Trend: 99.9% of tests may utilize simulation platforms in the future
Scenario library-based simulation testing is an important route to solving the lack of data for autonomous driving road tests. Simulation testing is mainly through the construction of a virtual scene library, to realize the closed-loop simulation test of automatic driving perception, decision planning, control and other algorithms, to meet the requirements of the automatic driving test. The scene library is the foundation of automatic driving simulation test, the higher the coverage of the scene library to the real world, the more realistic the simulation test results. Different stages of the development of self-driving cars have different requirements for scene libraries, which are needed to realize different test functions. In the development process of autonomous driving, the development process of pure model simulation-software-in-the-loop simulation-semi-physical simulation-closed-course road testing-open-road testing is the most economical and efficient development process.'Waymo self-driving test vehicle'
At present, self-driving simulation has been widely accepted by the industry. For example, Carcraft, the simulation platform of Waymo, a leading U.S. self-driving company, drives about 20 million miles on virtual roads every day, which is equivalent to driving in the real world for 10 years. As of May 2020, Waymo has simulated 15 billion miles, compared to 10 billion miles last June. In addition to Waymo, domestic and international automated driving solution providers such as GM's Cruise, AutoX, and Ponywise are also conducting a large number of simulation tests to perfect their own automated driving systems, and simulation testing has become the most important test for commercial automated driving. The current data shows that about 90% of the automatic driving algorithm testing is done through simulation platforms, 9% is done in the test field, and 1% is done through actual road tests. With the improvement of the level of simulation technology and the popularization of the application, the industry hopes that 99.9% of the test volume will be completed through the simulation platform, 0.09% will be completed in the closed field test, and the last 0.01% will be completed on the actual road. In this way, self-driving car research and development will reach a more efficient and economical state.『Autonomous Driving Simulation Test Scene』
Status quo: each track participant is actively laying out
At present, the main players of the autonomous driving simulation market mainly include: technology companies, vehicle enterprises, autonomous driving solution providers, simulation software enterprises, universities and research institutions, and intelligent network testing demonstration zones. Due to the different technical basis of each market subject in the automatic driving simulation, the research and development and cooperation in promoting the automatic driving simulation show different modes. Technology companies started relatively late in simulation and have less experience in exploring automotive functions, but have the advantage of big data and strong software development capabilities. Self-driving cars compared to traditional cars, the demand for software is greater, technology companies to explore simulation software, the purpose is to enter the automotive industry with a huge market, to establish a larger data platform, to form a new business growth point. At present, the main self-driving simulation technology companies include Tencent, Baidu, Huawei, Ali and so on.『Technology companies automatic driving simulation platform comparison』
Microsoft, NVIDIA and LG and other foreign technology companies are mainly for automatic driving simulation software research and development, and through the cooperation with industry chain enterprises to establish an automatic driving research and development ecosystem, become an important participant in automatic driving simulation. For vehicle manufacturers, synchronizing road testing and simulation testing is the best choice, and self-driving cars need to go through many functional and safety tests before they can really hit the ground, and road testing is one of them. Due to the low efficiency of road testing, many car companies are now inclined to choose the combination of self-driving simulation testing and road testing to complete the testing work before landing.'Vehicle enterprises utilize automated driving simulation software'
Automated driving solution providers mainly research and develop customized simulation software for their own needs, and seldom provide simulation services to the outside world. However, with sufficient funds, talent concentration and their own R & D drive, they have a strong innovation ability in automated driving simulation. The leading automated driving solution providers have their own simulation test software, such as Waymo, Cruise, Pony Intelligence, AutoX, and so on. Simulation software companies can be divided into two categories: traditional simulation software companies and startups. Traditional simulation software enterprises due to the deeper accumulation of technology, into the automatic driving simulation has an inherent advantage, and more partners, secondary development has an advantage. Startups due to the late start, technology accumulation is weak, the gap between domestic enterprises and foreign countries is large, but relying on strong capital and talent concentration, startups in the automatic driving simulation software research and development is expected to rise rapidly. Universities and research institutes mainly use automated driving simulation software to conduct forward-looking, basic research, but it is difficult to form a mature commercialized product. Domestic universities and research institutions engaged in automatic driving simulation research mainly include: Tsinghua University, Tongji University, Beijing University of Aeronautics and Astronautics, Jilin University, Tianjin University, Chang'an University, Nanjing University of Aeronautics and Astronautics, Wuhan University of Science and Technology and so on.『Shanghai Lingang Integrated Test Demonstration Zone for Intelligent Networked Vehicles』
