Child behavior recognition in social robot interaction using stacked deep neural networks and biomechanical signals

With the growing integration of social robots into pediatric environments, understanding and monitoring child-robot interaction has become increasingly important. Toward the advancement of biomechanical monitoring systems for pediatric applications, this study presents an innovative approach employing stacked Deep Neural Networks (DNNs) for the objective monitoring of interaction dynamics between children and social robots. The study focuses on quantitatively analyzing behaviors exhibited by children towards four types of social robots, each equipped with accelerometers and gyroscopes. These sensors capture vibration signals and angular displacements, translating them into statistical features-Kurtosis (K) for accelerometer data and Signal Magnitude Area (SMA) for gyroscope data. This transformation facilitates an objective analysis of the interaction dynamics. The model demonstrates high efficacy with accuracy, precision, recall, and F1 scores of 0.941, 0.94, 0.941, and 0.939, respectively. While the initial emphasis of this research is on the development of the stacked DNN model, the study also sets the stage for future applications in real-time mobile monitoring and biomedical robotics. This research contributes to the understanding of child-robot interactions by setting an objective and developmental stage-appropriate perspective, and paves the way for advancements in interactive technologies within developmental contexts.