Flexible Stochastic Microscopic Traffic Model for ADAS Testing

Methods to assess the safety of advanced driving assistance systems (ADAS) and further highly automated vehicles (HAV) are of paramount importance. There is a wide consensus that virtual testing will be part of them, as physical testing will not be even nearly exhaustive. Virtual testing requires models of the controlled vehicle, but also of the surrounding traffic, which for a long time will consist mainly of human-driven vehicles. Many driver and traffic models exist, but they are usually tailored for a specific situation or requirement. As the reaction of human drivers is affected by many different factors, like the traffic conditions, the time or the country, there is a need for a flexible structure which can be easily tuned to different situations. If we split the driver reaction in a decision and an actuation step, we argue that the actuation step can be represented by few stochastic actuation models which do not depend strongly on external factors. This paper shows such models and their performance with highway data both from China and Germany.