Scenario Identification for Validation of Automated Driving Functions

The continuous development and integration of Automated Driving Systems (ADS) leads to complex systems. The safety and reliability of such systems must be validated for all possible traffic situations that ADS may encounter on the road, before these systems can be taken into production. Test-driving with ADS functions requires millions of driving kilometers to acquire a sufficiently representative data set for validation. Modern cars produce huge amounts of sensor data. TNO analyses such data to distinguish typical patterns, called scenarios. The scenarios form the key input for validating ADS without the need of driving millions of kilometers. In this paper we present a newly developed technique for automatic extraction and classification of scenarios from real-life microscopic traffic data. This technique combines 'simple' deterministic models and data analytics to detect events hidden within terabytes of data.