ADAS Comfort-Critical Scenarios Extraction

The growing spread of assisted and autonomous systems in passenger vehicles has recently drawn attention to the comfort of such systems for end users. Studying comfort raises nontrivial challenges, as humans can recognize and realize in their own driving behavior what it is comfortable and what is not, based on their own subjective perception, which is dependent on numerous factors such as perceived risk, vibrations, predictability and motion sickness. Therefore, an Advanced Driver Assistance System (ADAS) can realize a comfortable driving, for its own passengers, if it has a human-like driving style. The objective is then to understand and replicate human driving in assisted and autonomous driving systems. To do so, it is needed to relate the subjectivity of the comfort to objective metrics that can be incorporated in a control system design Such metrics, defining human driving, can be extracted from datasets of naturalistic driving, which are, by nature, huge and unstructured. This thesis aims at tackling these problems by implementing a tool able to identify “comfort-critical scenarios”, i.e. scenarios where discomfort can be found, extract them from raw datasets with their related objective metrics and visualize them in a web-tool visualizer for further analysis. The thesis is structured as follows. First, it is covered the theory of comfort in ADAS and the objective metrics related to it are reviewed, according to the literature. Also, it is defined the concept of comfort-critical scenario and how they can be found in naturalistic driving. Then, it is described how the tool has been realized in detail. Finally, the tool is validated such that some comfort-critical scenario are possibile to be identified and extracted from collected data.