Toward a Criticality-Guided Sampling Strategy to Reduce Tests for Automated Driving Validation

The validation and certification of Connected and Automated Vehicles (CAV) are crucial prior to their deployment in the real world, and various approaches have been developed to tackle this challenge. The scenario-based approach holds particular promise as it reduces the number of scenarios required to test and certify CAV systems. Various sampling strategies might be applied at various stages of the scenarios’ modeling. We offer to fill a gap identified when considering scenarios modeled at the functional stage. We develop a criticality-guided sampling strategy to streamline the testing process of CAV by selecting functional scenarios that strike a balance between the diversity of encountered situation and their criticality level. This sampling process occurs in two phases: (i) a stratification based on clustering methods applied to the entire set of functional scenarios, (ii) a selection within these groups to encompass a broad spectrum of highly critical scenarios. It achieves a decision-making methodology to find a trade-off between criticality and diversity coverage when selecting a restricted set of scenarios to assess CAV.