虚拟仿真技术是汽车研发、制造、验证测试等环节不可或缺的技术手段，能有效缩短技术和产品开发周期，降低研发成本；随着汽车智能化、网联化趋势的发展，虚拟仿真技术有了更大的发挥空间，比如自动驾驶系统的仿真测试验证；虚拟仿真测试是实现高阶自动驾驶落地应用的关键一环，具备自动驾驶功能的车辆必须经过大量的虚拟仿真测试以及实车路测之后才能商用化；
Virtual imitation is an indispensable technical tool for vehicle R&D, manufacturing, validation tests, etc., that can effectively shorten the technology and product development cycle and reduce R&D costs; virtual imitation technology has gained greater scope as automobile intelligence and networking trends have developed, such as the simulation test validation of autopilot systems; virtual imitation testing is a key part of achieving high-level autopilot landings, and vehicles with autopilot functions have to undergo extensive virtual imitation tests and live road surveys before they can be commercialized;
自动驾驶汽车商用化需经历的三个测试阶段 ：仿真测试、封闭场地测试、开放道路测试。
Three test stages for the commercialization of auto-drive cars: simulation tests, closed-space tests, open-circuit tests.
自动驾驶仿真测试： 主要是以数学建模的方式将自动驾驶的应用场景进行数字化还原，建立尽可能接近真实世界的系统模型，无需实车直接通过软件进行仿真测试便可达到对自动驾驶系统及算法的测试验证目的。
Autopilot simulation tests: digitize autopilot applications in mathematical modelling, build systems models that are as close as possible to the real world, and do not need to be tested directly through software for testing validation of autopilot systems and algorithms.
仿真测试包括以下几种类型： 模型在环仿真（MIL）— 软件在环仿真（SIL）— 硬件在环仿真（HIL）— 整车在环仿真（VIL）
Simulation tests include the following types: Models in Simulation (MIL) - Software in Simulation (SIL) - Hardware in Simulation (HIL) - Vehicle in Simulation (VIL)
实车道路测试面临的问题：
Problems with physical road testing:
据美国兰德公司研究：一套自动驾驶系统至少需要通过 110 亿英里的驾驶数据来进行系统和算法的测试验证才能达到量产的条件；因此单纯依靠实车路测极难完成这一目标，并且实车路测还存在以下问题：
According to a study by the United States company Land, a system of automated driving systems requires at least 11 billion miles of driver data to test systems and algorithms to meet the requirements for volume production; it is therefore extremely difficult to achieve this goal by relying solely on live road surveys, and there are also problems with live road surveys:
- 道路测试周期长，成本高，效率低
- 覆盖的场景工况有限，尤其是对于 “corner case”， 很难复现
- 对于一些极端的危险场景，道路测试安全性无法保障
仿真测试的优势：
Advantages of simulation tests:
- 测试场景配置灵活，场景覆盖率高；
- 测试过程安全，且对于一些“corner case”能够进行复现再测试
- 可实现自动测试和云端加速仿真测试，有利于提升测试效率和降低测试成本
备注 ：● 真实? ○ 虚拟 ?◎ 虚拟或部分真实 数据参考 ：中国汽车工程研究院
仿真测试、封闭场地测试、开放道路测试三者之间互相补充，形成测试闭环，共同促进自动驾驶车辆的研发和标准体系建立：
Simulation tests, closed site tests, open road tests complement each other, forming test closed rings, and jointly promoting auto-drive vehicle development and standard-setting:
1）仿真测试结果可以在封闭场地和开放道路进行测试验证；
1) Simulation test results can be tested at closed sites and open roads;
2）通过道路测试得出的危险场景，将会反馈到仿真测试中，便可有针对性的去调整设定场景和参数空间；
2) Hazard scenarios derived from road tests will be fed back into simulation tests and can be targeted to adjust the setting of scenes and parameter space;
3）仿真测试和封闭场地测试的最终结果要进行综合评价，基于评价结果不断地去完善评价准则和测试场景库。
3) The final results of the simulation test and the closed site test are evaluated in a comprehensive manner, and evaluation results are used as a basis for continuous refinement of the evaluation guidelines and the testing landscape.
