Latent World Models for Automated Driving: A Unified Taxonomy, Evaluation Framework, and Open Challenges

Researchers propose a unified framework for latent world models in automated driving, organizing recent advances in generative AI and vision-language-action systems. The framework addresses scalable simulation, long-horizon forecasting, and decision-making through latent representations that compress multi-sensor data.

• →The paper introduces a taxonomy for organizing latent world models by representation types and structural priors in autonomous driving.
• →Five key internal mechanics are identified including structural isomorphism, temporal stability, and semantic alignment for robust AI systems.
• →A closed-loop metric suite is proposed to address evaluation gaps between open-loop and closed-loop performance in autonomous vehicles.
• →The framework connects design choices to critical factors like robustness, generalization, and real-world deployability.
• →Research directions are outlined for achieving decision-ready, verifiable, and resource-efficient automated driving systems.