Passenger Comfort Assessment via Motion Complexity Analysis for Autonomous Vehicles

Traditionally, passenger comfort in vehicles is perceived as being most influenced by acceleration and jerk. Consequently, the current research primarily focuses on developing control algorithms to limit the maximum acceleration and jerk of the vehicle in order to improve passenger comfort. However, naturalistic driving studies demonstrate that such simple characteristics are insufficient for accurately evaluating passenger comfort. This study identifies motion complexity as a key factor of passenger comfort. A series of naturalistic driving studies are conducted, during which passenger comfort is assessed using a 5-point Likert scale. Moreover, a real-time passenger comfort measurement based on electromyography (EMG) and stepwise regression is proposed to facilitate seamless data collection. Time-series features representing motion complexity are then introduced to better describe passenger comfort. Hierarchical regression confirms that simple characteristics of motion are insufficient to explain passenger comfort, and shows that the proposed motion complexity features have a substantial effect on passenger comfort. Finally, a machine learning-based real-time passenger comfort estimation method is developed according to the foregoing findings. Experimental results show that the proposed method can accurately estimate passenger comfort in real-time using only vehicle motion information. The findings of this study suggest that vehicle motion complexity should be considered in future passenger comfort studies.