Virtual Identification of Driver-Specific Characteristics for Assisted Lateral Control 2024-01-5090

Advanced driver assistance systems (ADAS) have become an integral part of today’s vehicle development. These systems are designed to provide secondary support to the driver, but the driver is primarily responsible for the driving task, e.g., lane-keeping assist (LKA). The driving setup and testing of these LKA systems is very time-consuming and usually applied in the car, based on experiences and subjective evaluation. This results in a cost-intensive calibration of the system. An objective-based calibration procedure can increase efficiency. For a targeted calibration of the system, it is necessary to define and identify key performance indicators (KPIs), which are able to describe the secondary support in sufficient detail. Usually, subjective feelings are used to derive KPIs. Vice versa, there are no results on how to design an LKA without any subjective assessment, before the calibration. With this in mind, this paper is focused on filling this unknown aspect by using virtual methods to identify driver-specific KPIs in a free driving scenario. A model sequence feedback control (MSFC) is used for the LKA. In addition, three different driver types (sporty, normal, and gentle) are parameterized, and the driving environment is modeled based on a statistical analysis of rural roads. Based on a design of experiments (DoE), the inputs of the LKA are varied, and the variation is measured using KPIs. The DoE output results in KPIs, which allow driver-specific conclusions to be drawn, in a closed-loop scenario. In addition, principal components (PCs) for the characteristic parameters were generated, and each type of driver can be described with sufficient precision with only three PCs. The drivers have 25 distinguishable KPIs in common. These KPIs aren’t vehicle-specific and can be used at a higher level for the driver-specific closed-loop description and for a model-based LKA calibration.