Minimum Critical Test Scenario Set Selection for Autonomous Vehicles Prior to First Deployment and Public Road Testing

Abstract

The growing complexity of autonomous vehicle functionalities poses significant challenges for vehicle testing, validation, and regulatory approval. Despite the availability of various testing protocols and standards, a harmonized and widely accepted method specifically targeting the selection of critical test scenarios—especially for safety assessments prior to public road testing—has not yet been developed. This study introduces a systematic methodology for selecting a minimum critical set of test scenarios tailored to an autonomous vehicle’s Operational Design Domain (ODD) and capabilities. Building on existing testing frameworks (e.g., EuroNCAP protocols, ISO standards, UNECE and EU regulations), the proposed method combines a structured questionnaire with a weighted cosine similarity based filtering mechanism to identify relevant scenarios from a robust database of over 1000 test cases. Further refinement using similarity metrics such as Euclidean and Manhattan distances ensures the elimination of redundant test scenarios. Application of the framework to real-world projects demonstrates significant alignment with expert-identified cases, while also identifying overlooked but relevant scenarios. By addressing the need for a structured and efficient scenario selection method, this work supports the advancement of systematic safety assurance for autonomous vehicles and provides a scalable solution for authorities and vehicle testing companies.