基于模型的

(1)

Siemens PLM Software

基于模型的ADAS和自动驾驶系统性能开发验证方法论

Digitalization of the automotive industry

Restricted © Siemens AG 2017

(2)

驱动汽车行业创新的趋势

自动化、电气化、网联化以及共享化

(3)

产品系统数字化双胞胎技术

全面满足智能网联新能源车的研发需要

智能网联新能源车研发解决方案

自动化

感知决策执行

电气化

电池电驱电控

网联化

人车环境从辅助驾驶

到自动驾驶

从内燃机到混动到纯电动

从孤立的车到系统的系统

确保协同、数字化的持续性、多领域的追溯性以及功能安全

校核&验证：仿真开发 & 测试

(4)

产品系统数字化双胞胎技术

全面满足智能网联新能源车的研发需要

智能网联新能源车研发解决方案

自动化

感知决策执行

电气化

电池电驱电控

网联化

人车环境从辅助驾驶

到自动驾驶

从内燃机到混动到纯电动

从孤立的车到系统的系统

校核&验证：仿真开发 & 测试

(5)

Continued investment in the Digital Enterprise 西门子在数字化工厂领域的持续投入

UGS acquisition establishes software foundation for

product development

Acquisition of LMS expands strategy for verification and validation of systems

Siemens establishes leadership in product and process simulation to enable digitalization

Product development

Vistagy

2007 2015 2016 2017

UGS

Perfect Costing Solutions

Kineo

LMS

TESIS

Polarion CD-adapco

Active

IBS Preactor

Camstar Orsi

Innotec

ETM

2008 2009 2010 2011

>$10 Billion since 2007

Production engineering and execution ²⁰¹⁶ ²⁰¹⁷

2013 2012

2011

2007

2012

Bentley Systems

(6)

智能网联新能源车研发解决方案校核和验证的解决方案

从纯模型、到半实物、到道路测试的解决方案

校核&验证：仿真开发 & 测试

软件在环 SiL

道路测试 Road Test 硬件在环

HiL 模型在环

MiL

车辆在环 ViL

试验场测试 Test Track

(7)

The ADAS/AD testing challenge

自动驾驶车辆开发和测试的挑战

 Dilemma:

“At what point do we sign-off on the automated functions

knowing that certain situations have not been tested?”

“ 明知无法遍历所有的交通情景，那么，究竟需要多少道路测试，

才能认为自动驾驶系统是成熟可靠的 ^?”

“14.2 billion miles of testing is needed”

Akio Toyoda, CEO of Toyota Paris Auto Show 2016

“Design validation will be a major – if not the largest – cost component”

Roland Berger, “Autonomous Driving” 2014

(8)

The ADAS/AD testing challenge

自动驾驶车辆开发和测试的挑战

 “Unlimited” number of possible scenarios to deal with 测试里程和工况场景，几乎是无限的

 System-critical situations and variant rarely happen 严苛工况及全面工况，在道路测试中可遇不可求

 Scenarios often not reproducible

即使能遇到严苛工况，也不可重现，无法重复验证

 “Ground-truth data” often unknown/inaccurate 在道路上，缺乏真值数据，来对比检验系统性能

 Scenarios often too dangerous/complicated to test 部分测试工况，具有危险性、复杂性

“14.2 billion miles of testing is needed”

Akio Toyoda, CEO of Toyota Paris Auto Show 2016

“Design validation will be a major – if not the largest – cost component”

Roland Berger, “Autonomous Driving” 2014

(9)

Siemens AD V&V Solution: Model Based Design & Validation

西门子自动驾驶测试解决方案: 基于模型的设计和测试验证

SiL

Proving ground Road test Concepts

MiL

Certification

校核&验证：仿真开发 & 测试

Vehicle in the loop ViL

System in Loop HiL

(10)

PreScan tool chain, state-of-the-art ADAS and AD simulation

核心工具-PreScan, 全球领先的ADAS和自动驾驶仿真技术

 PreScan is physics-based

simulation platform for development of Advanced Driver Assistance Systems (ADAS), that are based on sensor technologies such as radar, laser/lidar, camera and GPS. vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication applications

 PreScan是以物理模型为基础的汽车ADAS和自动驾驶仿真软件，支持基于摄像头、雷达、激光雷达、GPS、

V2V车车通讯等多种ADAS和自动驾驶的开发和测试。

(11)

ADAS & Automated driving systems in PreScan

高级驾驶辅助和自动驾驶系统在PreScan中的仿真实现

Scenarios

1

场景工况 ^Algorithms

3

算法 Sensors

2

传感器 ^Actuators

4

执行器

Traffic element database

134 scenario demos

Importing

 OpenDrive HAD Map

 IBEO lidar data

 GIDAS data

 CIDAS data

19 sensor types

 Camera

 Radar

 Lidar

 Ultrasonic

 V2X

 GPS

 …

Vehicle dynamics model

 PreScan model

 AMESim

Matlab Simulink

(12)

