厘清自动驾驶场景的三类定义，给出统一形式化方法。
On Scenario Formalisms for Automated Driving
- 用曲线、集合映射和时序逻辑分别形式化具体、逻辑与抽象场景。
- 对比了两类场景在表达力、复杂度、采样与监控上的优劣。
- 帮助从业者选择合适场景形式化工具，避免概念混淆。
场景及其多种类型（特别是逻辑场景与抽象场景）已成为保障自动驾驶系统安全的核心要素。然而，原始的语言定义常被模糊使用，导致实际中提出或标准化的场景描述语言与其术语基础脱节，造成对逻辑场景与抽象场景独特性及其优劣势的困惑。为此，本文为具体、逻辑与抽象场景提供了清晰的语言定义，并分别提出了通用统一的形式化方法：使用曲线、到曲线集合的映射以及时序逻辑。我们证明这些形式化方法可精准定位逻辑场景与抽象场景在表达能力、规格复杂度、采样效率与监控可行性方面的差异。本工作使实践者能够理解不同场景类型的特征，并选择合适的正式框架。
原文摘要 · Abstract (English)
The concept of scenario and its many qualifications -- specifically logical and abstract scenarios -- have emerged as a foundational element in safeguarding automated driving systems. However, the original linguistic definitions of the different scenario qualifications were often applied ambiguously, leading to a divergence between scenario description languages proposed or standardized in practice and their terminological foundation. This resulted in confusion about the unique features as well as strengths and weaknesses of logical and abstract scenarios. To alleviate this, we give clear linguistic definitions for the scenario qualifications concrete, logical, and abstract scenario and propose generic, unifying formalisms using curves, mappings to sets of curves, and temporal logics, respectively. We demonstrate that these formalisms allow pinpointing strengths and weaknesses precisely by comparing expressiveness, specification complexity, sampling, and monitoring of logical and abstract scenarios. Our work hence enables the practitioner to comprehend the different scenario qualifications and identify a suitable formalism.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。