Assessing driver subjectivity in a vehicle dynamic simulator through a deep learning-based framework

Artificial intelligence (AI) and data science techniques are integral to modern Engineering research, and this also holds for Automotive Engineering (AE). In this context, given the growing need to analyze relatively extensive datasets obtained from vehicle instrumentation and simulator use, this study aims to explore the relationship between drivers’ subjective assessments (SA) regarding vehicle dynamic behavior and its objective metrics (OM). To make this feasible, resources such as deep neural networks (DNN), virtual car models developed in VI-CarRealTime (VI-CRT) software, and 25 human drivers using a dynamic simulator with 9 degrees of freedom, the VI-Grade Driver in Motion (DiM) 150; were needed. By leveraging data from both virtual simulations and human drivers, the DNN demonstrated its ability to standardize vehicle evaluations across individuals with varying experience levels, ensuring a more consistent mapping between objective metrics and subjective assessments, and presenting satisfactory accuracy of around 90%. While one of the key outcomes of this paper is the capacity to evaluate the vehicle independently of driver subjectivity, another crucial one is the system’s ability to rely on less experienced drivers for vehicle classification. Also, another contribution is the significant reduction in vehicle development time and costs. Overall, this work advances vehicle evaluation by harmonizing subjective and objective assessments, fostering innovation in Automotive Engineering.