Simcenter Amesim

高级驾驶辅助和自动驾驶系统在PreScan中的仿真实现

自动驾驶

车辆动力学

(13)

高级驾驶辅助和自动驾驶系统在PreScan中的仿真实现

自动驾驶

车辆动力学

(14)

高级驾驶辅助和自动驾驶系统在PreScan中的仿真实现

自动驾驶系统、动力学、乘员模型的混合仿真方案

自动驾驶车辆动力学

乘员模型

轮胎模型

(15)

15

校核&验证

Design & validation methodology in model based simulation

开发和测试方法论：基于模型的仿真方法

MiL

(16)

Example 1: Massive simulation with PreScan cluster for an European OEM

案例1：大规模工况集群计算仿真方案 (某北欧OEM)

Massive Scenarios Inputs(thousands?) 大规模场景工况输入(成千上万种工况)

Regulation scenarios ISO

NHTSA NCAP

Accident database GIDAS

CIDAS

Recorded scenarios Road test data IBEO scanned

Critical scenarios OEM database Supplier database

…

Type, amount and representative of scenarios matters for verification and validation 测试工况的种类、数量、覆盖度对系统和算法的测试至关重要。

Scenarios

1

3

算法 Sensors

2

4 执行器

(17)

案例1：大规模工况集群计算仿真方案 (某北欧OEM)

PreScan Cluster solutions 大规模集群计算解决方案

Use massive scenarios and test automation on cluster to validate systems in an effective way 使用大规模测试工况库，利用集群计算、自动化测试功能，高效验证系统和算法。

Scenarios

1

3

算法 Sensors

2

4 执行器

(18)

案例1：大规模工况集群计算仿真方案 (某北欧OEM)

PreScan makes it possible to

import standard GIDAS database into virtual scenarios

导入GIDAS事故数据库数据，自动生成虚拟测试场景，用于仿真测试工况

19

(19)

Test automation enables massive scenarios

variants to improve the coverage

利用自动化测试功能，将单一场景衍生为大量的变种，提高测试覆盖度。

Example 2: Massive simulation for ACC-Adaptive Cruise Control

案例2： ACC自适应巡航大规模工况自动化仿真测试

115 kph

………

35 kph 30 kph

ACC cut-in工况

120 kph

………

45 kph 40 kph

Cut-in = 5/6/7/8/9/10 m

………

Overlap = 10%/20%/30%/40%/50% ………

Deceleration = -0.1/-0.2/-0.3/-0.4/-0.5 g

………

主车速度 20 种

目标车速度 20种

Cutin时机 10种

重叠度 10种

目标车减速度 5种

目标车类型 3种

……

工况组合

600,000

(20)

Example 3: ACC-Adaptive Cruise Control comfort analysis

案例3：ACC自适应巡航系统舒适性分析

How to determine the

parameters of each component - radar, ACC algorithm,

actuation, and chassis and tire dynamics, to get optimized

acc./dec. and jerk performance?

问题：如何分配传感器/控制算法/执行机构/底盘各部分的响应和延迟等参数特性，来实现加减速过程中的最优舒适性(jerk)？

Radar 雷达 ACC控制器和算法

Actuation 执行器底盘/轮胎 Chassis/tire

(21)

PreScan SiL Example 仿真案例

Radar ACC and TJA ：雷达ACC(自适应巡航)和TJA(堵车辅助)仿真

22

(22)

Example 4: LDW/FCW/TJA camera HiL – hardware-in-the-loop (Magna)

案例4: LDW/FCW/TJA硬件在环HiL台架 (Magna)

(23)

Example 5: LKA camera HiL – hardware-in-the-loop (Japanese OEM/tier)

案例5: 车道保持辅助硬件在环HiL台架 (日系OEM/供应商)

Real-Time PC PC

Vehicle Dynamics

Power steering model

Camera model

Video

Steering torque  x,y,θ

 x,y,θ

Driver model World Scenarios

LKA Camera

Vehicle speed ↑ Yaw rate

(24)

Q: how to fine tune Lane Keeping performance under various road

conditions? Obviously HiL is more efficient than road tests considering the amount of lane and lane marker combinations

问题：如何充分测试LKA系统在海量的车道工况下的最优性能呢？毫无疑问，

HiL测试效率远超道路测试。

海量LKA工况组合不同宽度车道

不同转弯半径各种车道组合多种标线形态

……

高效仿真测试

Example 5: LKA camera HiL – hardware-in-the-loop (Japanese OEM/tier)

案例5: 车道保持辅助硬件在环HiL台架 (日系OEM/供应商)

(25)

PreScan LDW/LKA Example 仿真案例

摄像头LDW/LKA(车道保持辅助)在PreScan中的仿真

26

(26)