自动驾驶仿真测试的重要构成： 场景库、仿真平台、评价体系；其中，场景库是 基础 ，仿真平台是核心 ，评价体系是 关键 ；三者紧密耦合，相互促进：场景库的建设需要仿真平台和评价体系作为指导，仿真平台的发展进化需要场景库和评价体系作为支撑，而评价体系的建立和完善也需要以现有的场景库和仿真平台作为参考基础；接下来笔者将从 场景库、仿真平台、评价体系 这三个重要方面依次展开来介绍。
An important component of autopilot simulation testing is a landscape library, a simulation platform, an evaluation system, in which the landscape library is the basis, the simulation platform is the core, and the evaluation system is the key; three are closely aligned and mutually reinforcing: the development of the landscape library requires the simulation platform and the evaluation system to guide it, and the evolution of the simulation platform needs to be underpinned by the development of the landscape repository and the evaluation system, and the establishment and refinement of the evaluation system needs to be based on the existing landscape library and the simulation platform; the next paper will be presented in turn from the three key areas of the scenario library, the simulation platform, and the evaluation system.
自动驾驶测试场景定义： 自动驾驶汽车与其行驶环境各组成要素在一段时间内的总体动态描述，要素组成由所期望检验的自动驾驶汽车的功能决定；（定义引自：中汽协团体标准-自动驾驶系统功能测试第7部分 – 仿真测试） 简言之，场景可以被视为是自动驾驶汽车行驶场合与驾驶情景的有机组合。具有场景无限丰富、极其复杂、难以预测、不可穷尽等特点。 ?
definition of autopilot test scenario: An overall dynamic description of autopilot and its constituent elements of the driving environment over a period of time, the elements of which are determined by the desired function of the autopilot vehicle; in short, the scene can be seen as an organic combination of auto-drive and driving scenarios.
测试场景要素：测试车辆自身要素以及外部环境要素 ；外部环境要素 又包括：静态环境要素、动态环境要素、交通参与者要素、气象要素等。
test scenario elements : test the vehicle's own elements as well as external environmental elements; external environmental elements, including static environmental elements, dynamic environmental elements, transport participants elements, meteorological elements, etc.
场景库定义：满足某种测试需求的一系列 自动驾驶测试场景 构成的数据库。场景库能够完成从场景数据的管理到场景测试引擎的桥接，实现从场景的自动产生、管理、存储、检索、匹配，到最后注入测试工具。
Sites Definition: is a database of a series of autopilot test scenarios that meet some test needs.
场景库包含 4 种典型测试场景（中汽中心基于数据来源不同的一种分类方法） ： 自然驾驶场景、危险工况场景、标准法规场景、参数重组场景。
The landscape contains four typical test scenarios (a classification based on different data sources by the Central Gas Centre): natural driving scenes, hazardous work scenes, standard regulatory scenes, parameter reorganization scenarios.
测试场景的数据来源主要包括三大部分：真实数据、模拟数据以及专家经验
The source of data for the test scenario consists mainly of three main parts: real data, analog data and expert experience
1）真实数据：即现实世界真实发生的，经过传感器采集到或以其它形式被记录保存下来的真实场景数据，包括自然驾驶数据、交通事故数据、路侧单元监控数据、封闭场地测试数据以及开放道路测试数据等；
1) Real data: , real scene data collected by sensors or recorded in other forms, including natural driving data, traffic accident data, side-by-side control data, closed site test data and open road test data;
2）模拟数据：主要包括驾驶模拟器数据和仿真数据；前者是利用驾驶模拟器进行测试得到的场景要素信息；后者是自动驾驶系统或车辆在虚拟仿真平台上进行测试得到的场景要素信息。
2) Simulation data: consists mainly of driving simulator data and analogue data; the former is scenario element information obtained from testing using driving simulators; the latter is scenario element information obtained from autopilot systems or vehicle testing on virtual simulation platforms.
3）专家经验数据：基于专家的仿真测试经验总结归纳出来的场景要素信息，其中标准法规就是专家经验数据的典型代表。
3) expert experience data: information on scenario elements based on expert simulation test experiences, in which standard regulations are typical of expert experience data.
