Abstract

More and more Advanced Driver Assistance Systems (ADASs) are entering the market for improving both safety and comfort. Adaptive Cruise Control (ACC) is an ADAS application that has high interaction with the driver. ACC systems use limited sensor input and have only few configuration possibilities. This may result in the behaviour of the ACC not matching user's preferences in all cases, resulting in lower acceptance of the system. In this work, we examine the possibilities for a Personalised ACC (PACC), which adapts the ACC settings such that it matches the driver preference in order to increase the acceptance. The driver preferred ACC behaviour is predicted using machine learning techniques and manual driving data. On-road experiments showed that the method is promising as it is able to discriminate between two preference clusters with an accuracy of 85%.

Abstract

More and more Advanced Driver Assistance Systems (ADAS) are entering the market for improving both safety and comfort by assisting the driver with their driving task. An important aspect in developing future ADAS and Automated Driving Systems (ADS) is testing and validation. Validating the failure rate of an ADS requires so many operational hours that testing in real time is almost impossible. One way to reduce this test load is virtual testing or hardware in-The-loop testing. The major challenge is to create realistic test cases that closely resemble the situation on the road. We present a way to use data of naturalistic driving to generate test cases for Monte-Carlo simulations of ADS. Because real-life data is used, the assessment allows to draw conclusions on how the ADS would perform in real traffic. The method, developed in EU AdaptIVe, is demonstrated by testing an Adaptive Cruise Control (ACC) system in scenarios where the predecessor of the ego vehicle is braking. We show that the probability of the occurrence of unsafe situations with the ACC system can be accurately and efficiently determined.