Emotion recognition and expression in therapeutic social robot design

To improve the healthcare and wellbeing of the elderly population, a therapeutic social robot was designed and its interaction with human was investigated in this research. The focus is to enable the designed robot to understand human's emotion and express emotion through gestures accordingly. To identify human emotion, tactile sensors were utilized and installed on the robot's head to detect users' touch signals, and classify them into four patterns (hit, pat, stroke and unknown). The classification task was performed based on three methods (Localist Attractor Network, Temporal Decision Tree and Naive Bayes Classifier). Exploration of the trade-off between high accuracy and low computing workload was also carried out for all three methods. Four gestures were designed to respond to the four emotion recognition patterns, which can express some basic human emotional statuses such as unhappiness, love, relaxation and confusion.