The protocol for this systematic review and meta-analysis was preregistered in the PROSPERO (International Prospective Register of Systematic Reviews; CRD420251266730). No deviations from the registered protocol were identified during the review process. This review was conducted and reported in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 statement (Checklists 1 and 2) [30].

The eligibility criteria were established in accordance with the Population, Intervention, Comparison, Outcomes, and Study design (PICOS) framework.

The exclusion criteria were as follows: (1) nonoriginal studies, including reviews, commentaries, conference abstracts, and case reports; (2) studies in which participants were not exclusively children or adolescents diagnosed with ASD; (3) interventions that did not involve VR-based exercise or physical activity components; (4) studies that did not report prespecified behavioral, executive function, or motor performance outcomes; and (5) studies that did not meet the predefined eligibility criteria for study design, such as single-case studies. These exclusion criteria were established to ensure sufficient specificity and comparability across studies regarding participant characteristics, intervention modalities, and outcome measures, thereby supporting subsequent evidence synthesis and meta-analysis.

Studies with extractable quantitative data were synthesized using meta-analysis, whereas studies lacking sufficient data for effect size calculation were summarized descriptively using qualitative synthesis methods. When relevant outcomes were reported but data were incomplete or could not be directly extracted, corresponding authors were contacted, when feasible, to obtain missing data. Studies for which complete data remained unavailable were documented accordingly and included only in the qualitative synthesis.

To comprehensively identify relevant studies, systematic searches were conducted in the following electronic databases and search platforms: PubMed (National Library of Medicine), Embase (Elsevier), PsycINFO (EBSCOhost), Scopus (Elsevier), Web of Science Core Collection (Clarivate), ProQuest, ScienceDirect, and IEEE Xplore. The final search was conducted on May 7, 2026. In addition to electronic database searches, gray literature sources and clinical trial registries, including ProQuest Dissertations and Theses and ClinicalTrials.gov, were searched to reduce the risk of publication bias. Backward citation searching of the reference lists of included studies and relevant review articles was also performed, alongside forward citation searching through Web of Science. Manual searches were also conducted to identify additional relevant studies.

The search strategy was developed using a combination of controlled vocabulary terms (eg, MeSH and Emtree terms) and free-text keywords and was adapted to the search syntax and requirements of each database. The search strategy incorporated the following key concepts: (1) ASD; (2) VR; (3) physical exercise or physical activity; and (4) behavioral, executive function, and motor outcomes. To maximize search sensitivity and specificity, Boolean operators (AND/OR) were used to combine search concepts, and database-specific subject headings were applied, when appropriate, to broaden the search scope. The complete and reproducible search strategies, including all search terms, limits, and filters, are provided in (S1 in Multimedia Appendix 1).

Only English-language full-text articles were included. No automated search translation tools, natural language processing techniques, or text-mining tools were used to generate or optimize search terms. All search strategies were manually adapted to the characteristics of each database. All retrieved records were imported into EndNote (Clarivate Plc) for reference management and duplicate removal before study screening. Before implementation, the search strategies underwent internal review to ensure the completeness and accuracy of the search syntax and content.

Two reviewers (YJ and KQ) independently conducted study selection using a 2-stage screening procedure. First, titles and abstracts were screened to identify potentially eligible studies, followed by full-text review to determine final eligibility. All retrieved records were imported into EndNote for reference management and duplicate removal. During screening, both reviewers independently completed each stage and documented the reasons for full-text exclusion. Any disagreements regarding study eligibility were resolved through discussion with a third reviewer (KC) until consensus was reached. Because only English-language full-text studies were included, no translation procedures were required. No automated screening tools or machine learning–assisted screening methods were used during study selection, and no additional information was requested from study authors to determine eligibility.

Data extraction was independently performed by 2 reviewers (YJ and KQ) using a prespecified Microsoft Excel data extraction form. Extracted information included study characteristics (author, publication year, and country), study design, participant characteristics, age, sample size, diagnostic methods for ASD, levels of VR immersion, VR devices and interaction modalities, exercise intervention type, intervention frequency, outcome measures, and primary findings. The 2 reviewers independently completed data extraction and cross-checked all extracted data for accuracy. Any disagreements arising during data extraction were resolved through discussion, with consultation from a third reviewer (KC) when necessary. When study data were missing or insufficiently reported, study authors were contacted, when feasible, to obtain additional information. No automated tools, machine learning techniques, or AI-assisted approaches were used for data extraction. Because only English-language full-text studies were included, no translation procedures were required. No specialized software was used to extract data from figures or graphs. When multiple reports originated from the same study, relevant data were combined and extracted as a single study. In cases of overlapping or inconsistent information across reports, priority was given to the report containing the most comprehensive data or the longest follow-up period.

