Toward a Wearable Affective Robot That Detects Human Emotions from Brain Signals by Using Deep Multi-Spectrogram Convolutional Neural Networks (Deep MS-CNN)

Abstract

Wearable robot that constantly monitors, adapts and reacts to human's need is a promising potential for technology to facilitate stress alleviation and contribute to mental health. Current means to help with mental health include counseling, drug medications, and relaxation techniques such as meditation or breathing exercises to improve mental status. The theory of human touch that causes the body to release hormone oxytocin to effectively alleviate anxiety shed light on a potential alternative to assist existing methods. Wearable robots that generate affective touch have the potential to improve social bonds and regulate emotion and cognitive functions. In this study, we used a wearable robotic tactile stimulation device, AffectNodes2, to mimic human affective touch. The touch-stimulated brain waves were captured from 4 EEG electrodes placed on the parietal, prefrontal and left and right temporal lobe regions of the brain. The novel Deep MSCNN with emotion polling structure had been developed to extract Affective touch, Non-affective touch and Relaxation stimuli with over 95% accuracy, which allows the robot to grasp the current human affective status. This sensing and decoding structure is our first step towards developing a self-adaptive robot to adjust its touch stimulation patterns to help regulate affective status.