The field of automated driving has been the focus of research in both, academia and industry in the recent decades. Thanks to the advancements in sensor technology currently used for Advanced Driver Assistance Systems (ADAS), automatization of several parts of the driving task was enabled. However, the introduction of new technologies into the market requires a testing phase. Different test methods can be applied to assess the automated driving functions and their features. Nevertheless, the systematic test of functions for driver assistance systems requires the identification and analysis of a huge number of traffic scenarios. Since most traffic scenarios are not challenging, an approach that automatically generates critical driving scenarios can reduce the testing efforts and costs.

This paper presents a methodology for automated definition of critical scenarios, which is composed of seven steps: define a critical maneuver, define the logical scenario, model the critical maneuver using a mathematical description, define the assessment criteria, obtain the concrete scenarios, specify a performance improvement method, and obtain the boundary scenarios. Firstly, Independent if the validation/homologation is based on a virtual or a field test, the tests should be focused on the most critical situations, since the test phase is costly. The selection of critical scenarios, based on critical maneuvers, is an important phase to develop testing methods. Thus, the critical maneuver selected was the lane change maneuver.

The logical scenario defined describes the event where Ego drives in highway, presenting certain speed and acceleration, and due to a traffic jam, the front vehicle and the rear vehicle, on the adjacent lane, are stationary and there is a considerable distance between Ego and the rear vehicle. After, the mathematical modelling of the lane change maneuver was performed. Composed of three parts, the trajectory planning, which is responsible to generate the desired trajectory that ensure a safe lane change maneuver based on different driver style, the second is the vehicle model, which describes the longitudinal and lateral motion of the Ego vehicle, and the last is the driver model, which is responsible to drive Ego vehicle along to the planned trajectory.

The assessment criteria was based on the safety distances and on the maximum lateral acceleration performed during the maneuver. The concrete scenarios were generated by variating three parameters, in this case, relative velocity between Ego and the front vehicle, the distance between Ego and the front vehicle, and alpha, the coefficient that affects the lane change time. Then, the performance improvement approach was applied, using the k-Nearest Neighbor Classifier, to reduce the computational elapsed time. After completing the first six steps, the concrete scenarios have been classified as critical and non-critical. Lastly, through the analysis of each neighbor point, it was possible to reduce the result space, to a few number of scenarios that can be easily manipulated, the boundary scenarios. Thus, the proposed methodology shown to be an efficient method for the definition of critical scenarios, since thousands of scenarios could be generated and classified in a matter of seconds.

Mr. Thaddäus Menzel, IDIADA Fahrzeugtechnik GmbH, GERMANY; Mr. Thiago de Borba, IDIADA Fahrzeugtechnik GmbH, GERMANY