Predicting emergency decision-making in conditionally automated driving from personality traits and heart rate

초록

Emergency situations in conditionally automated driving force the driver to quickly take over control from the artificial intelligence (AI). Studying which aspects can influence such driver decisions is crucial for developing safer automated driving technologies. In the present study, we hence ask whether decision-making both during and after take over can be predicted based on personal factors as well as physiological measures. Our experiment employed a realistic virtual reality (VR) scenario with an automated driving environment in which participants were able to change the driving mode. The scenario's driving course contained multiple forks indicated by warning signs for which one direction was "safe" and the other led to a cliff. During the testing phase, pedestrians suddenly blocked the safe direction, and the AI informed the driver that it would steer the car off the cliff. Here, participants had two, staged decisions to make: whether to switch to manual driving or to continue in automated driving, and if they had switched, whether to "self-sacrifice" to save the pedestrians or not. In our sample of N = 108 participants, 83.3 % made the switch, and of these, 56.7 % steered away from the cliff trying to save themselves. Importantly, statistical analyses showed significant influence of psychopathy, perspective taking, gender, and heart rate on the decision-making after take over. Based on this data, we created a hard-voting prediction model that outperformed other machine learning benchmark algorithms, showing that it is possible to predict drivers' decision-making after take over from personal and physiological data.

키워드