Is Your VLM for Autonomous Driving Safety-Ready? A Comprehensive Benchmark for Evaluating External and In-Cabin Risks

TL;DR

DSBench presents the first comprehensive driving safety benchmark that unifies external hazard perception with in-cabin driver-state monitoring to evaluate Vision-Language Models in autonomous driving. By building a fine-grained 10-by-28 safety taxonomy, aggregating 98K QA pairs and 3K test scenes from 10+ datasets, and employing GPT-4o-based evaluation, the framework reveals safety weaknesses in current VLMs. A DSBench-tuned model, DSVLM, achieves state-of-the-art safety performance across categories, notably in cockpit understanding, demonstrating the value of safety-focused fine-tuning. The work underscores the need for integrated, human-centric evaluation in safety-critical AI systems and provides a practical dataset, tooling, and methodology to advance trustworthy autonomous driving technologies.

Abstract

Vision-Language Models (VLMs) show great promise for autonomous driving, but their suitability for safety-critical scenarios is largely unexplored, raising safety concerns. This issue arises from the lack of comprehensive benchmarks that assess both external environmental risks and in-cabin driving behavior safety simultaneously. To bridge this critical gap, we introduce DSBench, the first comprehensive Driving Safety Benchmark designed to assess a VLM's awareness of various safety risks in a unified manner. DSBench encompasses two major categories: external environmental risks and in-cabin driving behavior safety, divided into 10 key categories and a total of 28 sub-categories. This comprehensive evaluation covers a wide range of scenarios, ensuring a thorough assessment of VLMs' performance in safety-critical contexts. Extensive evaluations across various mainstream open-source and closed-source VLMs reveal significant performance degradation under complex safety-critical situations, highlighting urgent safety concerns. To address this, we constructed a large dataset of 98K instances focused on in-cabin and external safety scenarios, showing that fine-tuning on this dataset significantly enhances the safety performance of existing VLMs and paves the way for advancing autonomous driving technology. The benchmark toolkit, code, and model checkpoints will be publicly accessible.

Figures (11)

• Figure 1: Overview of DSBench. A comprehensive benchmark for evaluating the safety of Vision-Language Models (VLMs) in autonomous driving. The diagram highlights 10 key categories related to in-cabin and external safety scenarios. Each category addresses critical aspects such as driver operation, environmental conditions, and situational awareness, all of which are essential for improving the safety and effectiveness of autonomous driving systems.
• Figure 2: The Pipeline of Data Extraction, Annotation, and Selection Process. Our annotation process consists of four key components: data extraction and integration (Sec. \ref{['sec:3.1']}), data analysis and preliminary categorization (Sec. \ref{['sec:3.2']}), expansion of question-answer templates (Sec. \ref{['sec:3.3']}), and manual quality control (Sec. \ref{['sec:3.4']}).
• Figure 3: Overview of the Detailed Categories in DSBench. Our benchmark is organized into 10 major categories and 28 more granular subcategories, encompassing a comprehensive spectrum of potential safety issues encountered in driving scenarios.
• Figure 4: Performance Comparison in DSBench. Our DSVLM has achieved state-of-the-art performance across all safety categories and made significant breakthroughs in Coc., a category where existing models have demonstrated limited capabilities.
• Figure 5: Visualization Results of Existing Models on DSBench. It reveals that, when confronted with both in-cabin and external safety challenges, our fine-tuned model, DSVLM, consistently delivers more comprehensive, reasonable, and factually grounded responses. Additionally, its outputs exhibit greater linguistic fluency and word choices that align more closely with natural human preferences.