A smart art toy interaction system based on multimodal biofeedback

This study aims to develop and preliminarily validate a multimodal biofeedback smart art toy framework that can (i) identify child-relevant emotional states from physiological signals and (ii) provide adaptive artistic feedback in a closed interaction loop. To address the limitations of single-modality or command-driven toy interaction, we integrate photoplethysmography/heart rate variability (PPG/HRV), galvanic skin response (GSR), and acceleration (ACC) signals with a hybrid Convolutional Neural Network-Graph Convolutional Network (CNN-GCN) model and a feedback mapping module. Using the public Wearable Stress and Affect Detection (WESAD) dataset for algorithm-level verification and simulation-based regulation experiments, the optimized CNN-GCN model achieved 91.3% accuracy and 90.6% recall, outperforming comparison methods. In closed-loop simulation, the regulation success rate reached 75%, with stronger effects in negative emotional states. These findings support the technical feasibility of multimodal affective perception and adaptive artistic feedback, while the current evidence should be interpreted as proof-of-concept under simulation rather than direct real-world clinical efficacy in children.