Example 6: PreScan synthetic data for central AD processing units

案例6: PreScan仿真数据用于自动驾驶中央处理器(Mentor DRS360)

PreScan - Virtual sensor data and ability to generate infinite challenging traffic scenarios, to test DRS 360 central AD processor

PreScan – 仿真无限数量的严苛交通工况，通过环境感知传感器，提供感知数据，用于测试DRS 360 中央处理器和算法

Simcenter PreScan

Virtual Sensor Image Generation

(LiDAR, Camera)

(27)

Perspective Update Auto Dynamic State

Actuator Control Virtual-HiL Control Platform

Sensor Signals

3D real-time SLAM Validation

Powertrain Model

Chassis Model

Real world Scenario

Simulation

Example 6: PreScan synthetic data for central AD processing units

案例6: PreScan仿真数据用于自动驾驶中央处理器

(28)

Example 8: Radar ViL(vehicle-in-the-loop) (PATAC)

案例8: 雷达ACC/AEB车辆在环ViL试验台 (泛亚)

Real Vehicle PreScan PC @ 100Hz

World scenarios

Intervention acc./dec. request

Radar model

Target information

Target detection

Radar

(radar antenna is bypassed so only ACC&AEB logic is tested)

Echo target detection  Vehicle velocity

Yaw rate Steering angle

Vehicle Chassis/Brake/Engine actuation response to ACC/AEB intervention request

 Vehicle motion message

ACC HMI CAN

(29)

Example 8: Radar ViL(vehicle-in-the-loop) (PATAC)

案例8: 雷达ACC/AEB车辆在环ViL试验台 (泛亚)

Q: how to have a tool to solve the problems encountered in road testing: uncontrolled scenarios, none-reproducibility, no test automation, and limit ground truth info

问题：如何克服ACC/AEB路试的诸多不利因素，使得：场景受控，测试可重复，批量化测试，以及提供道路交通信息的真值

海量ACC/AEB工况组合直道

弯道

不同转弯半径静止目标移动目标不同测试车速

(30)

Example 9: DiL(driver-in-the-loop)

案例9: 驾驶员在环DiL试验台

具备实时能力的

“数字化双胞胎”

有机结合了仿真和物理测试

Digital Twin

1D Multi-Physics

3D Multi-Body Co-simulation

PreScan

Real-Time “Digital Twin”

Bridging between Virtual and Physical

(31)

Euro NCAP accredit LDW&AEB Test Track

Euro NCAP 组织官方授权LDW&AEB试验场

TASS test track is accredit Euro NCAP test facility for new AEB/LDW/ISA testing protocols.

TASS的AEB/LDW/ISA试验场是首批获得Euro NCAP主动安全系统试验资质的试验场。

(32)

Dome camera (safety) Fixed camera (vehicle localization)

Communication unit (ITS G5)

Real-life validation of Connected Systems on a network-level Public Roads A270 between Helmond and Eindhoven operated by TASS

使用A270高速Helmond和埃因霍温段建立的自动驾驶测试场，长达7公里。由TASS运营，支持智能驾驶、

自动驾驶项目的道路测试

High way test facilities for automated driving

提供高速公路自动驾驶测试场 – 荷兰A270高速公路

(33)

Example 9: America Center for Mobility powered by PreScan

案例9：Willow Run ACM 使用PreScan进行测试场设计支持

(34)

35

校核&验证

Design & validation methodology in model based simulation

开发和测试方法论：基于模型的仿真方法

MiL

(35)

World scenarios rendering

场景渲染能力

(36)

Physics Based camera model

摄像头物理模型

Simulation from scene (light source), object material, camera optics to imager and AD.

仿真从光源、传播媒介、目标表面材质、光学镜头、成像器件、以及AD转换等各个成像的物理环节

(37)

Physics Based camera model

摄像头物理模型

(38)

Physics based radar model

基于物理的雷达模型

1, Object noise and resolution 雷达的精度、分辨率特性模型 2, Object level with energy

目标的能量反射模型

3, Probabilistic radar model

雷达性能缺陷的物理模型：漏检测/误检测/分列..

∆T

(39)

基于物理的雷达模型

• 检测精度特性

• 分辨率特性

• 误检测 – False Positives(‘Clutter’)

• 漏检测 – Fales Negatives(‘Missed Detections’)

(40)

7-6-2018

基于物理的雷达模型

(41)

Physics based lidar model

基于物理的激光雷达模型

42

(42)

Siemens AD V&V Solution: Model Based Design & Validation

西门子自动驾驶测试解决方案: 基于模型的设计和测试验证

SiL

Proving ground Road test Concepts

MiL

Certification

校核&验证：仿真开发 & 测试

Vehicle in the loop ViL

System in Loop HiL

(43)

Thank you.

siemens.com

基于模型的