The development in the areas of sensors and electronics has been bringing the automotive industry increasingly closer to automated driving in recent years. Automated functions that need to be continuously supervised by the driver are already on the market. Highly automated driving functions (HAD) will enter the market in the near future. The German Ethics Commission for automated and connected driving stated in its report that “the licensing of automated functions is not justifiable unless it promises to produce at least a diminution in harm compared with human driving, in other words a positive balance of risks” [1]. This leads to the question, how the traffic safety effect of automated driving functions can be assessed taking into account possible positive and negative aspects? This paper introduces comprehensively a method that is used by BMW for the prospective safety assessment of HAD by means of virtual experiments. This method is besides others part of the evaluation and safety assurance activities for HAD. The method is described from the scenario selection via the simulation up to the validation and verification. In contrast to other simulation approaches in this area, which mainly use accident re-simulation, this approach uses Monte-Carlo techniques, in which the initial starting conditions of the simulated driving scenario as well as the parameters of the involved drivers are randomly selected from distributions. These distributions base on accident data as well as on naturalistic driving data. A core aspect of this approach is the stochastic cognitive driver behavior model to describe the behavior of individually different traffic participants in a scenario. In contrast to accident re-simulation based approaches, this approach allows to analyze time-wise larger driving scenarios, which are of importance for HAD, since these functions act throughout the driving within the operational design domain and not only in critical situations.

The method for assessing the safety performance is applied to exemplary HAD. The results cover the positive effects, which are mainly achieved in today known accident scenarios, as well as scenarios, in which potentially newrisks compared to manual traffic can occur. One example for this is the minimum risk maneuver, for which the consequences of different implementation is discussed. Like all other methods (accident analysis, studies in driving simulator or on test track, field operational test) the simulation based approach has advantages and disadvantages. The main criticism is that the assessment is done virtual, which poses the question on the validity of the simulation. In order to tackle this aspect the validation and verification of the method and tool is a key aspect. Therefore, our current conceptual considerations regarding validation and verification are described in this paper.