Simulation test research: foreign autonomous driving simulation companies forge ahead steadily with localization services.

As the functions of ADAS and autonomous driving systems are developed and the expected development cycle of SOTIF functions shortens, the launch of new vehicle models in the competition among automakers is inseparable from a mass of tests. Wherein, the simulation test has been widely adopted by Chinese and foreign automakers. Ideally, about 80% to 90% of the autonomous driving algorithm tests are completed through simulation platforms, 9% to 20% in test fields, and 1% on actual roads.

1. The iteration of simulation tools accelerates, and the 3D realistic and visual simulations provide ever higher test confidence.

Macroscopic, mesoscopic, and microscopic simulation tools and technologies advance. Especially the increasingly refined functions of microscopic simulation tools enable more flexible control over traffic flow simulation, simulate and reproduce road environment, weather conditions (including extreme weathers, e.g., rain, snow, heavy fog and light intensity) and extreme working conditions (accident trigger, etc.), control the simulation settings of various sensors (radar, LiDAR, camera, etc.) and reconstruct scene variants.

All types of simulation companies expedite the iteration of their simulation software, and keep expanding and verifying corner cases, long-tail scenarios and hard examples. They continuously narrow down various abnormal scenarios that may appear in the function development and even expected function development by automakers, and output high-fidelity 3D visualization results to verify the bugs of different models and algorithms of auto companies, for higher confidence in their simulation tools.

A simulated road environment needs to define multiple components, such as roads (lane lines, pavement materials, etc.), traffic signs, traffic lights, traffic participants (motor vehicles, non-motor vehicles, pedestrians, etc.), elements around the road (green belts, stations, buildings, etc.) and weather conditions (day, night, sun, rain, etc.). A variety of sensor models and user-defined sensors can be used to detect these objects. In general, static scenes are constructed by collecting actual environmental information combined with existing HD maps, or the needed environmental elements are artificially created.

Scenario simulation sensors include camera, LiDAR, radar, ultrasonic radar, GPS/BDS, IMU, V2X and other modules, of which the camera simulation needs to simulate multiple complex real weather conditions, automatically adjust the weather, and support camera simulation in different weather and light conditions; the LiDAR simulation referring to the scanning modes of real LiDAR, simulates the emission of each real ray, intersects with all objects in the scene, and generates real point cloud data.

2. Realize the comprehensive testing and verification of ADAS/ADS digitalization through unlimited coverage of scenario variants.

The scenarios in the real world are infinitely rich, extremely complex, and unpredictable. It is very hard to completely reproduce these scenarios in a virtual environment. How to use limited test scenarios to map an infinitely rich world is the key to effective testing and verification of autonomous driving. The simulation test based on scenario library is an important route to solving the problem of insufficient autonomous driving road test data. The higher the real-world coverage of the test scenarios in the scenario library, the higher the accuracy of the simulation test results.

In autonomous driving simulation, ASAM’s OpenX Standards have gained extensive attention from all over the world. The standards cover five parts: OpenDRIVE, OpenSCENARIO, OSI, OpenLABEL, and OpenCRG. Wherein, OpenDrive defines the description method of static scenarios; the OpenSCENARIO-defined contents involve the description of dynamic scenes, location and speed of the car owner, and information of other traffic participants; OpenCRG focuses on the description of physical information related to road surfaces, and is mainly used for friction between tires and the ground.

The simulation test is to simulate dangerous working conditions, including a great many harsh weather environments, complex road traffic, and typical traffic accidents. The parameter reorganization scene is to parameterize the existing simulation scenes, and complete the random generation or automatic reorganization of simulation scenes, featuring infinity, scalability, batching, and automation. The purpose of the parameter reorganization scene is to supplement the uncovered, unknown scenes, such as natural driving, standards and regulations, and dangerous working conditions, so as to well cover blind spots in autonomous driving function tests.

When verifying uncovered, unknown scenarios, Israel’s Foretellix provides a verification and validation platform Foretify using the Coverage-Driven Verification (CDV) Methodology (on the one hand, highly automated generation and regulation of millions of test vectors to verify various scenarios; on the other hand, the safety and production dashboard with big data analytics shows objective work status of quantifiable, measurable verification and validation). The Foretify solution can detect system bugs, edge cases and unknowns in the early stage of development, help prevent costly recalls due to design defects (some of which are fatal), and port scenarios to different maps and ODD. Its current customers include Denso, Valeo, NVIDIA, Mobileye and Volvo.

3. Foreign autonomous driving simulation companies continue to expand cooperation with Chinese companies, and make steady progress in localization services.

Foreign autonomous driving simulation companies are working harder on layout in the Chinese market. For example, Germany’s PTV Group, France’s ESI Group, and Israel’s Cognata have established subsidiaries in China to facilitate simulation business expansion. Among them, PTV has covered over 600 customers in more than 90 cities of China. In addition, NI, dSPACE, VI-grade and the likes go on deepening their partnerships with Chinese automakers and solution providers such as Automotive Intelligence and Control of China (AICC) and RoboSense.