This systematic review focused on the following 3 core outcome domains:

For each outcome domain, all eligible measurement instruments, assessment time points, and available statistical results were extracted. In the meta-analysis, postintervention outcome data were preferentially extracted. When multiple postintervention assessment time points were reported, data collected immediately after the intervention were prioritized for quantitative synthesis, whereas results from other time points were summarized descriptively. When multiple measurement instruments were used for the same outcome, the instrument with the most complete reporting, the broadest use, or the greatest relevance to the outcome domain was prioritized for analysis. No modifications were made to the predefined outcome domain definitions or outcome selection criteria during the review process. When relevant data were missing, unclear, or unverifiable, they were recorded as “not reported,” and no assumptions or estimations were made.

Two reviewers (YJ and KQ) independently assessed the risk of bias of the included studies using Cochrane Collaboration tools, applying Risk of Bias 2 (RoB 2.0; Cochrane Bias Methods Group) for RCTs and Risk of Bias in Nonrandomized Studies of Interventions (ROBINS-I; Cochrane Bias Methods Group) for non-RCTs. Any discrepancies arising during the risk-of-bias assessment were resolved through consultation with a third reviewer (KC). Under RoB 2.0, 5 domains were evaluated, including the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selective reporting. Each domain was rated as having low risk, high risk, or some concerns, in accordance with established criteria [31].

The ROBINS-I tool assesses 7 domains of potential bias in nonrandomized studies. For the first 3 domains—bias due to confounding, participant selection, and intervention classification—the risk-of-bias assessment in nonrandomized studies differs substantially from that in randomized trials, primarily due to the absence of randomization. In contrast, for the remaining 4 domains—deviations from intended interventions, missing data, outcome measurement, and selective reporting—there is substantial overlap between the 2 assessment tools [32]. Accordingly, RoB 2.0 was applied to all included studies, with ROBINS-I additionally used to assess the first 3 domains in non-RCTs. No modifications were made to the RoB 2 or ROBINS-I tools, and no automated tools, machine learning techniques, or AI-assisted methods were used in the risk-of-bias assessment. No additional information was requested from study authors during the risk-of-bias assessment process. The results of the risk-of-bias assessment were used to inform interpretation of the findings and the overall evaluation of evidence certainty.

Ethical approval was not required for this systematic review. This systematic review was a systematic review and meta-analysis of previously published studies and did not involve the collection of individual-level data.

The study selection process is presented in the PRISMA flow diagram (Figure 1). A total of 11,576 records were identified through database searching, and 4669 records remained after duplicate removal for title and abstract screening. Following title and abstract screening, 329 full-text articles were assessed for eligibility. During full-text screening, 314 articles were excluded for the following reasons: review articles (n=93), insufficient relevance to the research topic (n=125), failure to report relevant outcomes (n=29), interventions not involving physical activity (n=38), single-case study designs (n=25), and participant samples not exclusively comprising children and adolescents with ASD (n=4). An additional 6 reports were identified through other sources. However, these reports overlapped with already included studies and were therefore not treated as independent studies. Ultimately, 15 [1,2,4-8,10,39-45] studies were included in the systematic review. Of these studies, 4 used [1,6,39,45] single-group pre-post designs, 1 [4] did not provide sufficient quantitative data for effect size calculation (eg, means and SDs), and 1 [45], although reporting quantitative data, was not included in the meta-analysis because behavioral outcomes were reported in only a small number of studies and substantial heterogeneity existed across measurement methods. Therefore, these studies were included only in the narrative synthesis. In total, 9 studies [2,5,7,8,10,40-43] were included in the meta-analysis. Details of the study selection process are shown in Figure 1.

This systematic review included 15 eligible studies, with detailed study characteristics summarized in Tables 1 and 2. Across these studies, 439 children and adolescents aged 6‐18 years with ASD were included. ASD was diagnosed using the DSM-5 (Diagnostic and Statistical Manual of Mental Disorders [Fifth Edition]) in 6 [1,5,7,8,41,42] studies, the Autism Diagnostic Observation Schedule–Second Edition (ADOS-2) in 2 [10,43] studies, and the World Health Organization (WHO)’s ICD-10 (International Statistical Classification of Diseases, Tenth Revision) in one [4] study. In 6 [2,6,39,40,44,45] studies, ASD diagnoses were established by qualified physicians.

