The rapid evolution of autonomous vehicles (AVs) has exposed the need for fast-paced development and testing processes of a variety of perception, planning, and control algorithms. To expedite development, the AV industry and researchers leverage virtual vehicle environments to simulate a range of test scenarios that may otherwise be costly or difficult to conduct on a real test track. However, the various virtual environments may have different results depending on the fidelity of various simulation features, such as vehicle dynamics, sensor simulation, and environment recreation. This tutorial article examines a proposed framework for constructing, parameterizing, and validating a virtual vehicle environment using an existing AV data set. First, an overview of several open source and commercially available simulation tools, including their associated workflows, for scene and scenario creation is presented. Next, various open AV data sets are examined to inform the data set selection for the validation framework. Then, an example workflow of recreating a real-world scene from the selected data set in a simulation tool with various emulated sensors parameterized to match the data set is demonstrated. Finally, an example AV-perception algorithm is subjected to data streams from virtual and real-world environments and suggested metrics for analyzing the results are discussed.

中文翻译：

---

模拟自动驾驶未来：了解虚拟车辆环境以及如何使用公共数据集验证模拟

自动驾驶汽车 (AV) 的快速发展暴露了对各种感知、规划和控制算法的快节奏开发和测试过程的需求。为了加快开发速度，AV 行业和研究人员利用虚拟车辆环境来模拟一系列测试场景，否则这些场景在真实的测试轨道上进行可能会很昂贵或难以进行。然而，各种虚拟环境可能会产生不同的结果，这取决于各种模拟特征的保真度，例如车辆动力学、传感器模拟和环境再现。本教程文章检查了使用现有 AV 数据集构建、参数化和验证虚拟车辆环境的建议框架。首先，概述几个开源和市售的仿真工具，包括它们相关的工作流程，用于场景和场景创建。接下来，检查各种开放的 AV 数据集，以告知验证框架的数据集选择。然后，演示了使用模拟工具中的选定数据集重新创建真实世界场景的示例工作流程，其中各种模拟传感器参数化以匹配数据集。最后，一个示例 AV 感知算法受到来自虚拟和现实世界环境的数据流的影响，并讨论了用于分析结果的建议指标。演示了从模拟工具中的选定数据集重新创建真实世界场景的示例工作流程，其中各种模拟传感器参数化以匹配数据集。最后，一个示例 AV 感知算法受到来自虚拟和现实世界环境的数据流的影响，并讨论了用于分析结果的建议指标。演示了从模拟工具中的选定数据集重新创建真实世界场景的示例工作流程，其中各种模拟传感器参数化以匹配数据集。最后，一个示例 AV 感知算法受到来自虚拟和现实世界环境的数据流的影响，并讨论了用于分析结果的建议指标。