Effectiveness and Adherence of a Robot-Enhanced Imitation Therapy in Autistic Children

Robot-Enhanced Therapies (RET) are gaining recognition in autism interventions, though limited empirical evaluations of their effectiveness exist. This study introduces a novel RET focused on imitation training to promote social skills, incorporating gestures from children's songs to enhance engagement. This work presents preliminary findings from eight autistic children who underwent a 12 -week intervention, assessing improvements in adaptive abilities using the Adaptive Behavioral Assessment Scale (ABAS-II) and Primo Vocabolario del Bambino (PVB), with the Social Adaptive Domain (SAD) of ABAS-II as the primary outcome. Treatment adherence is measured as the proportion of correctly imitated gestures, which is also quantified by leveraging a CNN-based gesture recognition algorithm. While results show overall trends of improvement, the primary outcome does not reach statistically significant. However, adherence was significantly higher in subjects with improved social skills, underscoring the importance of active participation The gesture recognition algorithm achieved an F1-score of 0.57. While findings are promising, the comparison with a control group will permit differentiating the outcomes of the therapy protocol and the developmental trajectory. Future efforts should focus on improving the gesture-imitation algorithm and refining adherence assessment tools.