Dance imitation AI detects autism with 79.2% accuracy via new SCSI

A study from University of Coimbra (Pereira et al., arXiv:2608.12548) introduces a computational framework that treats dance imitation as a window into motor behavior differences in autism. The researchers captured 3D motion data from autistic and neurotypical adults as they imitated dance sequences solo and in dyads with social framing. Using Dynamic Time Warping to measure consistency, they proposed the Social Context Sensitivity Index (SCSI), which quantifies how much movement variability changes when a social partner is present. This approach moves beyond static motor assessments to capture real-time social adaptation.

Results showed a clear distinction: neurotypical participants increased movement variability across upper and lower limbs in the socially-framed condition, while autistic participants maintained highly consistent movements across both contexts. The resulting classifier distinguished autistic from neurotypical adults with 79.2% balanced accuracy, suggesting that social modulation of motor imitation is a robust, objective biomarker. The authors argue this could enable more inclusive human-centric technologies—such as adaptive robotics, interactive training tools, and clinical assessment systems—that account for neurodiverse motor signatures.