Integrating robotic kinematics and dynamics with online handwriting features for dysgraphia classification

Abstract

Dysgraphia, a learning disorder affecting handwriting fluency and legibility, can significantly hamper children's academic development. Early and accurate classification is essential for timely intervention and tailored educational support. This study presents a novel deep learning framework that integrates robotic kinematic and dynamic features from a robotic arm replicating handwriting with traditional online kinematic and temporal features extracted from digitized writing samples. By integrating these complementary features, we aim to enhance dysgraphia classification by capturing detailed handwriting patterns. We transform the multidimensional time-series data into a structured tabular format and process each feature set independently using TabNet, a deep learning model optimized for tabular data. To maximize classification performance, we employ a Tanh-based score fusion strategy, dynamically balancing the contributions of both models. Evaluations on a publicly available dysgraphia dataset demonstrate state-of-the-art performance, achieving 91.7% accuracy and 95.2% precision on the most comprehensive classification task. These results highlight the effectiveness of robotic motion analysis in improving handwriting-based dysgraphia classification, offering a promising tool for clinical and educational screening.