Abstract
This study introduces and evaluates a novel virtual reality (VR) adaptive system that advances the state of the art by tailoring social-cognitive training to individual performance, addressing gaps in traditional interventions for children with Autism Spectrum Disorder (ASD). A sample of 23 children (aged 6-8 years) diagnosed with Level 1 ASD participated in a four-session program integrating traditional Theory of Mind tasks with gamified, adaptive VR environments. The online adaptive engine provides real-time support with hints (question repetition), feedback, and reinforcement strategies, while the offline engine adjusts difficulty levels between sessions based on performance. Pre- and post-test assessments using the Theory of Mind Test Battery (ToMTB) and Social Responsiveness Scale evaluated participants' progress. Statistical analyses, including generalized linear mixed models and the Aligned Rank Transform, were used to examine behavioral and biosignal metrics such as motor movement and electrodermal activity (EDA). Results revealed significant improvements in ToMTB post-test scores, suggesting that the adaptive system positively influenced ToM abilities, while no significant changes were observed in the SRS-2 scores. Behavioral analysis indicated progressive accuracy gains and reduced reliance on hints across sessions. Kinematic analysis revealed that whole-body acceleration and displacement increased across sessions, while EDA indicated a significant rise in tonic mean, tonic variance, and phasic variance levels. These findings demonstrate the feasibility and potential efficacy of using adaptive training VR systems to address social-cognitive challenges in individuals with ASD. Future research should expand sample sizes and explore follow-up effects to optimize VR-based interventions.