Modeling Active Perception Sensors for Real-Time Virtual Validation of Automated Driving Systems

The validation of the functionality is up until now always an insolvable problem for the development of automated driving systems (ADS). The simulation-based test methods (e.g. X-in-the-Loop test) with sensor models are nowadays being developed and considered by the most players in the automotive industry as an economically feasible solution for the validation of ADS. However, in order to ensure the reality and the validity of test results, the sensor performances and the driving environment should be modeled realistically. This reality presents a challenge for modeling technology. In this dissertation, a novel grey-box method for modeling the active automotive perception sensors with neither the efficiency disadvantages of the white-box methods nor the reality disadvantages of black-box methods. After a statement of the basic problems in the modeling, a framework for the grey-box modeling is introduced in the dissertation. According to this framework, the sensor model to be developed should consist of the following parts: sensor performance model, wave propagation model and environment model. For modeling the sensor performances, a novel modeling method, so-called Cell-Volume Concept (CVC), is developed. Three kinds of variants of this method are introduced and compared. After an analysis of the advantages and disadvantages of the variants, the so-called vector-projection variant is chosen for the modeling of a radar and implemented exemplarily. Based on the vector-projection Cell-Volume Concept, a wave propagation model, which simulates the propagation of the electromagnetic waves from the sensor into the atmosphere, is developed. The possible physical phenomenon during wave propagation are analyzed and selectively modeled regarding the modeling necessity. The sensor radiation pattern will also be modeled. Furthermore, to ensure simulation efficiency and avoid unnecessary computing effort, an ergodic method is developed. As one of the most important part of the environment model, a model for representing the reflectivity (distribution) of different vehicles at different aspect angles is developed. This model plays an essential role in calculating the performance of waves. The modeling of reflectivity distribution is especially meaningful for the simulation of the detection and measurement in short range. Finally, the developed sensor model is verified via comparing simulation results with real sensor outputs. By using some verification test cases, the developed sensor model has demonstrated the capability of representing sensor performances dynamically and efficiently. In summary, the developed sensor model in this dissertation is appropriate to be applied for radar sensor modeling. The simulation efficiency and fidelity can be ensured simultaneously. For finding the application possibility of this sensor model in modeling the other types of active perception sensors, some discussions and suggestions are also given and summarized