备注： 测试场景数据来源参考《自动驾驶系统功能测试第7部分：仿真测试（征求意见稿）》
搭建流程：
built-up process:
- 确定单个虚拟场景的数据存储方式与标准
- 构建单个自动驾驶虚拟测试场景
- 在众多的虚拟测试场景中根据特征标签选取适当场景作为场景库的组成部分
（图片参考：中汽数据有限公司）
自动驾驶研发测试与场景库的搭建形成闭环：测试场景库的搭建，能有效驱动自动驾驶的研发测试工作，自动驾驶的研发测试反过来也能够为场景库提供反馈意见，丰富场景库。 ?
autopilot R & D testing and setting-up of a landscape:
现状分析： Status Analysis: 场景库覆盖范围：自然驾驶场景库（1228种）、功能安全场景库（110种）、V2X场景库、危险事故场景库（206种）、中国特有交通法规场景库（82种）、和预期功能安全场景库（70种）等。 Sites Coverage: Natural Drive Sites (1228), Functional Safety Sites (110), V2X Sites, Dangerous Accidents Sites (206), China's Special Traffic Regulations Sites (82), and Expected Functional Safety Sites (70). a、自然驾驶场景库 a, natural driving garage 基于中国不同道路拓扑结构、交通基础设施、环境条件、车辆信息的动静态要素特征，从不同维度对数据库进行分类构建和更新迭代： Based on the dynamic static element characteristics of China's different road topping structures, transport infrastructure, environmental conditions and vehicle information, the database is classified and updated from different dimensions: b、功能安全场景库 功能安全场景库生成过程： Function security site generation process: 第一步：参考 ODD 分类的国际标准，对驾驶区域、天气、光线等 7 大类场景要素以及 30 类子要素进行排列与重组，进而形成体系化功能场景库； First step: , by reference to the international standards of the ODD classification, organizes and reconfigures elements of major scenarios, such as driving areas, weather, light, etc., and 30 sub-elements, leading to the development of a systemized, functional landscape pool; 第二步：将上述功能场景，基于真实路采场景数据，调整相关参数范围从而得到相应的逻辑场景，最后在仿真软件中搭建驾驶场景与周边环境，形成所需的复现场景； Step 2: will adapt the range of parameters to obtain the corresponding logical scenario based on real road-seed data, and eventually build the driving scene and the surrounding environment in the simulation software to create the required re-site scene; 第三步：在仿真验证软件中，激活 ADAS 功能，对复现场景下的电子电气失效展开仿真验证，并对产生的风险进行评估，根据公式计算出严重度（S）、暴露度（E）和可控性（C），最后根据 ISO26262 标准最终确定 ASIL 风险等级。 Step 3: activates the ADAS functionality in the simulation software, carries out simulations of electrical and electronic failures in the re-site setting and evaluates the risks associated with the calculation of severity (S), exposure (E) and controlability (C) based on formulae, resulting in the final determination of the ASIL risk classification based on ISO2662. c、 V2X 场景库
- 根据 LTE-V 终端及 DSRC 终端应用层开发经验以及《合作式智能运输系统车用通信系统应用层及应用数据交互标准》中的场景要求进行场景库构建
- 从数据安全、通信安全、控制安全、身份安全等四个方面进行 V2X 场景数据积累研究
- 建立了 V2X 测试场景库创建流程，形成测试用例 3500 条 ?
场景库数据来源：标准法规、交通事故、人工经验以及自然驾驶数据。
site database data sources: standard regulations, traffic accidents, manual experience and natural driving data.
场景库生成方案：场景数据采集、场景分类提取、场景数据标注、场景聚类、场景重构、虚拟场景转化等。
scenario generation programme: scene collection, scene classification extraction, scenario data labelling, landscape clustering, scenario re-engineering, virtual landscape transformation, etc.