Of the 15 studies, 3 were conducted in China, 3 in Korea, 2 each in Australia, Brazil, and Iran, and one each in Serbia, Saudi Arabia, and Italy. Regarding study design, 11 studies were RCTs, whereas 4 used single-group pre-post designs.

Regarding intervention modalities, 8 studies used Xbox 360 Kinect (Microsoft)–based systems, 3 used head-mounted displays (HMDs), 2 used cave automatic virtual environment (CAVE; Electronic Visualization Laboratory), and 2 used the Nintendo Wii Fit Plus (Nintendo). The physical activity interventions primarily consisted of sport-based games, including tennis, golf, bowling, soccer, cycling, dancing, and rugby. Intervention duration ranged from 3 days to 16 weeks across the included studies.

For behavioral outcomes, social communication was assessed using the SRS-2 in 3 studies, while repetitive and stereotyped behaviors were measured using the Repetitive Behavior Scale–Revised in one study. Executive function was assessed using the WCST (3 studies), the Flanker Task (3 studies), the Stroop Color–Word Test (2 studies), the DCCS (one study), and the CoSAS-S (one study). Motor abilities were evaluated using the TGMD-2/3 (2 studies), the MABC-2 (2 studies), the BOT-2 (2 studies), the PDMS-2 (1 study), the Biodex Balance System (1 study), the Reactive Postural Balance (1 study), and the MDS (1 study).

Figure 2 [1,2,4-8,10,39-45] and Figure 3 [1,2,6,45] summarize the risk-of-bias assessment for all included studies. Across the 5 domains, 7 studies were judged to be at low risk of bias, 3 raised some concerns, and 5 were rated as high risk, primarily due to deficiencies in the randomization process in nonrandomized studies and the lack of reported means and SDs for outcome data in several studies. Overall, the included studies showed a relatively low risk of bias in the domains of deviations from intended interventions and outcome measurement. For the 4 nonrandomized controlled studies, 3 domains related to the absence of randomization were additionally assessed. Of the 12 judgments assessed in these domains, one was rated as moderate risk, whereas the remaining judgments were classified as low risk.

According to the GRADE framework, the certainty of evidence differed across outcome domains (Table 3). The certainty of evidence for motor performance outcomes was rated as low, mainly due to considerable heterogeneity and moderate imprecision. In contrast, the certainty of evidence for executive function outcomes was rated as moderate and was downgraded solely due to concerns regarding risk of bias in some studies. Overall, although VR-based physical exercise interventions showed relatively consistent beneficial effects on executive function, the strength of evidence supporting behavioral and motor outcomes remains limited. Detailed GRADE assessment procedures and the specific reasons for downgrading each outcome are provided in the Summary of Findings table (S2 in Multimedia Appendix 1).

This study systematically reviewed and meta-analyzed the effects of VR–based exercise interventions on behavioral outcomes, executive function, and motor performance in children and adolescents with ASD. Unlike previous reviews that primarily focused on general VR training [28], social skills interventions [46], or single functional domains [47], this review specifically examined VR interventions centered on physical activity participation and integrated behavioral, executive function, and motor outcomes to better clarify the potential clinical value of VR-based exercise interventions within comprehensive ASD intervention frameworks.

Overall, the findings indicate distinct evidence patterns across outcome domains. Current evidence suggests that findings for behavioral outcomes remain inconsistent, whereas executive function outcomes demonstrate relatively consistent positive effects and motor outcomes exhibit substantial between-study variability and contextual dependence. Moreover, certainty of evidence, heterogeneity, and risk of bias differed substantially across outcome domains. Therefore, interpretation of intervention effectiveness should emphasize evidence consistency and contextual applicability rather than relying solely on pooled effect estimates.

In addition, the VR interventions included in this review exhibited substantial conceptual heterogeneity, encompassing AVGs [18], motion-based interactive training [43], AR-based exercise training [2], and VR systems with varying levels of immersion [44]. Therefore, future research should more precisely differentiate interventions according to immersion level, feedback modality, social interaction format, and exercise task specificity rather than pooling different VR exercise modalities into a single intervention category, thereby improving interpretability and clinical applicability.

