Drive-P2D: A Progressive Perception-to-Decision Benchmark for VLMs in Autonomous Driving

The introduction of Drive-P2D, a progressive perception-to-decision benchmark, aims to enhance the evaluation of vision-language models (VLMs) in autonomous driving by providing 6,650 questions across various levels of complexity. This benchmark addresses the limitations of existing evaluations that often separate perception and decision-making, thereby improving the reliability of autonomous driving systems.

This development is significant as it allows for a more nuanced understanding of how VLMs can be assessed in high-risk scenarios, ultimately contributing to safer autonomous driving technologies. By adopting a separated reasoning-and-answer protocol, Drive-P2D enables a detailed analysis of error modes, which is crucial for advancing the field.

The emergence of Drive-P2D reflects a broader trend in autonomous driving research that emphasizes the integration of perception and decision-making processes. This aligns with recent advancements such as OmniDrive-R1 and AutoMoT, which also focus on enhancing the reliability and performance of VLMs. As the industry grapples with challenges like object hallucination and safety-critical applications, these benchmarks and models collectively push the boundaries of what is achievable in autonomous driving.