中国典型场景库 V2.0：
China Model Scene Bank V2.0:
- 参考德国 PEGASUS 项目的场景分类体系及 ASAM 推出的 OpenDRIVE 和 OpenSCENARIO 仿真格式
- 包括标准法规场景 1000 例、经验式场景 3000+ 例、重构场景 5 万 + 例、事故场景 150+ 例
2020 年 12 月，“中国典型场景库”在 V2.0 版本的基础上升级到 V3.0：
- 在 2.0 版本的基础上，新增了典型事故场景、自动驾驶事故场景和预期功能安全场景三大板块
- 涵盖总计 20 余项自动驾驶功能；总量大于 10 万例的 OpenX 标准格式场景文件
- 为了便于企业的自动驾驶技术开发，把场景库分为基础、初级、中级、高级四个模块等级
涵盖场景类型：
covers the type of scene:
- 法规标准场景：满足基本的标准法规要求
- 危险工况场景：保证自动驾驶的安全性
- 能力评估场景：评价自动驾驶能力
典型场景类型数量：200 种左右
Number of typical scenarios: 200 or so
涵盖：不同的道路类型、障碍物类型、道路规划、红绿灯信号
covers: different types of roads, types of barriers, road planning, red light signals
测试场景数据来源： 自然驾驶路采数据 、 交通数据库、人工经验等。
test scene data source: Natural DriveLoging Data, Traffic Database, Manual Experience, etc.
测试场景基于生成方案的不同又分为： Logsim 场景和 Worldsim 场景， 目前总共提供了 220 个 Worldsim 场景和 17个 Logsim 场景。
The test scenes, based on different generation scenarios, are divided into: Logsim and Worldsim scenes, with a total of 220 Worldsim scenes and 17 Logsim scenes currently available.
a、Logsim： 由路测数据提取的场景，提供复杂多变的障碍物行为和交通状况，场景充满不确定性
a, Logsim: scenes extracted from road survey data, providing complex and variable barrier behaviour and traffic conditions, scenes full of uncertainty
b、Worldsim： 由人为预设的障碍物行为和交通灯状态构成的场景 ， 场景简单 ?
路测实采数据积累：截止到 2020 年，已经积累了超过 50 万公里交通场景数据。 ?
涵盖场景类型 ：车辆避撞能力、交通合规性、行为能力、视距影响下交叉路口车辆冲突避免、碰撞预警、紧急制动、危险变道、无信号交叉口通行、行人横穿等方面。 covers the types of scenes: vehicle collision avoidance, traffic compliance, mobility, vehicle conflict avoidance at intersections under visual influence, collision warning, emergency brakes, dangerous variants, non-signal intersections, pedestrian crossings, etc. 典型场景类型数量：1000 种左右 Number of typical scenarios: 1000 or so 场景生成方案：通过 Agent AI 能力，可以自由生成各种随机的驾驶场景 ? scene generation program: is free to generate random driving scenes through Agent AI capabilities? 仿真平台一般包括 仿真框架 、 物理引擎 和 图形引擎 ；其中仿真框架是平台软件平台的核心，支持传感器仿真、车辆动力学仿真、通信仿真、交通环境仿真等； Simulation platforms typically include imitation frames, physical engines and graphic engines; imitation frames are at the core of the platform's software platform and support sensor simulations, vehicle dynamics simulations, communications simulations, traffic environment simulations, etc.; 1 ） 传感器仿真 ： 支持支持摄像头、激光雷达、毫米波雷达以及GPS/IMU等传感器仿真； 1) Sensor Simulation: Support to support sensors such as camera, laser radar, millimetre wave radar and GPS/IMU; 2 ）车辆动力学仿真 ：基于多体动力学搭建的模型，将包括转向、悬架、制动、I/O硬件接口等在内的多个真实部件进行参数化建模，进而 实现车辆模型运动过程中的姿态和运动学仿真模拟； 2) Vehicle kinetic Simulations: Models based on multi-body dynamics will be used for parameter modelling of multiple real components, including diversions, suspensions, braking, I/O hardware interfaces, etc., which will lead to simulations of postures and kinetic simulations during vehicle modeling campaigns; 3 ） 交通场景仿真： 包括静态场景还原和动态场景仿真两部分，静态场景还原主要通过高精地图和三维建模技术来实现； 3) Traffic scene simulation: including both static and dynamic scene simulations, which are mainly achieved through high-precision maps and 3D modelling techniques; 动态场景仿真既可通过把真实路采数据经过算法抽取后，再结合已有高精地图进行创建，也可通过对随机生成的交通流基于统计学的比例，经过人工设置相关参数后自动生成复杂的交通环境; Dynamic scenario simulations can be created either by extracting the true routing data from algorithms and combining