This review did not identify a stable or consistent overall effect of VR-based exercise interventions on behavioral outcomes in children and adolescents with ASD. Nevertheless, several individual studies reported improvements in social interaction or repetitive stereotyped behaviors [2,39,44]. This pattern—where positive findings were reported in individual studies despite inconsistent overall evidence—suggests that behavioral outcomes may be highly context-dependent, with intervention effects potentially influenced by training content, interaction modalities, outcome measures, and study design characteristics.

Social motivation theory and mirror neuron system models suggest that interactions with virtual avatars in low-threat virtual environments may facilitate social engagement and imitative behaviors [48]. However, the VR interventions included in this review primarily emphasized physical activity participation, with intervention goals focused on exercise engagement, motor control, task completion, and cognitive processing rather than directly targeting the core social behavioral impairments associated with ASD. Consequently, behavioral improvements may depend largely on whether VR tasks incorporate training mechanisms aligned with the targeted behavioral outcomes [39]. For example, Nekar et al [39] reported that although their AR cognitive-motor training included interactive tasks, the absence of interactive verbal dialogue systems within the game environment may have limited its capacity to promote social communication and language development. The authors further suggested that improvements in social behaviors and higher-order cognitive functions may require longer-term and sustained training exposure. This interpretation is consistent with the findings of the present review, suggesting that although most VR-based exercise interventions may enhance engagement and task involvement, their social interaction components are often limited, particularly regarding real-world verbal communication, emotional feedback, and bidirectional social interaction, thereby restricting the transfer of intervention effects to real-world social behaviors.

In contrast, several studies reported improvements in repetitive stereotyped behaviors. Nekar et al [2] reported that AR-based gamified cognitive-motor training reduced purposeless movements and repetitive behaviors. The authors suggested that these improvements may be associated with the sensory stimulation and physical exertion generated through simultaneous motor and cognitive engagement. Previous studies have suggested that higher levels of physical activity may reduce stereotyped behaviors through postexercise fatigue mechanisms [49]. In addition, the visual, motor, and sensory stimulation provided by AR/VR environments may partially substitute for self-stimulatory needs, thereby decreasing the frequency of repetitive stereotyped behaviors [50]. These findings suggest that the mechanisms underlying VR-based exercise interventions may differ across behavioral outcome types. Improvements in social communication may depend more heavily on sophisticated interaction designs, whereas reductions in repetitive behaviors may be more closely related to physical activity engagement and sensory stimulation [51-55].

Therefore, the present review does not support the simplistic conclusion that VR-based exercise interventions are ineffective for behavioral outcomes. Rather, the available evidence suggests that their behavioral effects are highly task-specific and intervention-dependent. Substantial differences across VR systems in interaction modalities, feedback mechanisms, immersion levels, and integration of social elements may represent one important source of the inconsistencies observed in current behavioral outcome findings. These findings highlight a critical issue within the field: VR technology itself does not inherently produce behavioral improvement; rather, its effectiveness may depend on the specific skills and behaviors targeted within the VR environment. Future studies should further differentiate the behavioral training mechanisms embedded within different types of VR-based exercise interventions, such as verbal interaction, cooperative tasks, social feedback, multiplayer interaction, or contextual generalization training, and investigate which intervention designs are most likely to facilitate the transfer of effects to real-world social behaviors.

Among the outcome domains included in this review, executive function demonstrated the most consistent evidence base. Both the systematic review and meta-analysis suggested that VR-based exercise interventions may positively affect multiple domains of executive function in children and adolescents with ASD, including inhibitory control, working memory, attention, and cognitive flexibility, with relatively low between-study heterogeneity [2,5,10,41].

The positive effects of VR-based exercise interventions on executive function are generally consistent with previous findings. Sepehri Bonab reported that VR-based exercise games integrate engaging task designs, real-time feedback, interactivity, and immersive experiences, thereby providing children and adolescents with ASD with training environments that simultaneously require physical participation and cognitive processing, which may facilitate executive function development [5]. Related studies further suggest that virtual exercise games may not only improve executive function performance but also promote physical health, self-perception, and participation in moderate-to-vigorous physical activity [41,56]. These interventions typically require children to process multiple stimuli across physical and virtual environments while making rapid decisions, thereby imposing substantial cognitive demands that may activate executive control–related brain networks more effectively than exercise modalities focused primarily on physical activity alone [57]. However, some studies have reported that virtual training is not consistently superior to conventional physical exercise for improving executive function, suggesting that intervention effectiveness may depend on factors such as game type, immersion level, and training content [4].

