Method development for evaluation of Automated Driving Systems: Investigating the performance of two reference driver models

Abstract

This work deals with the subject of safety verification and validation of automated driving systems with a high level of automation, in this particular case, SAE level three. The method investigated is scenario based simulation, with the use of refer ence driver models as performance comparison. The operational design domain is a traffic jam at a highway, and the response is evaluated. Specifically, the simulated scenarios are the cut-in and cut-out scenarios. The two investigated reference driver models are: the ALKS UNECE models and the FSM, which is a proposed performance model by the UNECE. A parameter sweep was done for the two scenario types, where the parameter limits were chosen to target difficult or unavoidable scenarios. Time to collision (no crash) and impact speed (crash) were analysed. First, results from the ALKS model is presented by itself. Then, the performance of the FSM is compared to the ALKS model in the cut-in scenarios. Results show that the ALKS model crashes more than twice as many times as the FSM. Also, the FSM has a more spread out distribution of impact speeds, while the ALKS model is sharply peaked at the initial velocity. This suggests that, for SAE level tree automation systems, the FSM is a more challenging performance target than the ALKS model, in the analysed scenarios.