existing high-precision maps, or by automatically creating a complex traffic environment by artificially setting the relevant parameters through statistically based ratios of randomly generated traffic flows; 例如可模拟自动驾驶汽车在现实世界中可能遇到的极端情况和危险情况，从模拟暴雨和暴雪等恶劣的天气条件到较弱的光线照明，再到周围车辆的危险操作等； For example, extremes and dangerous situations that autopilots may encounter in the real world can be simulated, ranging from harsh weather conditions such as simulating storms and snowstorms to weak lighting of light to dangerous operations of surrounding vehicles; 4 ）V2X仿真（通信仿真）： 支持创建真实或虚拟传感器插件，使用户能够创建特殊的V2X传感器；既可以用来测试V2X系统，又可生成用于训练的合成数据； 4) V2X emulation (communication simulation): supports the creation of real or virtual sensor plugins to enable users to create special V2X sensors; can be used both to test V2X systems and to generate synthetic data for training; 1）浙江天行健智能科技 – Panosim ? 1) Zhejiang Skyline Intelligent Technology - Panosim? 平台类型： 面向汽车自动驾驶 技术与产品 研发的 一体化 仿真与测试平台 ? Platform type: Integration of autopilot technology and product development Simulation and testing platform? 平台特点： Features of the platform: 2 ）51WORLD – 51Sim-One ? 平台类型： 覆盖自动驾驶全流程的一体化集成的仿真测试平台 Type of platform: Simulation test platform that covers the integrated integration of autopilots 平台特点： Features of the platform: 3 ）腾讯 – TAD Sim ? {\bord0\shad0\alphaH3D}TAD Sim? 平台类型： 基于虚幻引擎打造的虚实结合、线上线下一体化的仿真测试平台 Type of platform: Simultaneous testing platform for virtual integration, off-line integration based on imaginary engines 平台特点： Features of the platform: 4) ? 沛岱（上海）技术有限公司– Pilot-D GaiA ? 4 > — Pilot-D Gaia (Shanghai) Technology Co. Ltd. 平台类型： 基于德国自动驾驶仿真核心技术所研发的仿真测试平台 Type of platform: Simulation test platform developed on the basis of German autopilot imitation core technology 平台特点： Features of the platform: 1 ）西门子 – PreScan ? {\bord0\shad0\alphaH3D}PreScan? 类型： 以物理模型为基础的传统汽车仿真平台 Type: Traditional car simulation platform based on physical models 平台特点： Features of the platform: 2 ）MSC Software ? – VTD ? 类型： 传统汽车仿真平台 Type: Traditional Automobile Simulation Platform 平台特点： Features of the platform: ? ? ? ? ? —— 提供图形化的交互式路网编辑器 Road Network Editor (ROD)，? 在构建路网仿真环境的时候，可以同步生成OpenDrive高精地图 ♪ ♪ ♪ -- provide a graphical interactive network editor, RoadNetwork Editor (ROD) ♪ The OpenDrive high-precision map can be created simultaneously in the construction of the network simulation environment ? ? ? ? ? —— ? 对于动态场景构建， 提供了图形化的交互式场景编辑器 Scenario? ? ? Editor ♪ ♪ ♪ for dynamic scenario construction, provides a graphical interactive scene editor Scenario ♪ ♪ Editor ? 3 ）德国IPG – CarMaker ? Germany IPG – CarMaker? 类型： 以传统动力学仿真为基础优势发展起来的自动驾驶 仿真平台 Type: Autopilot based on traditional kinetic simulations, imitation platforms 平台特点： Features of the platform: ? ? ? ? —— IPG Road：可以模拟多车道、 十字路口等多种形式的道路， 并可通过配置 GUI生成锥形、 圆柱形等形式的路障 ♪ ♪ ♪ -- IPG Road: Multiple lanes, intersections, etc. can be simulated, and roadblocks can be created by configuration of GUI in the form of conicals, cylindricals, etc. ? ? ? ? —— IPG Traffic：提供丰富的交通对象模型，如车辆、 行人、 路标、交通灯、 道路施工建筑等 ♪ ♪ ♪ ♪ ♪ IPG Trafic: Provide a rich model of traffic objects, such as vehicles, pedestrians, signs, traffic lights, road