Perceived Training Utility of Artificial Intelligence in Virtual Reality Taekwondo Sparring: A Repeated-Measures Study

Abstract

Virtual reality (VR) can provide repeatable sport-specific practice, but the training value athletes assign to AI-controlled opponents remains uncertain. This study compared perceived training utility between AI-controlled and human-controlled VR Taekwondo sparring and examined whether bout duration modified this response. Thirty-one collegiate Taekwondo athletes completed four conditions in a 2 × 2 within-subject design: AI-controlled 30-s, AI-controlled 60-s, human-controlled 30-s, and human-controlled 60-s sparring. Perceived training utility was the primary outcome; condition-specific measures of presence, realism, opponent perception, biomechanical perception, usability, enjoyment, satisfaction, behavioral intention, perceived exertion, and cybersickness were also collected. Repeated-measures ANOVA showed higher perceived training utility for AI-controlled than human-controlled sparring, F(1, 30) = 6.32, p = .018, ηp² = .174, whereas the duration effect and opponent × duration interaction were not significant. No significant opponent-type or duration effects were detected for the other condition-specific outcomes. In an exploratory linear mixed-effects model, realism, opponent perception, and biomechanical perception were positively associated with perceived training utility. Within this VR system and brief exposure protocol, athletes assigned somewhat greater training utility to the AI-controlled opponent. This finding concerns perceived usefulness rather than demonstrated performance gains and supports further testing of whether realistic opponent behavior and movement fidelity translate into objective training benefits.