Automatic Generation of Safety-Critical Test Scenarios for Collision Avoidance of Road Vehicles

Abstract

It is apparent that one cannot rely solely on physical test drives for ensuring the correct functionality of autonomous vehicles. Since physical test drives are costly and time consuming, it is advantageous to accompany them with computer simulations. However, since most traffic scenarios are not challenging, even simulations are often too time consuming. To address this issue, we present an approach that creates automatically critical driving situations, i.e., situations with a small solution space for avoiding a collision. Our approach combines reachability analysis for determining the size of the solution space with optimization techniques to shrink it. The solution space is reduced by shifting the initial states of traffic participants, demanding an immediate and correct action of the vehicle under test. We demonstrate our approach by automatically increasing the criticality of several initially uncritical situations recorded from real traffic.

Generating Critical Test Scenarios for Automated Vehicles with Evolutionary Algorithms

“…Our previous work [20] is the first approach that generates critical scenarios with a small solution space for the vehicle under test, which we refer to as the ego vehicle from now on. In that work, the drivable area is used as a measure for criticality.…”

“…This measure enables the quantification of criticality in many more driving situations than the Time-To-Collision. However, [20] only allows the generation of simple scenarios on straight, non-intersecting roads and only realized translation of other traffic participants.…”

“…We present an approach to generate critical scenarios for testing motion planners in complex traffic situations. Based on the optimization of the drivable area of the ego vehicle, our method provides the following improvements compared to [20]: 1) A new parametrization of scenarios is presented for realizing more complex and diverse scenarios.…”

“…2) By using evolutionary algorithms (EA), we find better local minima over a larger range of scenarios. This is especially advantageous compared to the local linearization used in [20] despite the highly nonlinear optimization problem at hand. 3) To improve the optimization process and to consider only relevant driving situations, a method to prune the parameter space of a scenario is presented.…”

“…The subsequent formulation of the underlying optimization problem is similar to [20], except that we consider collisions among other traffic participants. For an illustration of the introduced variables, we refer to Fig.…”