With the continuous improvement of the level of autonomous driving functions put into the market, how to ensure the safe driving of cars in various complex scenarios is a problem that many technology companies, car companies, and autonomous driving solution providers are currently facing.
For complex road networks and real scenarios, the time and cost of using traditional road tests to optimize automatic driving algorithms are too high, and open road tests are still restricted by regulations, extreme traffic conditions and scene reproduction are difficult, and there are certain hidden dangers in test safety. Based on the above reasons, large-scale simulation testing in the virtual environment has basically become the "consensus" of the industry, and virtual simulation testing seems to have become a "new highland" for competition in the autonomous driving industry.
The importance of simulation testing
According to the world-renowned consulting firm RAND, autonomous vehicles need to go through 150 million miles of testing to reach a level of safety comparable to that of human drivers, of which 95% use virtual simulation and 5% require real vehicle road testing.
Autonomous driving simulation test scenario
Compared with traditional cars, autonomous vehicles have undergone tremendous changes in terms of test evaluation content, working conditions, tool chains, etc., and diversified traffic scenarios will become the core elements of automatic driving testing.
A core element of simulation testing
Simulation test is an important way to simulate the road test of self-driving cars, through the simulation platform scene library, quickly build a simulation test scene of various working conditions, so that the car runs in a virtual state.
Simulation test scenarios under various operating conditions
Overall, a simulation test library needs to have the following features:
1 The scene library has rich data sources: in addition to the basic data of the road, it also includes the weather, time, and special situations of the road facilities. and other working conditions, and even based on traffic accident data to establish a thematic simulation scenario.
2 Scene modeling software can be efficient modeling: modeling software, automated scene feature extraction tools, 3D scene editing tools, and scene library management tools must be professional and efficient.
3 Applicable to L2+ or above automatic driving simulation test scenarios: it can cover AEB, ACC, FCW, LDW and other functional test scenarios.
4 Project visual management: data collection, vehicle status, real-time working status of sensors, and project progress can be visually managed.
The advantages of road testing through the virtual simulation test platform are very obvious - in the case of no need for mass production hardware at all, and no need for real cars, the algorithm can be comprehensively tested by means of simulation, and the bugs in the algorithm part can be found out and solved in advance as much as possible.
High-precision maps are the basis for the operation of the autonomous driving simulation test platform
Because the high-precision map provides an accurate description of the lane-level road network, road model, road identification, three-dimensional representation of actual road participants (centimeter-level accuracy), and their locations. This detailed road condition information is of great significance for scene extraction and modeling of autonomous driving simulation tests.
High-precision maps are used for simulation testing
For complex road networks and real scenes, the more rich the scene library, the stronger the functionality of the automatic driving simulation test platform, and Yitutong has rich data resources, such as SD/HD maps, actual road acquisition data, etc. The import of these data can help the automatic driving simulation test platform to develop a large number of simulation scenarios.
It can be said that high-precision maps are the basis for the operation of the automatic driving simulation test platform.
What services can Edotong provide for simulation testing?
Since the results of road surveying and mapping involve confidential data of National Geographic, depots and simulation platforms will have some barriers to entry when processing map data, which requires a certain mapping qualification of the cartographer.
The data compliance service process provided by E-Tutong for simulation testing
Yitutong has Grade A surveying and mapping qualifications, and at the same time lays out the field of automatic driving simulation earlier, which can provide customers with integrated solutions such as simulation test data acquisition, desensitization, storage, simulation test management and other integrated solutions that meet laws and regulations and meet customer requirements. Up to now, ETutong has cooperated with a number of OEMs and Tier 1 in the verification and simulation of autonomous vehicles, and has completed 30+ models. Collect 2 million+ kilometers of road data from Chinese cities (covering Beijing-Tianjin-Hebei, Yangtze River Delta, Pearl River Delta, major cities in the central and western regions, etc.), and accumulate thousands of T data, including scenarios: road facility weather, time, special situations...
In addition to participating in the autonomous driving simulation test projects of many well-known car manufacturers at home and abroad, Yitutong has also established close cooperation with third-party institutions such as mathWorks, an internationally renowned software developer, in the field of automotive simulation, jointly promoting the development of automatic driving simulation test tools and promoting the commercial application of simulation testing.