Preliminary Study on Detection of Behavioral Features at Conversation Breakdown in Human-Robot Interactions

Abstract

Communication robots are expected to help elderly people that live alone. However, it is difficult for existing conversation systems to communicate smoothly. One reason for this is that the robot cannot recognize conversation breakdowns. To solve this problem, we tried to detect human motion characteristics when conversation is not established. The conversation experiment was conducted as a Wizard of Oz experiment. The behavior during the conversation was measured using an RGBD sensor, and the measured data analyzed through FFT analysis. The results showed that characteristic movement was found in head yaw movement.