Advancing autonomous vehicle safety assessment: A novel methodology for moving from functional to concrete scenarios using kinetic 3D-LiDAR and SHAP

Abstract

The commercialization of autonomous driving systems is gaining momentum, yet ensuring safety remains an ongoing challenge. To address these safety concerns, autonomous vehicle safety assessment employs a scenario-based approach, which can be categorized into knowledge-based and data-driven methods. In this study, we propose a framework that extends existing data-driven approaches by addressing three critical limitations: data, method, and scenario development. Using actual driving data, including 3D LiDAR Point Cloud Data (3D-LiDAR PCD), we extracted kinetic properties such as vehicle speed, acceleration, and detected critical accident triggered vehicle (ATV). Subsequently, we employed SHAP (SHapley Additive exPlanations) to assess the importance of kinetic properties and set criteria for selecting the configuration value into scenario. Specifically, we used SHAP value to determine the optimal configuration values for each variable in concrete scenario. This study presents a comprehensive scenario development framework that not only overcomes data limitations but also provides a methodological foundation for developing scenarios that accurately reflect real world. It offers an innovative approach to addressing safety concerns in the commercialization of autonomous driving systems.

Keywords

• Autonomous vehicle
• Safety assessment
• Concrete scenario
• LiDAR
• Kinetic property
• XAI