Page - 43 - in Integration of Advanced Driver Assistance Systems on Full-Vehicle Level - Parametrization of an Adaptive Cruise Control System Based on Test Drives

3.2. System and Component Levels or a recorded video of a real situation. An example for such an application is given in [GS11]. There, Lane-Departure Warning (LDW) and Lane-Keeping Assistant (LKA) systems are tested. The advantage of HIL is that complicated situations can be tested in a time and cost-efficient way. Nevertheless the correct behaviour of the system has to be defined in advance. Another example for HIL testing is given in [ABF+03]. One significant disadvantage of the HIL method in general is that for tests of an automotive Electronic Controller Unit (ECU), the whole data bus has to be simulated. Otherwise, the ECU will not work. Another challenge is the significant diversity of variants a system is developed for. The aim of vehicle manufacturers is to use the same system in many different vehicles or derivates, [WSSR10]. This leads to a very high number of combinations of different systems or modules with which the ADAS should work. Therefore, a structured devel- opment and test plans are needed to design a stable, affordable and innovative system. To accomplish these goals, Model-in-the-Loop (MIL) and HIL methods are used. With HIL tests, the device being tested is a physical component. In contrast, MIL tests assess software. The advantage is that MIL tests can be performed before any physical part is available. In [WSSR10], Wehner et al. show that a time and cost-efficient development process requires realistic sensor data, even in the early development phase. The challenge is that the right sensors may not be available at that time. Magosi describes in [Mag13] a method for dealing with this problem. Commercially available simulation packages provide optimal sensors as environmental recognition sensors, meaning they make no errors and have no delays. Magosi developed a phenomenological sensor model that applies noise and random signal losses to the optimal data. Thus, the components in the simulation can be trained to handle such special situations. In [BMLSE13], this sensor model was compared to a commercially available sensor model by simulating a frequently occurring motorway situation. The ego vehicle, equipped with an AEB system, is travelling at 130km/h and overtakes a platoon of different vehicles travelling at 90km/h. At a certain distance, one of the vehicles of the platoon moves into the lane of the ego vehicle. The driver of the ego vehicle does not react to the emerging object, and the AEB system applies the brakes. The outcomes of the simulations with the two different sensor models were completely different. If the optimal sensor model is used, the AEB system prevents the crash. The simulation with the proposed model leads to a speed reduction of the ego vehicle, but the AEB system does not prevent the collision. The main difference between the two simulations is that the phenomenological model requiresacertain time fromtheappearanceof theobjectuntil thedataprocessing passes the information to the ADAS. This realistic behaviour can be used to improve the simulation quality and to find ways to deal with realistic data. 43