Post published on Bluesky on 29 Mar 2026 11:29

Abstract

ABSTRACT While repetitive behaviors such as hand flapping are associated with autism, non‐autistic toddlers also exhibit this behavior (e.g., when excited), making clinically meaningful differences difficult to discern prior to age 2. Computer vision, a subfield of artificial intelligence (AI), has the potential to capture subtle movement differences that may otherwise be overlooked. We examined whether movement amplitude and frequency of repetitive hand flapping differed between autistic and non‐autistic toddlers. Based on prior literature, we hypothesized that autistic toddlers would exhibit more intense or vigorous flapping, as reflected by increased amplitude and frequency of repetitive hand flapping. We analyzed 81 manually annotated video segments from 28 toddlers (autism spectrum disorder (ASD): n = 14; no ASD: n = 14), ages 13–16 months (M = 14.78, SD = 0.85), participating in a 3‐min bubble play activity (1–10 events per child). Using AlphaPose, we automatically identified arm key points and quantified amplitude and frequency. Event‐level analyses revealed that, compared to non‐autistic toddlers, autistic toddlers showed significantly greater amplitude (p = 0.028) but did not significantly differ in the frequency of repetitive hand flapping (p = 0.464). In contrast, event‐averaged results showed no group differences in amplitude (p = 0.997) or frequency (p = 0.383). These findings highlight the variability of repetitive hand flapping and suggest that averaging may mask potentially important nuances of repetitive hand flapping. Clinically, the dynamic nature of these early, subtle behaviors may require more fine‐grained observation to identify meaningful differences during this highly dynamic period of early development. Summary Video‐based pose estimation was used to examine kinematic differences in repetitive hand flapping during a 3‐min bubble play activity in autistic and non‐autistic toddlers. Event‐level analyses of all repetitive hand flapping instances revealed a significantly elevated amplitude of hand flapping in autistic toddlers relative to non‐autistic toddlers. Event‐averaged analyses appeared to mask group amplitude differences, suggesting that an event‐level analysis of repetitive hand flapping may better capture clinically meaningful kinematic differences. Findings underscore the possible added value of examining the quality of repetitive hand flapping to identify clinically meaningful differences during a highly dynamic developmental period.