Multi-Modal ASD Screening System: A Preliminary Study

Early screening for autism spectrum disorder (ASD) can significantly improve intervention outcomes. However, acknowledging the diverse nature of ASD and the variety of its symptoms, a multi-modal approach becomes necessary for effective autism screening. In this study, we propose a framework powered by the Robotic Operating System (ROS) to integrate various ASD screening tools as separate modalities seamlessly. With a focus on behavioral data, our preliminary study involves the analysis of three distinct ASD symptom groups using three modalities: the sensorized toy car for assessing repetitive behaviors, the Lightwheel for analyzing sensory input sensitivity, and the Roboparrot for simulating social interaction initiation. By fusing the results of each modality, we achieved enhanced accuracy and reliability in ASD screening, surpassing the individual modality accuracy by more than 10% on the average.