Scenario-based testing is becoming increasingly important in safety assurance for automated driving. However, comprehensive and sufficiently complete coverage of the scenario space requires significant effort and resources if using only real-world data. To address this issue, driving scenario generation methods are developed and used more frequently, but the benefit of substituting generated data for real-world data has not yet been quantified. Additionally, the coverage of a set of concrete scenarios within a given logical scenario space has not been predicted yet. This paper proposes a methodology to quantify the cost-optimal usage of scenario generation approaches to reach a certainly complete scenario space coverage under given quality constraints and parametrization. Therefore, individual process steps for scenario generation and usage are investigated and evaluated using a meta model for the abstraction of knowledge-based and data-driven methods. Furthermore, a methodology is proposed to fit the meta model including the prediction of reachable complete coverage, quality criteria, and costs. Finally, the paper exemplary examines the suitability of a hybrid generation model under technical, economical, and quality constraints in comparison to different real-world scenario mining methods.

翻译：基于场景的测试在自动驾驶安全验证中日益重要。然而，若仅使用真实世界数据，实现场景空间的完整且充分覆盖需要耗费大量资源。为此，驾驶场景生成方法被开发并广泛应用，但生成数据替代真实数据的效益尚未量化。此外，在给定逻辑场景空间内，具体场景集的覆盖程度也未能预测。本文提出一种方法论，用于量化场景生成方法在给定质量约束与参数配置下实现场景空间确定完整覆盖的成本最优方案。通过元模型对知识驱动与数据驱动方法进行抽象，对场景生成与使用的各流程环节开展研究评估。进一步提出涵盖可达完整覆盖率预测、质量准则及成本评估的元模型拟合方法。最后，通过实例对比分析混合生成模型在不同技术、经济与质量约束下，相较于多种真实场景挖掘方法的适用性。