Estimating Children's Personalities Through Their Interaction Activities with a Tele-Operated Robot.

Based on the little big-five inventory, we developed a technique to estimate children's personalities through their interaction with a tele-operated childcare robot. For personality estimation, our approach observed not only distance-based but also face-image-based features when a robot interacted with a child at a close distance. We used only the robot's sensors to track the child's positions, detect its eye contact, and estimate how much it smiled. We collected data from a kindergarten, where each child individually interacted for 30 min with a robot that was controlled by the teachers. We used 29 datasets of the interaction between a child and the robot to investigate whether face-image-based features improved the performance of personality estimation. The evaluation results demonstrated that the face-image-based features significantly improved the performance of personality estimation, and the accuracy of the personality estimation of our system was 70% on average for the personality scales.