Previous research suggests that exercise may improve executive function in children and adolescents with ASD through multiple interrelated mechanisms [55]. From a neurobiological perspective, exercise may promote structural and functional neuroplasticity by enhancing the functioning of brain regions closely associated with executive processes, such as the prefrontal cortex and hippocampus, thereby supporting improvements in working memory and cognitive flexibility [51]. Furthermore, functional near-infrared spectroscopy studies have demonstrated that exercise can significantly increase prefrontal cortical activation in individuals with ASD, and this enhanced activation appears to be closely associated with improved executive task performance [58]. At the psychosocial level, physical activity may further enhance the beneficial effects of exercise interventions on executive function by promoting social participation, fostering positive experiences, and increasing motivation [52-55]. Therefore, the potential value of VR technology may lie primarily in its ability to enhance engagement and adherence through gamification, real-time feedback, and structured task design, thereby strengthening the effects of exercise interventions on executive function rather than producing independent effects through immersive technology alone.

Compared with executive function outcomes, motor outcomes demonstrated substantially greater heterogeneity and uncertainty. Although several studies reported positive trends in balance performance [7], object control skills [1], and overall motor proficiency following VR-based exercise interventions [10,40,43], substantial variability was observed across studies, and the prediction intervals suggested that intervention effects may vary considerably across different research contexts. These findings indicate that the observed pooled effects may not be generalizable across all forms of VR-based exercise interventions or all subgroups of children and adolescents with ASD.

This inconsistency may primarily reflect substantial intervention heterogeneity across included studies. The VR interventions included in this review involved various device types, including Xbox Kinect, Nintendo Wii, immersive head-mounted display systems, and AR platforms, as well as differing immersion levels and exercise training content, all of which may influence motor learning processes and intervention outcomes [59]. For example, some studies implemented structured exercise training characterized by explicit movement goals, repetitive practice, and immediate motion feedback, and these studies were generally more likely to report positive outcomes [7,8,43,45]. In contrast, other studies emphasized entertainment, free interaction, or general physical activity participation, which may have produced more limited improvements in specific motor skills while primarily enhancing training tolerance and participation motivation, both of which are considered important determinants of exercise intervention effectiveness and sustainability [8,60,61].

Additionally, differences across VR systems in feedback modalities, motion-capture precision, forms of social interaction, and sensory stimulation complexity may influence exercise participation experiences and training responses in children and adolescents with ASD [62,63]. For some children and adolescents with ASD who exhibit sensory sensitivities or attentional regulation difficulties, highly immersive environments may not necessarily produce superior training outcomes, because complex sensory stimuli may increase cognitive load or sensory stress in some individuals with ASD [28]. Accordingly, the findings of this review suggest that the effectiveness of VR-based exercise interventions may be highly context-dependent and that different intervention modalities should not be simplistically treated as homogeneous interventions.

Notably, several studies have indicated that when VR training is used as an adjunct to conventional physical activity or physical therapy, positive changes are more frequently observed in postural control, balance performance, and training engagement [40,42]. These findings suggest that VR-based exercise interventions may be more appropriately positioned as adjunctive or enhancement-based interventions within the motor domain rather than as complete replacements for traditional exercise training. In other words, their potential value may lie primarily in enhancing training enjoyment, increasing opportunities for movement repetition, and strengthening real-time feedback rather than independently providing comprehensive motor rehabilitation [4]. Based on these findings, future research should more precisely differentiate interventions according to immersion level, feedback modality, social interaction format, and exercise task specificity rather than pooling different VR exercise modalities into a single intervention category, thereby improving interpretability and clinical applicability.

The findings of this systematic review suggest that VR-based exercise interventions may hold promise within comprehensive intervention frameworks for children and adolescents with ASD. However, this potential should be understood primarily as adjunctive and context-dependent rather than as a direct replacement for conventional intervention approaches. From a clinical perspective, VR-based exercise interventions may be particularly suitable for children and adolescents with ASD who demonstrate low motivation for physical activity participation, heightened sensory sensitivity, or difficulty maintaining long-term engagement in traditional exercise programs [6,8,10]. Compared with conventional training approaches, the real-time feedback, gamified features, and structured task designs embedded within VR environments may improve engagement and adherence while providing some children with more predictable and lower-stress training settings [41]. However, this evidence remains insufficient to support the widespread implementation of VR-based exercise interventions as stand-alone treatments, particularly for improving core behavioral symptoms [2,39,44]. Therefore, their clinical value should be interpreted cautiously. At present, VR-based exercise interventions may be more appropriately positioned as complementary tools within conventional exercise training, rehabilitation programs, or comprehensive behavioral interventions to enhance participation and training experiences rather than replace real-world social interaction and exercise training.

