Safety2Drive: Safety-Critical Scenario Benchmark for the Evaluation of Autonomous Driving

TL;DR

Safety2Drive tackles the lack of regulatory-compliant, closed-loop scenario libraries for autonomous driving safety evaluation by introducing a comprehensive benchmark with 70 regulatory test items implemented in OpenSCENARIO. It couples scenario construction with LLM-assisted generation, and introduces scenario generalization through natural environmental corruptions and adversarial attacks to stress-test perception and driving systems. The framework supports end-to-end evaluation of both perception tasks and driving agents, and provides a driving-agent leaderboard to compare multiple AD systems under realistic safety threats. The work enables standardized, multi-dimensional safety testing with practical implications for regulator-aligned validation and robust deployment of autonomous vehicles.

Abstract

Autonomous Driving (AD) systems demand the high levels of safety assurance. Despite significant advancements in AD demonstrated on open-source benchmarks like Longest6 and Bench2Drive, existing datasets still lack regulatory-compliant scenario libraries for closed-loop testing to comprehensively evaluate the functional safety of AD. Meanwhile, real-world AD accidents are underrepresented in current driving datasets. This scarcity leads to inadequate evaluation of AD performance, posing risks to safety validation and practical deployment. To address these challenges, we propose Safety2Drive, a safety-critical scenario library designed to evaluate AD systems. Safety2Drive offers three key contributions. (1) Safety2Drive comprehensively covers the test items required by standard regulations and contains 70 AD function test items. (2) Safety2Drive supports the safety-critical scenario generalization. It has the ability to inject safety threats such as natural environment corruptions and adversarial attacks cross camera and LiDAR sensors. (3) Safety2Drive supports multi-dimensional evaluation. In addition to the evaluation of AD systems, it also supports the evaluation of various perception tasks, such as object detection and lane detection. Safety2Drive provides a paradigm from scenario construction to validation, establishing a standardized test framework for the safe deployment of AD.

Figures (10)

• Figure 1: The pipeline of Safety2Drive.
• Figure 2: Example of double lane changer scenario.
• Figure 3: Visualization of four scenarios: Cut In, Cut Out, Decelerating, and Pedestrian Crossing, as well as the corresponding BEV. The blue box is the Ego vehicle and the surrounding vehicle and pedestrian are in red.
• Figure 4: The visualization of safety-critical scenario generalization. The top is natural environmental corruption and the bottom is adversarial attack.
• Figure 5: The visualization of prediction results for different perception tasks.