The role of simulation technology in the development of intelligent connected vehicles

0 Introduction

Intelligent connected vehicles refer to a new generation of vehicles equipped with advanced on-board sensors, controllers, actuators and other devices, and integrating modern communication and network technologies [1-2]. From the perspective of national industrial development and market demand, intelligent connected vehicle technology with vehicle safety as the core goal will inevitably receive more and more attention. The national intelligent connected vehicle development plan roadmap points out that by 2025, my country's partially autonomous driving and conditionally autonomous driving intelligent connected vehicles will be expected to account for 50% of the annual automobile market sales, and highly autonomous driving intelligent connected vehicles will begin to enter the market (see Figures 1-3). Based on the national intelligent connected vehicle development plan, we can draw the following conclusions.

① Intelligent connected vehicles are gradually developing from single-vehicle intelligence to the integration of intelligence and networking;

Figure 1. Milestones in the development of intelligent connected vehicles

Figure 2 Overall technology roadmap for intelligent connected vehicles (1)

Figure 3 Overall technology roadmap for intelligent connected vehicles (2)

② Autonomous driving will drive the evolution of new electronic and electrical architectures, and software-defined and data-driven cars will become the future development trend;

③ New technologies for intelligent connected vehicles will be applied in specific scenarios first. As the technology continues to be verified and mature, it will gradually expand to urban and suburban roads, highways and other scenarios;

④ In the future, roadside infrastructure will accelerate the process of intelligence, and connecting the cloud control platform with intelligent connected vehicles will form a multi-level intelligent connected transportation system.

1 Intelligent Connected Vehicle Architecture and Development Process

Different from traditional fuel vehicles and electric vehicles, the development of intelligent networked vehicles focuses more on the development of the vehicle's intelligent driving performance and control system. The core is the development of control strategies, algorithms, software, hardware, and testing in its control system. Therefore, it is very necessary to revise the development process of traditional vehicles to meet the needs of intelligent networked vehicle product development.

Different from the traditional forward development process, the V model emphasizes the control of collaboration and speed of software and hardware development. It organically combines software and hardware design and verification, so that each development activity in the software and hardware life cycle corresponds to a testing activity, and the two are carried out simultaneously, which can shorten the development cycle while ensuring high software and hardware quality (as shown in Figures 4 to 6).

Figure 4 Functional architecture of intelligent connected vehicles

Figure 5 Intelligent connected vehicle system architecture

Figure 6 V-shaped process of intelligent connected vehicle product development

Simulation tests for intelligent driving must also meet the V-shaped development process of automobiles and have the ability to test and verify the entire development process including MIL (model in the loop), SIL (software in the loop), HIL (hardware in the loop), and VIL (vehicle in the loop).

2 Importance of simulation technology

Simulation technology is a requirement for the development of intelligent connected vehicle products. Before going into production, intelligent connected vehicles need to undergo a large number of road tests and verifications to meet the requirements of product commercialization. It takes too much time and cost to use road tests to optimize the performance of autonomous driving. According to research by the Rand Corporation of the United States, if autonomous driving algorithms want to reach the level of human drivers, at least 17.7 billion kilometers of driving data are needed to improve the algorithms; road tests are restricted by traffic regulations and insurance claims mechanisms; the safety of driving in extreme scenarios and dangerous conditions is not guaranteed, and the conditions are difficult to reproduce; the traffic environment and rules of countries around the world are different, and it is difficult to form a universal industrial chain system. By around 2025, the simulation system will achieve 90% coverage of MIL scenarios and 80% coverage of HIL scenarios; it will have MIL and HIL simulation test systems that support CA-level intelligent connected vehicle testing and verification, and initially establish a real vehicle traffic environment in-loop platform. By around 2030, the V2X test conditions and test evaluation system in the simulation environment will be completed, and MIL 95% coverage of scenarios and HIL 90% coverage of scenarios will be achieved. It has MIL, HIL, and VIL simulation test systems that support HA-level intelligent connected vehicle testing and verification. The "China Autonomous Driving Simulation Blue Book 2020" points out that currently about 90% of autonomous driving algorithm testing is completed through simulation platforms, 9% is completed on test sites, and 1% is completed through actual road testing. With the improvement of simulation technology and the popularization of applications, the industry will complete 99.9% of the test volume through simulation platforms, 0.09% in closed field tests, and the last 0.01% on the actual road, which can make the development of autonomous driving vehicles more efficient and economical. Therefore, simulation technology plays a vital role in the development of intelligent driving.