construction, etc. ? ? ? ? —— IPG Driver：提供可自学习的驾驶员模型 ♪ ♪ ♪ ♪ ♪ ♪ ♪ IPG Driver: Provide self-learning driver models ? ? ? ? —— 支持从HERE HD Live Maps导入地图数据 ? ? ? ? ♪ ♪ support the import of map data from Here HD Live Maps ♪ ? ? ? ? —— 支持ROAD5和OpenDrive 格式导出地图数据 ♪ ♪ ♪ ♪ -- supporting the export of map data in ROAD5 and OpenDrive formats 4 ）英伟达- ? Drive Constellation ? {\bord0\shad0\alphaH3D}Drive Construction {\cHFFFFFF}{\cH00FFFF} {\cHFFFFFF}{\cH00FFFF} {\cHFFFFFF}{\cH00FFFF} {\cHFFFFFF}{\cH00FFFFFF} {\cHFFFFFF}{\cH00FFFF} 类型 ：基于虚幻引擎开发，由两台服务器构成的自动驾驶仿真平台 Type: Autopilot imitation platform based on false engine development consisting of two servers 平台构成： Composition of the platform: a、第一台服务器硬件构成：由8个英伟达RTX Turing GPU a. First server hardware: made up of 8 British Wedger RTX Turing GPU 作用： 运行DRIVE Sim软件来模拟仿真自动驾驶车辆上的传感器数据（包括摄像头、毫米波雷达、激光雷达、IMU和GNSS）以及驾驶场景数据； Activation: Run DRIVE Sim software to simulate sensor data (including camera, millimetre wave radar, laser radar, IMU and GNSS) on autopilot vehicles, as well as driving scene data; b、第二台服务器硬件构成：自动驾驶车辆目标AI ECU b. Hardware composition of the second server: Autopilot target AI ECU 作用： 用于处理第一台服务器传输过来的模拟数据，如传感器仿真数据 Activation: Simulation data used to process first-server transfers, e.g. sensor simulation data 平台特点： Features of the platform: 5 ）微软 – AirSim ? {\bord0\shad0\alphaH3D}Microsoft? {\bord0\shad0\alphaH3D}AirSim? 类型： 建立在虚幻引擎（Unreal Engine）上的无人机及自动驾驶开源仿真平台 Type: Uranium-based drones and autopilot open-source simulation platforms on a false engine (Unreal Engineering) 平台特点： Features of the platform: 6 ）巴塞罗那自治大学（联合丰田研究院和英特尔实验室） – CARLA ? 类型 ：基于虚幻引擎开发，采用服务器和多客户端架构的开源平台 Type: An open-source platform using server and multi-client structures based on the development of a vanity engine 平台特点 ： Platform characteristics: 7 ）LG电子 - LGSVLSimulator ? LG Electronic - LGVL Simulator ? 类型： 基于游戏引擎-Unity研发的自动驾驶开源仿真平台 ? Type: Autopilot open source simulation platform based on game engine-Unity? 平台特点： Features of the platform: —— 提供 Python API，使用户能够通过编写脚本来创建所需的虚拟场景 _Providing Python API to enable users to create the virtual scene required by writing scripts —— 高精地图： 支持创建、编辑和导入/导出现有 3D 环境的高清地图 - high-precision maps: support the creation, editing and import/guiding of high-resolution maps with 3D environments 注：高精地图支持的导入/导出格式： a ．支持的导入格式： Apollo5.0 高清地图、Autoware Vector map、Lanelet2和OpenDrive 1.4 b. ? 支持的导出格式： Apollo 5.0 、Lanelet2和OpenDrive 1.4 传统的仿真测试一般是单机仿真测试，但高阶自动驾驶系统的商业化应用，需要进行庞大数量测试场景的仿真测试；海量的数据存储，运算及处理已经成为摆在自动驾驶研发道路上一道障碍，而云平台的 分布式架构 、 加速计算能力 以及 高数据算力 的特性，能够可以很好的解决这一问题，进而能够极大地提升系统研发和验证测试效率；因此，云平台仿真必将是未来自动驾驶仿真测试技术的发展趋势，同时也会是企业开展自动驾驶开发的核心技术壁垒之一。 Traditional simulation tests are typically single machine simulation tests, but the commercialization of high-level autopilot systems requires a large number of simulation tests of the test scenario; big-volume data storage, operation and processing have become an obstacle on the path of autopilot R&D, while cloud platforms’ distributional structures, accelerated computing capabilities and high data numeracy features can best address this problem, which in turn can greatly enhance the efficiency of system R&D and validation testing; therefore, cloud platform simulation is bound to be one of the core technical barriers to the development of autopilot testing technologies in the future, as well as to the development of automated driving by firms. ? 