自动驾驶模拟：用于验证 ADAS 和 ADS 的道路仿真

每年，汽车製造商和供应商都在不断扩大 ADAS（高级驾驶辅助系统）和 ADS（自动驾驶系统）的范围，以补充或取代人类驾驶员。其主要目标之一是透过减少碰撞来提高道路安全。然而，驾驶是人类经常执行的高度复杂的任务。光是在美国，人们每年驾驶 3.2 兆英里，但平均每 50 万英里仅发生一次事故。

验证 ADAS 和 ADS 实际上比人类驾驶员更安全是一个非常高的障碍，特别是考虑到驾驶环境几乎无限的变化以及重现测试条件的难度。过去几十年来，模拟已成为验证汽车安全系统的重要工具，对于证明 ADAS/ADS 的有效性也至关重要。

ADAS/ADS 开发人员使用开发和测试过程的每一步来验证新概念，并确保对已证明有效的系统进行更改不会引入错误。在各个阶段使用各种模拟工具。将软体、硬体和推动程式合併到环路中的开环和闭环模拟都被广泛使用。还需要用于产生测试场景的自动化工具，以确保测试套件的足够覆盖范围。大多数模拟工作流程结合了来自多个供应商的各种工具，有助于确保 ADAS 和 ADS 在部署到公共道路上之前有助于提高安全性。

目录

• 序论
• 背景情况
• 推荐事项
• 由于 ADAS/ADS 的出现，对模拟的需求不断增加
• 在虚拟空间中再现实体驾驶环境
  • 运算平台
  • 模拟的种类
    • 单元/子系统模拟
    • 全端模拟
    • 循环中的软体
    • 循环中的硬体
    • 循环中的推动程式
  • 模式和情势
    • 车辆的物理的建模
    • 感测器建模
    • 景色建模
    • 情势建立
    • 模式的检验
• 安全自动化需要虚拟验证
  • 收集和分享基础设施数据
  • 走向监管

With each passing year, automakers and suppliers are continuing to expand the scope of what advanced driver assistance systems (ADAS) and automated driving systems (ADS) can do to supplement or replace human drivers. One of the primary goals is to improve road safety by reducing the number of crashes. However, driving is a very complex task that humans do with a very high frequency. In the US alone, people drive as much as 3.2 trillion miles per year and only crash about once every half million miles on average.

Validating that ADAS and ADS are actually safer than human drivers is a very high bar, particularly given the nearly infinite variability of the driving environment and the difficulty of reproducing test conditions. Simulation has become a crucial tool for validating automotive safety systems over the past several decades, and it is essential for proving the efficacy of ADAS/ADS.

ADAS/ADS developers rely on a range of simulation tools at all stages of the development and testing process to validate new concepts and ensure that changes have not caused errors in systems that are already demonstrated to work. Open- and closed-loop simulations with software, hardware, and drivers in the loop are all being used extensively. Automated tools to generate testing scenarios are also needed to ensure sufficient coverage of the test suite. Most simulation workflows combine a range of tools from multiple vendors to help guarantee that ADAS and ADS contribute to improved safety before the technology is deployed on public roads.

Table of Contents

Spark

Context

Recommendations

The Emergence of ADAS and ADS Drives Simulation Demand

Replicating the Physical Driving Environment in Virtual Space

Compute Platforms

Simulation Types

Unit and Subsystem Simulation

Full-Stack Simulation

Software in the Loop

Hardware in the Loop

Driver in the Loop

Models and Scenarios

Vehicle Physics Modeling

Sensor Modeling

Scene Modeling

Scenario Building

Model Validation

Safe Automation Needs Virtual Validation

Collecting and Sharing Infrastructure Data

Looking toward Regulations

List of Figures

• Annual Light Duty Vehicle Deployments by Automation Level, World Markets: 2022-2031

NVIDIA DRIVE Sim's Virtual Driving Environments and Simulated Sensor Inputs

VI-grade's DiM150 Dynamic Driving Simulator at Ford Product Development Center

Vehicle Physics Simulation Model