For virtual testing of automated driving and driver assistance functions, digital models of the environmental perception sensors, such as cameras, LiDAR, and radar, installed in the vehicle are required. These models must accurately replicate the real behavior of the sensors while also placing feasible demands on the hardware used for simulation, so they can be economically used in large simulation campaigns. To address this trade-off between model accuracy and efficiency, research is being conducted on how a sensor’s behavior can be approximated based on data utilizing methods from the field of machine learning.
Within the project, a suitable data representation for the surroundings detected by environmental perception sensors is defined. Additionally, a model architecture to approximate the behavior of the environmental perception sensors in various driving scenarios will be developed. The resulting model will be evaluated and trained on data from different test setups. Ultimately, the integration of the model into multiple test setups will be demonstrated.