The construction of intelligent networked test demonstration zones has formed a certain scale. At present, there are more than 10 national and several provincial-level intelligent networked test demonstration zones, mainly through the deployment and application of new technologies such as 5G, V2X vehicle-road cooperation, simulation, and Telematics, etc., to provide systematic test services for autonomous driving and networked communication providers, and to promote the establishment of a comprehensive standard system for automobiles, information and communication, and road facilities. In order to promote the simulation test work of intelligent networked vehicles, there are already enterprises and intelligent networked demonstration zones carrying out road tests combined with virtual simulation.Challenge: test evaluation system lacks specification
At present, the simulation test of automatic driving has initially formed a complete industrial chain system, forming the upstream simulation software providers mainly by science and technology companies, automatic driving solution providers and simulation software enterprises, and the downstream application providers of simulation software mainly by vehicle enterprises and automatic driving test organizations. Analyzing from the industrial chain point of view, there are still many problems in the current automatic driving simulation test. On the one hand, the construction of simulation scene library and cooperation mechanism needs to be improved. The scene library construction is inefficient and expensive. At present, the construction of the scene library also needs to rely on a large number of artificial collection, labeling, and then scene analysis and mining, testing and verification, the whole process is inefficient and costly, and at present, the global annual cost of artificial labeling is in the order of 1 billion U.S. dollars.'Domestic *** to build automatic driving simulation test scene library still need to overcome the difficulties'
The scene library is not large enough, diversity, coverage, scalability is not strong. The existing scenario library is not large enough to cover common traffic scenarios, and cannot effectively cover the diversity of the real world with limited resource input. As different elements of the scene can be expanded to different scenes, the current scene scalability is not enough to meet the requirements of simulation testing. The effectiveness of the scenarios needs to be improved. The existing scenarios are based on real-time data collection, and cannot meet the requirements of dynamic changes in autonomous driving scenarios. In the scene, people, cars, roads, driving environment and other dynamic and static elements coupled, a change in one element will cause changes in other elements, and different traffic participants have their own behavioral logic. For example, if you change the vehicle behavior and trajectory, the behavior of the surrounding vehicles and pedestrians will also change. Scene data collection format and storage issues. Existing test scene acquisition, is based on different vehicles and sensor configuration, can not be applied to all kinds of models and technical routes of research and development and testing, the format of high-precision maps is also the focus of industry attention. The data format of the scene library such as system architecture, data interface, database management system and other unified is also a key concern. The test true value and evaluation system in the test scenarios. Test scene data collection needs to consider the collection requirements, collection methods, data pre-processing, data transmission and storage, the number of collection, collection accuracy, time synchronization, collection range, collection integrity and other factors, the lack of any one of the factors will lead to the scene of the authenticity and effectiveness. Moreover, the evaluation index system for automatic driving test vehicles under different scenarios is still not perfect.Lack of cooperation in the establishment of scene libraries and duplication of resources. At present, it is difficult for a single enterprise to complete the construction of scene libraries covering all scenarios. At present, the scene library construction of each enterprise is their own, resulting in duplicative investment of resources, and the lack of scene library construction cooperation between enterprises. In particular, natural driving scenarios, standard regulatory scenarios, and other ****ual scenarios can be built through cooperation **** to achieve the use of **** enjoyment, and there is currently very little cooperation in this regard.
『Tencent TAD?Sim part of the scene show』
On the other hand, the automatic driving simulation test evaluation system lacks standardization.On the other hand, there is a lack of standardization in the evaluation system of automatic driving simulation test, due to the different simulation software system architectures and scenario library construction methods, which makes it difficult to establish a unified and standardized evaluation system of simulation test. At present, the research direction of the domestic simulation evaluation system is mainly to evaluate the simulation test results of automatic driving vehicles from the aspects of driving safety, comfort, traffic coordination, standard matching, etc., and there is a lack of unified evaluation standards for the evaluation of the simulation software itself, such as the realism of the scene of the simulation software, the coverage of the scene, the simulation efficiency, etc.
Automatic driving as a smart driving vehicle is not a good idea.
As an intelligent product, self-driving cars need to apply deep learning algorithms in the future so that the car has self-learning capabilities, such as the ability to reproduce road obstacles, the ability to migrate to the generalization of the scene, so self-learning evolution is also an evaluation index of self-driving cars, and there is a lack of corresponding evaluation specifications for the learning evolution of self-driving cars.Summary:
There are two routes in the evolution of self-driving technology, from L2 to L3 and L4 to L5, respectively, the former is the route taken by most car companies, and the latter is often the choice of tech companies, and the main representatives of the two are Tesla and Waymo, respectively. this year, a number of companies have said they already have the ability to mass-produce L3-level self-driving vehicles; tech companies have also been launched Robotaxi commercial operation tests. It can be seen that all the competing forces are trying to be the first to apply autonomous driving technology. Who will come out on top in the race? Cost and efficiency are undoubtedly the most critical factors, and the mature application of simulation testing may become the key. (Text/Auto Home? Xiao Ying)