This review also provides several implications for future research. First, future studies should further standardize the reporting of VR interventions, including key parameters such as immersion level, feedback format, exercise task structure, social interaction modality, and training dosage, to improve comparability across studies. Second, future research should place greater emphasis on the potential moderating effects of different VR characteristics on intervention outcomes, including differences between immersive and nonimmersive systems, individual versus socially interactive training, and structured motor learning versus entertainment-oriented activities. Third, because most existing studies focus primarily on short-term outcomes, future investigations should incorporate long-term follow-up assessments to evaluate the durability and real-world generalizability of intervention effects. In addition, substantial variability exists in the behavioral, executive function, and motor outcome measures used across studies, highlighting the need for more standardized and clinically interpretable assessment frameworks.

More importantly, future research should move beyond simply asking whether VR is effective and instead investigate which types of VR-based exercise training are most likely to produce beneficial effects, for which subgroups of children and adolescents with ASD, and under which contextual conditions. Compared with simply contrasting “VR versus non-VR” interventions, this more precise and mechanism-oriented research framework may better facilitate the clinical translation and individualized implementation of VR-based exercise interventions in the ASD field.

Several limitations of this systematic review should be acknowledged. First, the overall number of included studies was limited, and some outcome domains included only a small number of studies, which may have reduced the stability and precision of the pooled effect estimates. In particular, the wide CIs and low certainty of evidence for behavioral outcomes indicate substantial uncertainty in the current findings.

Second, the VR-based exercise interventions included in this review demonstrated substantial conceptual heterogeneity. Considerable variability was observed across studies in terms of VR device types, immersion levels, feedback mechanisms, training content, and implementation approaches, all of which may influence intervention effectiveness. Accordingly, pooling heterogeneous VR exercise modalities into a single intervention category may have reduced interpretability and limited conclusions regarding the effects of specific intervention models.

Third, some meta-analytic findings demonstrated substantial heterogeneity and wide prediction intervals, suggesting that intervention effects may vary considerably across research settings. This finding suggests that the observed pooled effects may not be consistently generalizable across all clinical or educational settings. Therefore, the findings of this review should be interpreted as reflecting overall trends within the current evidence base rather than definitive conclusions applicable to all forms of VR-based exercise interventions.

Furthermore, several studies were subject to risks of bias related to randomization procedures, outcome measurement, and selective reporting, while small sample sizes and short follow-up durations further reduced the certainty of the evidence. In addition, most included studies focused on children and adolescents with mild-to-moderate ASD, and evidence regarding the applicability of these interventions to individuals with more severe ASD, different age groups, and diverse cultural backgrounds remains limited.

Finally, owing to the limited number of included studies, this review was unable to comprehensively examine the potential effects of different VR characteristics, training dosages, and participant characteristics through meta-regression or detailed subgroup analyses. Consequently, interpretations regarding differences among various VR exercise modalities should be made cautiously at this stage.

This systematic review systematically synthesized the current evidence regarding VR-based exercise interventions for children and adolescents with ASD. Unlike previous reviews that primarily focused on general VR training, social skills interventions, or single functional domains, this review specifically emphasized VR interventions centered on physical activity participation and simultaneously integrated behavioral, executive function, and motor outcomes, thereby providing a more comprehensive understanding of their multidimensional effects. Current evidence suggests that VR-based exercise interventions may produce relatively consistent positive effects on executive function, whereas their effects on motor and behavioral outcomes remain uncertain and highly context-dependent. Accordingly, these interventions may currently be more appropriately regarded as promising adjunctive tools rather than fully established stand-alone intervention alternatives.

Consistent with these findings, this review suggests that future VR research in the ASD field should move beyond focusing solely on whether VR is used and instead place greater emphasis on how VR interventions are designed, which goals they target, and the mechanisms through which they exert their effects. From a clinical perspective, VR-based exercise interventions may provide new opportunities to enhance training enjoyment, participation motivation, and individualized adaptation to intervention programs. From a research perspective, this review highlights the importance of developing more standardized, mechanism-oriented, and longitudinal study designs. These findings may help facilitate the transition of VR-based exercise interventions from emerging technological applications to evidence-based adjunctive intervention tools.