Move or sit: Preschoolers' in-kindergarten physical activity and its association with fundamental motor skills using AI-assisted movement assessment.

This observational study examined physical activity (PA) across three kindergarten contexts and its association with fundamental motor skills (FMS), with a proprietary AI-assisted video/action-analysis tool used as a supplementary assessment aid. Fifty-seven children (61.11 ± 12.73 months; range 45-81 months) from one kindergarten in Chengdu, China, were observed during morning exercise, organized physical education (PE), and free play using the Observational System for Recording Physical Activity in Children-Preschool Version (OSRAC-P); FMS were assessed with the standardized Fundamental Movement Skills Development Test for Children Aged 3-10 Years. The AI component was the proprietary "Xiaoneng Classmate" action-analysis platform developed by Chengdu Tixiaoneng Education Technology Co., Ltd.; manufacturer materials document AI-assisted action/video analysis and digital testing/data functions but do not disclose a public model architecture or study-specific validation metrics. Mean activity level was highest during morning exercise (3.29 ± 0.09), followed by free play (2.94 ± 0.40), and lowest during organized PE (2.46 ± 0.26). Organized PE had the highest sedentary frequency (46.58% ± 12.69%), whereas free play had the highest MVPA frequency (33.92% ± 13.03%). To address multicollinearity in the original high-dimensional regression, parsimonious individual-level models adjusted for age, sex, and BMI and included MVPA during organized PE and free play. Organized-PE MVPA was positively associated with locomotor skill score (B = 0.471 per percentage point, HC3 SE = 0.141, β = 0.407, 95% CI [0.187, 0.754], p = .002) and showed a weaker association with total FMS (B = 0.484, HC3 SE = 0.235, β = 0.234, 95% CI [0.011, 0.957], p = .045); no PA predictor was significant for object-control or stability scores. Morning-exercise MVPA was not entered in the individual-level regression because it had no within-class variability. These findings are context-specific associations and do not establish causal direction.