3. Current status of development of intelligent driving simulation technology at home and abroad

ASAM is a non-governmental standardization organization in the automotive field. As of 2019, a total of 295 vehicle manufacturers, suppliers and research institutions from Asia, Europe and North America have joined as members. The standards launched by ASAM cover multiple automotive standard fields, including simulation, Internet of Vehicles, measurement and calibration, diagnosis, automated testing, software development, ECU network and data management and analysis. OpenX launched by ASAM includes 5 simulation test standards (as shown in Figure 7).

Figure 7 ASAM simulation format standard

In 2019, China Data (China Automotive Data Co., Ltd., a subsidiary of China Automotive Technology and Research Center) cooperated with ASAM to jointly establish the C-ASAM working group to expand the simulation scenario standard of the ASAM OpenX standard.

Figure 8 Intelligent driving simulation system framework

4 Application of intelligent network simulation technology

The complete intelligent driving simulation platform includes functions such as static scene library, dynamic case simulation, perception sensor simulation, vehicle dynamics simulation, and path planning decision algorithm verification. In general, the core algorithm of autonomous driving includes three major links: perception fusion algorithm, decision planning algorithm, and control algorithm. Correspondingly, the intelligent driving simulation test platform should also have the ability to complete the simulation test of the above three algorithms. The perception fusion algorithm simulation requires a high-fidelity 3D reconstruction scene and an accurate sensor model; the simulation of the decision planning algorithm requires a large number of scene libraries as support; the simulation of the control algorithm requires the introduction of an accurate vehicle dynamics model. Virtual scene construction: The simulation test of intelligent driving cars first needs to simulate and construct a vehicle operation scene that is consistent with the real world, and the scene construction can be divided into two levels: static scene construction and dynamic scene construction. The role of static scene construction is to restore the static elements related to vehicle driving in the scene, such as roads (including materials, lane lines, speed bumps, etc.); static traffic elements (including traffic signs, street lights, stations, tunnels, surrounding buildings, etc.). The most commonly used method is to complete the scene construction based on high-precision maps and 3D reconstruction technology, or to build the scene based on augmented reality methods. In a broad sense, dynamic scene elements include dynamic environmental elements such as dynamic indicator facilities and communication environment information, as well as traffic participants (including motor vehicle behavior, non-motor vehicle behavior, pedestrian behavior, etc.), meteorological changes (weather conditions such as rain, snow, fog, etc.), time changes (mainly changes in light at different times), etc. Perception system simulation: The general method of camera simulation is to build a three-dimensional model of the object based on the geometric spatial information of the environmental object, which is to generate realistic images; millimeter wave radar simulation: Generally, according to the configured field of view and resolution information, a series of virtual continuous frequency modulation millimeter waves are emitted in different directions, and the reflected signal of the target is received. The radar echo intensity of different vehicles can be calculated by the micro-surface model energy radiation calculation method, which is calculated by the vehicle model and the vehicle orientation, material, etc.; LiDAR simulation: Referring to the scanning method of the real LiDAR, simulate the emission of each real radar ray and intersect with all objects in the scene; In the simulation test of intelligent driving cars, it is necessary to simulate the vehicle with the help of the vehicle dynamics model to objectively evaluate the decision and control algorithm. Because the complex vehicle model can ensure that the vehicle has good simulation accuracy and make the reaction of the controlled object closer to the real world. Vehicle dynamics simulation: The vehicle dynamics model is a model built based on multi-body dynamics, which includes vehicle models of multiple real components such as the body, suspension system, steering system, braking system, power system, transmission system, vehicle dynamics system, hardware IO interface, etc. After parameterizing these controlled object models, the real wire control brake, wire control steering system and intelligent driving system can be integrated into the large system for joint simulation testing. Cloud accelerated simulation: When performing simulation tasks, the simulation system needs to access a large amount of collected or generated data, and use CPU and GPU resources to reprocess and restore the data based on the generated data, or perform GPU rendering and reproduction on the structured data. This requires the use of a mechanism to distribute simulation tasks to multiple machines and let all machines work together, which can reduce the performance requirements of a single machine, thereby enabling large-scale simulation tasks to be achieved.