Application Story
ADGenerator: An AD/ADAS simulation testing scenario generation and management platform
CATARC - ADGenerator: An AD/ADAS simulation testing scenario generation and management platform
Summary
The current Simulation of autonomous driving scenarios faces problems such as difficulty in applying road data collection, inconsistent scenario data formats, low scenario generation efficiency, small scenario coverage, and incomplete scenario data management system. An integrated solution and tool are needed to overcome a series of difficulties in scenario data before the Simulation application. The ADGenerator of CATARC Intelligent and Connected Technology Co., Ltd. automatically converts road acquisition data containing sensors such as cameras, LiDAR, and inertial navigation into ASAM OpenDRIVE and ASAM OpenSCENARIO DSL format AD/ADAS Simulation testing scenario files in bulk, thereby solving the problems of difficult application of road acquisition data, inconsistent scenario data formats, and low scenario generation efficiency; To solve the problem of low scenario coverage, ADGenerator uses the ASAM OpenSCENARIO DSL Standard to generate Simulation scenario generalization of road acquisition data by changing parameter values such as vehicle speed, vehicle action start time, and vehicle action duration, and selects more dangerous Simulation testing scenarios through TTC and THW indicators to improve the efficiency of scenario Simulation testing; In addition, ADGenerator provides a powerful data management system (Label based scenarios management system) that can automatically classify, quickly filter, and retrieve ASAM OpenSCENARIO DSL scenarios.
Application story
CATARC Intelligent and Connected Technology Co., Ltd. has 1 million kilometers of road acquisition data including camera, LiDAR, inertial navigation, and other sensor data. To apply road acquisition data with real environmental information and real dynamic scenario information to AD/ADAS system testing, and achieve effective management and rapid retrieval of AD/ADAS system testing scenarios, CATARC Intelligent and Connected Technology Co., Ltd. has launched the research and development work of the ADGenerator project, And based on the ASAM OpenDRIVE and ASAM OpenSCENARIO DSL standards, automatic batch generation generalization from data collection to Simulation scenarios has been successfully achieved, greatly improving the efficiency of scenario generation and AD/ADAS testing Verification.
During the ADGenerator project, we encountered four challenging issues: how to automatically convert the collected static road information into a simulated usable Map? How to automatically convert dynamic scenario information into simulated dynamic scenarios and effectively combine it with Simulation Map files? How to achieve automatic batch generalization generation of data collection to Simulation scenarios? How can the Simulation scenarios generated by generalization be effectively classified, managed, and quickly retrieved?
Road information is mainly obtained by fusing camera data and LiDAR data, including parameters such as number of lanes, type of lane lines, and curvature of lane lines. How to convert road information into a high-precision Map that can be used for Simulation is an urgent problem that needs to be solved. We have studied the Simulation Map construction logic of multiple domestic and foreign commercial software for autonomous driving Simulation testing, including Carmaker, PreScan, 51sim one, etc. We can use the collected road information to convert it into a Map unique File Format that can be run by any commercial software for autonomous driving Simulation testing, but we cannot achieve cross-simulator use of scenarios. We found that the high-precision maps defined by the ASAM OpenDRIVE Standard can be compatible with multiple commercial software for autonomous driving Simulation testing. Due to the good compatibility of ASAM OpenDRIVE maps and the strong logical construction and easy conversion of ASAM OpenDRIVE maps, we chose the ASAM OpenDRIVE Standard as the basis for converting Simulation Map files.
The collected dynamic scenario information is obtained by fusing data from cameras, LiDAR, millimeter-wave radar, and other sensors, including parameters such as traffic participant types, traffic participant actions, and weather. To convert the collected dynamic scenario information into dynamic Simulation files, it is necessary to solve the compatibility problem of scenario files in multiple simulators, At the same time, it is also necessary to solve the problem of association between dynamic Simulation files and Simulation Map files, as well as the problem of automatic batch generalization generation of Simulation scenarios. We found that the ASAM OpenSCENARIO DSL Standard can completely solve all the above problems, so we used the ASAM OpenSCENARIO DSL as the conversion basis for dynamic Simulation files in ADGenerator.
To solve the problem of effective classification and fast retrieval of Simulation scenarios, ADGenerator provides a Label-based scenario data management system. ADGenerator refers to ISO 34503 and ISO 34504 standards to provide a Label library and a mapping relationship between the labels in the Label library and the parameter fields of ASAM OpenDRIVE and ASAM OpenSCENARIO DSL. When an external OpenX Simulation scenario is uploaded to ADGenerator or when ADGenerator generalizes to generate a Simulation file, ADGenerator will traverse the parameter fields within the Simulation file and automatically classify the Simulation file into appropriate categories based on the mapping relationship. Users can also quickly filter and retrieve Simulation files through tags.
At present, ADGenerator has approximately 50000 scenarios and is compatible with VTD, Carmaker, dSPACE, CARLA, and SCANeR5 commercial software for autonomous driving Simulation testing. ADGenerator has already served many universities and car companies in China.
For example, NIO Inc. uses ADGenerator to transform Simulation scenarios of RosBag format road data, and verifies its self-developed ADAS algorithm with Carmaker software; China Automotive Innovation Corporation uses ADGenerator to classify and manage Simulation scenarios, including CIDAS accident scenarios, Standard regulatory scenarios, expert experience scenarios, and so on.
In the development of the autonomous vehicle, the ASAM OpenX Standard plays an important role, providing a unified data interface Specification for the entire industry, promoting the seamless flow of Simulation data between autonomous vehicle simulators, and greatly reducing the complexity and cost in the development process of the auto drive system.
Shuai Zhao / chief expert of CATARC
BACK TO OVERVIEW
Follow us on
