RESIST - Resilienzbewertung von Wahrnehmungs- und Planungsansätzen in kooperativ interagierenden Automobilen bei unerwarteten Störungen
Cooperative interacting vehicles are an important step towards fully automated driving. However, a broad applicability ranging from highway to city traffic at different seasons and daylight situations, bring up manifold risks. To guarantee safety and robustness considerations, new methods for validation and evaluation of all sensor data that have been aggregated within a common event horizon are necessary. Special care has to be taken with respect to the resilience against unexpected disturbances originating from the environment, weather conditions or faulty communication channels. Recent methods for validation, e.g. by recording real test data, seem to have reached the limits of complexity due to the huge number of possible environmental conditions. The required standard qualification process would imply the need of roughly one billion kilometers of test driving according to latest industrial sources.
In this project proposal we address these problem through a new approach of generating synthetic disturbances on real test data. We aim at first aggregating the sensor data of all cooperating vehicles into a common image of the environment. Based on this, a methodology to evaluate the resilience is developed. Through the consequent inclusion of environmental conditions and potential interferences on communication channels as well as the usage of virtual prototypes, multi-sensor systems can be verified and evaluated efficiently under a wide range of ambient conditions. By the application of fitness landscaping techniques, specifically those conditions can be determined under which the complete system does no longer guarantee full or sufficient functionality. Additionally, the human influence by situation-dependent time to react or take over control is being analyzed and integrated into our simulation environment. The feedback of the simulation results allows for a targeted optimization of the involved algorithms and system components, to finally ensure safe operation in real driving situations.