The safety validation of automated driving of SAE level 3 and higher (AD) is still an unsolved issue. In the validation process, criticality metrics can be used for two different purposes. First, for the identification of test scenarios from recorded data that are later tested in simulation. Secondly, for an estimation of the safety of a specific AD system based on the likelihood of critical situations in test drives or in other words as a safety surrogate. In the past, different metrics for those purposes have been defined that work well in specific scenarios such as longitudinal traffic. However, a metric that describes criticality in all situations and is applicable to human and AD traffic is currently not available. In this paper, an approach to define a criticality metric is introduced. The metric is based on the definition of criticality as the level of driving requirements in the specific situation. The computation of the proposed metric uses elements of model predictive control using an objective function that contains four elements that describe the difficulty of the driving task. Based on those demands and a simplified driving dynamics model, the solution with the minimal criticality is computed. Finally, the metric is tested in four test scenarios that are typical for highway traffic. A short parameter variation study is conducted in order to study certain effects of the algorithm and to identify room for improvement.
Criticality Metric for the Safety Validation of Automated Driving using Model Predictive Trajectory Optimization
2018-11-01
470430 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch