Analysis and Adaptive Estimation of Human Car Following Behavior for Advanced Driver Assistance Systems

In the field of advanced driver assistance systems (ADAS) the capability to accurately estimate and predict the driving behavior of surrounding traffic participants has shown to enable significant improvements of the respective ADAS in terms of economy and comfort. The interaction between the different participants can be an important aspect. One example for this interaction is the car following behavior in dense urban traffic situations. There are different phenomenological or psychological models of human car following which also consider variations between different participants. Unfortunately, these models can seldom be applied for control directly or prediction in vehicle applications. A different way is to follow a control oriented approach by modeling the human as a time delay controller which tracks the inter-vehicle distance. The parameters are typically chosen based on empirical rules and do not consider variations between drivers. In this work a time delay controller approach is applied and extended. First real world measurements in urban test drives are recorded by a test vehicle equipped with forward and reward radar sensors. These datasets are analyzed and used to identify the varying parameters and their probability distribution functions for different human drivers. An advantage of the applied model structure is that it makes online learning and adaptation during the driving possible. This allows adapting prediction models during real world drives even in closed loop control and hence improves the prediction quality. Further, the identified driver models can be used to establish virtual multi vehicle scenarios and build up a virtual traffic environment. The obtained prediction results for different test drives show satisfactory results and could well capture the differences between drivers.