Virtual radar sensor with characteristic properties
Rain, glare or dirt – factors any driver is familiar with as interfering with the perception of the environment affect sensors as well. Advanced driver assistance systems (ADAS) have to cope with them and initiate vital actions in spite of faulty sensor data. To reproducibly validate this in virtual test driving, researchers from the Institute for Automotive Engineering of TU Graz together with Magna Steyr Engineering developed a radar sensor model that realistically simulates the sensor properties and can be used as early as in the concept stage. The model was comprehensively examined and tested using CarMaker.
Development of a radar sensor model that models the characteristic properties of real-world sensors while minimizing the required computing time.
A phenomenological sensor model efficiently calculates physical sensor effects based on mathematical correlations. The model has been validated and optimized in virtual test driving.
A phenomenological sensor model operates with geometrical information from the simulation environment. It analyzes whether the simulated traffic objects are located within the acquisition range of the sensor and if they are obstructed by other objects. The list of objects generated this way is subsequently loaded with radar-specific measurement noise. In this process, the properties of real-world sensor components are modeled by means of simple mathematical correlations in a way that makes efficient use of computing time. Environmental influences such as weather and the properties of the detected objects are taken into account here as well. Additionally, the virtual sensor randomly loses previously detected objects for a short period of time. Like in real-world sensors, the noisy signals are processed with a Kalman filter and subsequently made available to the assistance system.
The simulation results illustrate the relevance of testing ADAS in the context of dealing with realistic, i.e. faulty, sensor signals. The phenomenological sensor model developed by TU Graz and Magna Steyr Engineering can be used as early as in the concept stage for virtual development and validation; at a time when only a small amount of data from measurements is available. It minimizes the required computing time and is easy to parameterize based on data sheets. The utilization of this model in the virtual development with CarMaker can lead to a high level of maturity of a system even in early stages. As a result, costly development loops can be avoided and development times significantly reduced.