How Can Machine Learning Models Be Used for Subjective Assessment of Safety and Comfort: Application on Free Lane Change Maneuver

Abstract

Subjective evaluation plays a key role in autonomous driving feature validation for safety, comfort, and driving quality. As for being objective, it is also important to evaluate the autonomous features with key performance indicators (KPI) depending on physical parameters before stepping into the delivery phase. To provide better driving experience for autonomous features, calibration parameters need to be tuned carefully while considering safety and comfort. Calibration parameters can be evaluated in terms of safe, unsafe, comfortable, or uncomfortable states through questions that allow the evaluation of the passenger’s feelings during real-world testing which includes predefined scenarios and environments. In this paper, we proposed a method that performs the rating of the free lane change maneuver in terms of safety and comfort by employing the machine learning algorithms to model the passenger feedback according to the questionnaire for the subjective evaluation of the test maneuver execution. After trying several machine and deep learning regression techniques, we have shown that Extreme Gradient Boosting (XGB) regressor can be used to model drive feeling accurately for validation and calibration purposes. The constituted evaluation model can be utilized to improve quality of the autonomous driving, optimize calibration parameters and achieve user acceptance.