1) ? 百度 – Apollo云仿真平台 ? ? > 100 degrees — Apollo Cloud Simulation Platform 主要应用： Main applications: 平台特点： Features of the platform: 2 ）华为 – 自动驾驶云服务Octopus ? 2) - Auto-driving Cloud Service Octopus? 平台提供三大服务： 数据服务、训练服务、仿真服务；服务覆盖自动驾驶 数据、模型、训练、仿真、标注 等全生命周期业务； The Platform provides three main services: data services, training services, imitation services; services covering all life-cycle operations such as autopilot data, models, training, simulations, labelling, etc.; 平台特点： Features of the platform: 3) ? 腾讯 – TAD Sim Cloud ? - TAD Sim Cloud? 平台特点： Features of the platform: 4）阿里 – 混合式仿真测试平台 4) - Mixed Simulation Test Platform 平台特点： Features of the platform: 1 ）两个重要评价维度： 真实性和有效性 1) Two important evaluation dimensions: authenticity and validity a、真实性评价 ：主要是针对场景库真实合理性的评价，分为场景信息真实度、场景分布真实度两个方面 a. Evaluation of authenticity: evaluation of the true legitimacy of the site library, in terms of both the trueness of the information on the scene and the true distribution of the scene 场景信息真实度：在场景在构建过程中，需要合理准确的在虚拟环境中渲染测试场景中的静态环境要素（如交通设施、道路及障碍物等）、动态环境要素（动态指示设施、通信环境信息）、交通参与者要素（机动车/非机动车、行人等）以及气象环境要素（环境温度、光照条件、天气情况）等信息。 The trueness of scene information: During the construction of the scene, reasonably accurate static environmental elements (e.g. transport facilities, roads and barriers), dynamic environmental elements (dynamic indicator facilities, communication environmental information), transport participants elements (motorized/non-motorized vehicles, pedestrians, etc.) and meteorological environmental elements (environmental temperature, light conditions, weather). 场景分布真实度：在参数重组场景中对于由特征元素组合和人工编辑合成的场景，由于人工修改参数后可能会出现真实世界不存在的场景，因此在人工编辑场景时需参考真实世界场景的参数值范围去合理化地设置参数重组场景。 The trueness of the scenario distribution: In the case of a feature element combination and artificially edited synthesis in the context of a parameter reorganization scenario, it is necessary to rationalize the parameter reorganization scenario by reference to the range of parameters of the real world scenario in the case of an artificially modified scenario that may not exist in the real world. b、有效性评价 b. Evaluation of effectiveness ?目前尚未统一的有效性评价标准 ， 当前可参考的标准如下： ? The effectiveness evaluation criteria are not yet harmonized, and the following criteria are currently available for reference: 2 ）具体评价内容 ：仿真测试自身评价以及自动驾驶车辆驾驶性能、驾驶协调性、标准匹配性、学习进化性等方面； 2) Specific evaluation: Simulation of self evaluation and autopilot driving performance, driver coordination, standard matching, learning evolution, etc.; 注：评价项目内容参考《2020中国自动驾驶仿真蓝皮书》 ? Note: The contents of the project are evaluated by reference to the China Autopilot Simultaneous Blue Book 2020? 3.2 ? 仿真测试评价标准示例 ? 3.2? Example of imitation test evaluation criteria? 评价方法 ：（两种） Evaluation methodology: (two) 评价指标：（两类） Evaluation indicators: (two categories) 参考：评价标准参考《自动驾驶系统功能测试第7部分：仿真测试（征求意见稿）》 ? Reference: Evaluation criteria reference, Part 7: Simulation Test (Consultation Draft) 1 ）场景库建设 1) Site library 2 ）仿真平台 < strong>2) Simulation platform 3 ）测试评价体系 3) Test evaluation system 1）基于云平台的高并发测试、加速测试是仿真测试未来重要的发展方向，是自动驾驶仿真企业的核心竞争力之一；云平台具备海量数据存储、处理和管理的能力，支持大规模仿真构建，覆盖海量驾驶场景；可有效建立城市级地理信息+虚拟交通流信息，可实现跨专业多用户信息交互与共享。 ? 1) High-synchronous testing, accelerated testing of cloud-based platforms is one of the key future directions of simulation and one of the core competitiveness of autopilot simulation enterprises; cloud platforms have the capacity to store, process and manage big-volume data, supporting large-scale simulations and covering large-scale driving scenes; and can effectively create city-level geo-information + virtual traffic flow information that allows multi-professional multi-user information to interact and share? 2）功能安全场景库、 V2X场景库以及预期功能安全场景库将是未来场景库建设的重点；不同厂商 采用统一数据格式标准，共建基础场景库，形成通用的、可移植的场景库。 ? 2) The functional security site library, the V2X site library and the expected functional security site library will be the focus of future site development; different manufacturers will build a common, portable site repository using uniform data format standards.? 3）混合交通仿真测试将是未来自动驾驶仿真技术的一个重要研究领域；自动驾驶汽车取代传统汽车不是一蹴而就的事情，必然在相当的一段时间内处于两者共存的局面，因此传统汽车与自动驾驶汽车混行将是一种常见的交通场景。 甚至还需要建立自动驾驶车辆在不同渗透率下的混合交通仿真模型。 ? 3) The hybrid traffic simulation test will be an important area of study for the future autopilot simulation technology; autopiloting instead of traditional cars is not a one-off event, and is bound to coexist for a considerable period of time, so that the mix of conventional and auto-driving cars will be a common traffic scene. Even a hybrid model of auto-driving vehicles at different penetration rates is needed. 4）数字孪生技术将为自动驾驶仿真测试发展道路上的一大助力，是仿真测试的增效利器； ? 数字孪生就是在虚拟环境中建立一个和现实世界一致的模型。自动驾驶数字孪生测试VRIL(Virtual Reality in the Loop)即真实的车辆在真实的测试场地中进行测试，将同时映射到虚拟的测试环境中，从而完成虚、实状态的同步，实现整个数字孪生系统的闭环实时仿真测试。 ? 4. Digital twinling technology will be a major contribution to the development of autopilot simulations on the road, and is a simulator of simulations; Digital twining is a model that is consistent with the real world in a virtual environment. Autopilot digital twin testing is a real-time simulation of VRIL, the real-life vehicle, in a real-life test site, that will map both into a virtual test environment, thus completing the virtual synchronization of the state, and achieving a real-time simulation of the closed loops of the entire digital twin system. 5）当前自动驾驶仿真测试主要还是偏向于传统的主被动层面的安全性及可靠性等方面的测试验证， 以及用于感知、规划决策及控制算法的研发或迭代升级；但是每个自动驾驶车辆都不是孤立存在的，必然是网联化的，是作为整个智慧交通体系中的一个智能节点而存在，因此信息安全方向的验证测试与评价也必然是未来自动驾驶仿真测试的一个重要研究方向。 (5) The current testing of autopilot simulations is mainly in favour of tests of safety and reliability at the traditional primary passive level, as well as research and development or iterative upgrades for perception, planning decision-making and control algorithms; however, each autopilot is not isolated, necessarily networked, and exists as an intelligent node in the entire intelligent transport system, so the testing and evaluation of the direction of information security must also be an important research direction for future autopilot simulation tests.
注： √- 有此功能 ×-无此功能?— Unknown
注： √- 有此功能 ×-无此功能?— Unknown
注册有任何问题请添加 微信：MVIP619 拉你进入群
打开微信扫一扫
添加客服
进入交流群
发表评论