Behavioral and kinematic outcomes of adaptive pauses in VR social cognition training for autistic children

Autism Spectrum Disorder (ASD) is characterized by persistent challenges in social communication and restricted, repetitive behavioral patterns. Virtual Reality (VR) has emerged as a promising tool for social-cognitive rehabilitation, yet optimizing these interventions through personalized pacing remains underexplored. Adaptive Pauses (APs)-brief regulatory breaks designed to regulate arousal and prevent cognitive overload-have emerged as a key mechanism for sustaining engagement and optimizing learning.

This study expands on a preliminary, validated VR-based social cognition intervention and examines the comparative efficacy of three AP implementation methods: rule-based, Wizard of Oz (human-triggered), automated-random (algorithm-driven), and no-AP.

Forty-six autistic children with level 1 were randomly assigned to one of the four conditions. Each VR session embedded APs consisting of guided breathing and sensory relaxation. Behavioral and kinematic data were continuously recorded across sessions, alongside pre- and post-intervention assessments of social cognition.

Significant improvements in social-cognitive outcomes were observed from pre- to post-test, replicating earlier findings. Behavioral indices showed a trend toward improvement across sessions, and kinematic analyses indicated increases in acceleration and displacement, suggesting more comfortable environmental conditions. However, no significant differences emerged among the four groups in any outcome measure.

VR-based interventions effectively improved social cognition in autistic children, as measured by ToMTB, with robust session-driven gains in behavioral performance and kinematics, supporting the main result. While AP methods did not produce differential intervention outcomes, the findings establish the feasibility and safety of integrating structured pauses into immersive interventions without impeding progress. However, conclusions regarding the efficacy of APs remain preliminary due to limited statistical power and absence of generalization beyond the experimental context.