German-Japanese research cooperation in automated and connected driving: Virtual validation

The goal of VIVID is the development of a virtual validation toolchain, which includes the simulation of the environment as well as sensors.

The approval of automated and connected driving requires new efficient procedures for validation. One promising approach is the scenario based testing to control the parameter space in a targeted manner. The application in a virtual environment for purpose of validation promises benefits regarding costs, increased efficiency as well as safety for the execution. The main goal of the project is therefore the development of a virtual validation toolchain, which includes the simulation of the environment as well as sensors.

Further significant goals are the standardization and the use of open interfaces. These further global cooperation and support the transparent validation process, which is indispensable for the approval of automated driving functions.

The consortium consists of ten project partners including automobile manufacturers, suppliers, various research institutes as well as medium sized companies.

The project VIVID is a German-Japanese research cooperation for a virtual validation methodology for intelligent driving systems. This cooperation adequately addresses the necessity of international standards for the safety assurance relying on simulation.

A common approach to improve the perception quality as well as the redundancy in case of failures is to utilize multiple sensors. However, the evaluation criteria for such a combination of sensors are still unclear.

Therefore, perception requirements on object level are formulated as sensor-neutral criteria. These can then be used to quantify the strengths and weaknesses of each individual modality. Emphasis is also placed upon understanding the complementarity between different perception algorithms and sensors. The same criteria can then also be applied to any sensor fusion algorithm in order to quantify performance.

Another aspect is the redundancy or the robustness of the perception towards algorithm or component failures. New methods are developed to adequately evaluate the robustness towards failure as well as redundancy.

For the safety argumentation of automated and connected vehicles, simulations are increasingly used in addition to real drives. Simulation environments in the form of 3D worlds with different material properties serve as input for sensor system models in order to reproduce the corresponding behavior of the sensors. However, there are no quantifiable requirements for the simulation environment for rendering-based Radar and Lidar system models.

Therefore, based on comparisons performed at different levels of the signal processing chain. Experiments and simulations are used as tools and compared with each other. A new methodology serves as the basis of the approach, which quantifies geometric regions in the simulation with different deviations using selected metrics.

This new approach is applied to different radar sensors and different environments. In addition to static scenarios with simple reference targets, dynamic runs and complex scenarios are analyzed.