusing on safety, with empirical analysis of prediction faults in autonomous driving and robotics.

Reasoning

The paper provides a comprehensive literature review and includes an empirical analysis of state-of-the-art models, identifying common faults (pathologies) and offering quantitative evaluation, which is a strength. However, it is limited to autonomous driving and robotics domains, and the abstract does not detail specific methodologies or results beyond fault categorization.

Read-first score

Read-first score 57.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 42.

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=analysis,evaluation,result

Topical relevance 42%

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: 269.

Keyword Scores

world model

world dynamics prediction

generative world model

model-based reinforcement learning world model

world simulator

interactive world model

video world model

Deep Analysis

Innovations

Comprehensive literature review of World Models in autonomous driving and robotics with a specific focus on safety implications of scene and control generation tasks.
Empirical analysis that collects predictions from state-of-the-art models, identifies and categorizes common faults (pathologies), and provides a quantitative evaluation.

Methodology

The authors conduct a comprehensive literature review of World Models in autonomous driving and robotics, focusing on safety implications of scene and control generation tasks. They complement this with an empirical analysis where they collect predictions from state-of-the-art models, identify and categorize common faults (pathologies), and provide a quantitative evaluation of the results.

Key Results

The empirical analysis identifies and categorizes common faults (pathologies) in predictions from state-of-the-art World Models, with a quantitative evaluation of these faults.

Resources

Paper