In-depth analysis of autonomous driving test scenarios

The content of this article mainly focuses on the connotation, function, scale, perspective and data source of autonomous driving vehicle testing scenarios.

01. Scenario elements and scenario classification When constructing a test scenario, you first need to clarify the elements covered by the test scenario. The scenarios in the real world are endless and the elements are complex. Decomposing the scenarios and extracting the types of elements contained in the scenarios are the basic methods for dimensionality reduction and abstraction of real-world scenarios. In order to facilitate the analysis and organization of the elements, the scene elements need to be classified. There are many ways to divide the scene elements according to different organizational structures. From the perspective of the needs of autonomous driving tests, scene elements can be divided into two types of elements: environmental elements and self-driving tasks.

Based on the basic attributes of scene elements, they can be divided into static elements and dynamic elements, etc. Based on the topological relationship of scene elements, they can be divided into road elements, traffic participant elements, meteorological elements, etc. Scene elements are the basis for test scene construction, play an important role in the test and evaluation process of autonomous vehicles, and are the main support for the evaluation system.

Based on the different attributes of scene elements, the complete test scene elements include: Static environmental elements within a certain spatial range: road types, traffic facilities, geographic information, static obstacles, etc.
Dynamic environmental elements within a certain time and space range: such as dynamic traffic sign facilities and communication environment information.

Traffic participant elements in the driving environment: motor vehicles, non-motor vehicles, pedestrians, animals, etc. The tested vehicle does not belong to this element category, mainly because its behavior does not necessarily have to be predefined.

Meteorological environmental elements in the driving environment: light, temperature, humidity, climate, etc.

Initial state of the vehicle: Ego’s initial state, goals, and behavioral elements of the vehicle being tested.

Test scenarios are classified. Specific classification methods include the following categories: 1. Classification by the data source of the test scenario, such as test scenarios formed by natural driving data, dangerous working condition data, standards and regulations, etc., which are mainly used to test the effectiveness and safety of autonomous driving functions. 2. Classification by road structure level mechanical energy, such as road basic road network scenarios, unstructured road scenarios, static scenarios and dynamic scenarios. Test scenarios of different levels such as scenes are mainly used to meet the needs of different stages of function development. 3. Classification by the degree of abstraction represented by the test scenario, such as logical scenarios, functional scenarios, and specific scenarios. 4. Classification by the application method of the test scenario, such as simulation test scenarios and field test scenarios. 5. Classification by the attribute characteristics of the elements contained in the test scenario, such as structured road fields.

02. Test Methods Test scenario research methodology involves the classification of test scenario components, scenario construction methods, key technologies for test scenario application, etc.

2.1 Test scenario generation steps Test scenario construction requires steps such as data collection, data analysis and mining, and scenario generation, and the scenario is reproduced in the actual test environment through virtual simulation and physical testing technology.

2.2 Test case design theory and method Design test case theory and method: There are many test case design methods, commonly used methods include orthogonal experiment method, boundary value analysis method, equivalence class partition method, decision table method and error guessing. Comparative test, enumeration test, coverage test, Alaska test method, etc.

2.3 Common test theory methods Orthogonal experiment method: Orthogonal experiment method is a test case design method based on orthogonal table. It covers different value combinations of multiple factors by selecting a limited set of test cases, thereby reducing the number of test cases and improving test efficiency.

Boundary value analysis method: Boundary value analysis method is a test case design method based on system boundaries. It selects boundary values ​​and special values ​​of system input as test cases to detect the behavior of the system in boundary conditions.

Equivalence Class Partitioning Method: Equivalence Class Partitioning Method is a test case design method based on the characteristics of input data. It divides the input data into different equivalence classes and then selects representative test cases to cover each equivalence class.

Decision table method: The decision table method is a test case design method based on system rules. It creates a decision table, lists the system input conditions and corresponding output results, and then selects test cases based on the decision table.

Error guessing method: Error guessing method is a test case design method based on false assumptions. It verifies the fault tolerance of the system by assuming possible errors in the system and designing corresponding test cases.

03.

Direction of autonomous driving simulation testing

In the process of standardization of autonomous driving test scenarios, the compatibility of different acquisition platforms and technical solutions and the intercommunication and sharing of database data can be achieved mainly from the following aspects: 1. Formulate a unified data format standard: First, it is necessary to define a common data exchange format or protocol, such as AVP (autonomous driving verification and validation) data format, so that various acquisition devices and systems can output data according to the same standard. This ensures that the data obtained from different platforms can be understood and parsed by each other. 2. Build a standardized test scenario library: Establish a standardized test scenario library including various basic scenarios, special scenarios, etc., and clearly describe the parameter indicators of each scenario to facilitate reproduction and verification in different test platforms and solutions. 3. Build a cross-platform data interface: Develop a general data access and conversion tool to support the access and conversion of multiple data sources, so as to realize data interaction between different platforms. 4. Build a data sharing platform or data center: Through cloud service technology, a safe and reliable data storage and sharing center is established, and all participants can upload data that meets the standards to the platform to achieve the aggregation and intercommunication of data resources. 5. Strengthen information security and privacy protection: While ensuring data sharing, take strict encryption and desensitization measures to ensure the security of sensitive information and comply with relevant laws and regulations. 6. Actively participate in industry organizations and alliances: Join domestic and foreign autonomous driving related standard organizations and industry alliances to jointly promote and adopt industry-wide data formats, interface specifications and security standards to accelerate the realization of interoperability and consistency at the data level.

04. Test scenario

Scenarios that serve specific testing purposes are called test scenarios. Test scenarios can help people place specific research objects in specific situations for research, examine their performance and reactions, and thus form certain test conclusions. For self-driving car test scenarios, their meaning can be understood from the following aspects.

4.1 The connotation of test scenarios

When scenarios are applied to autonomous vehicle testing, they describe a certain type or certain driving environment and the tasks of the vehicle under test in the above driving environment.

The driving environment describes the basic traffic environment conditions and the status and behavior of traffic participants, and can present or reflect the environment and process of traffic scenarios in the real world.

The goals and behaviors of the tested vehicle describe the tasks that the tested vehicle needs to complete or is expected to complete in the above driving environment. The above two parts together constitute a specific test scenario or a certain type of test scenario. In the test scenario, the performance of the specific functions of the tested vehicle in the driving environment can be examined and analyzed.

4.2 The role of test scenarios

Test scenarios are used to test, verify or evaluate the functions or performance of autonomous vehicles. Application test scenarios must have clear test purposes, such as testing the expected behavior and performance requirements of the vehicle. The vehicle can be verified and evaluated by its performance in the test scenario.

4.3 Scale of the test scenario The test scenario describes the traffic driving environment within a certain time and space range and the test tasks of the vehicle under test.

The test scenario contains dynamic elements, whose behavior reflects a dynamic process with a certain time span; all elements contained in the test scenario are arranged in an environment of a certain spatial scale.

The time and space scales of the test scenario are determined based on the test task. For the time scale, an emergency obstacle avoidance scenario usually lasts for a few seconds, while a vehicle following scenario may last for minutes or even hours. For the space scale, the test scenario may include a section of road or a road network consisting of several roads.