Applications of Virtual Reality to Support Social Communication in Individuals with Autism: A Systematic Review of Immersive Interventions

Abstract

Virtual reality (VR) has emerged as a promising tool to support social communication in individuals with autism spectrum disorder (ASD). This study presents a systematic review of 28 empirical studies sourced from ERIC, APA PsycInfo, and Scopus. Articles in English published between 2014 and 2015 were included, with the last search being 10 June 2025, that implemented immersive VR interventions for children, adolescents, and young adults with ASD. Following PRISMA guidelines, we analyzed participant characteristics, intervention features, research designs, and reported outcomes with 91.6% IOA. An analysis of the risk of bias was performed using the RoB 2 framework for randomized trials and ROBINS-I for non-randomized studies. Results from the two assessments showed low to significant bias. However, due to the relative novelty of this field of study, all data was deemed valuable and was therefore included in this review. Results show that immersive VR interventions are generally effective in improving skills such as eye contact, emotion recognition, empathy, and conversational abilities, while also being engaging and well accepted by participants. Nevertheless, most studies were limited by small sample sizes, lack of control groups, and scarce evidence for long-term maintenance or real-world generalization. The findings underscore the potential of immersive VR as an innovative and scalable approach for enhancing social communication in ASD, while highlighting the need for more rigorous and longitudinal research. No funding was obtained for this systematic review. No protocol registration was carried out for this review as it was unfunded and exploratory in nature.

1. Introduction

With the rapid increase in technology over the last decade, recent advancements have led to widespread use of digital tools in education, therapy, and communication among all kinds of populations [1]. Among these tools are immersive technologies such as virtual reality (VR). These immersive technologies have gained significant interest for their ability to simulate real-life contexts, support immersive learning, and increase personalized interaction [2,3]. Because of these unique benefits, the acceptance of VR in health, education, and clinics has expanded rapidly, especially for populations with diverse and unique needs [4].

Individuals with autism spectrum disorder (ASD) are one of these populations that could benefit significantly from using immersive technologies. The World Health Organization [5] estimates that about 1 in 100 children have ASD, although statistics vary on the percentage of the population with ASD due to the rising percentage of individuals with ASD. ASD is a neurodevelopmental disorder that is characterized by deficits in social communication, restricted interests, and repetitive behaviors [6]. These challenges impact many aspects of an individual’s life, including their ability to initiate or sustain conversations, interpret non-verbal cues, engage in joint attention, and navigate complicated social environments [7,8,9]. Thus, social communication is a common area of need for many individuals with ASD.

Typical interventions that are used to improve social communication skills in individuals with ASD include Social Skills Training (SST), visual supports and social stories, structured play, and peer-mediated interventions [10]. While these strategies have proven effective in many instances, they are still limited by a lack of teachers and therapists, and individuals can face challenges in generalizing skills they have learned [11,12].

VR is defined as technology that creates a simulated environment which can be explored in 360 degrees. It places the user inside the virtual environment, providing an immersive experience [13]. Head-mounted displays (HMDs) and the Cave Automatic Virtual Environment (CAVE) are examples of equipment that is used to create these immersive VR environments. CAVE is a room-sized immersive VR system where images are projected onto multiple walls, creating a 360-degree virtual environment [14]. HMDs are devices worn over the head that use projection technology integrated into eyeglasses or mounted over a helmet or hat [15]. Immersive VR shows immense potential due to its safe, customizable, and independent learning environments that simulate real-world social interactions while maintaining high user engagement and motivation [16]. This technology allows individuals to practice social skills with less anxiety than in-person situations, to receive immediate feedback, and to learn at their own pace. This can lead to increased confidence, attention, and motivation [17].

VR has been helpful for individuals with ASD due to the novel interventions it can provide. VR has been used to simulate social scenarios, create game-based learning modules, and create repeatable social practice [18]. Although the use of VR is growing in regard to ASD intervention, research is fragmented on the subject. Previous reviews vary greatly in their methodological scope [19] and often employ a very broad purview of the research. Some reviews have explored technology-based interventions for individuals with ASD more broadly, focusing on non-immersive video games, robotics, or computer-assisted instruction [20]. Many reviews fail to differentiate between immersive and non-immersive VR [21]. Some reviews have also focused on the use of VR to teach a variety of skills such as academic or vocational skills to individuals with ASD [22]. However, very few systematic reviews have focused specifically on immersive VR interventions that teach social skills to individuals with ASD. As immersive VR technologies become more advanced, less expensive, and generally more accessible, it is important to understand their effectiveness in being applied to individuals with ASD.

Previous reviews have been able to offer valuable information, helping create a foundation of comprehensive understanding of this relatively new field of study, such as those by Dechsling, (2022) and Thai & Nathan Roberts (2018) [23,24]. Thai & Roberts (2018) [24] offered valuable information on the emerging potential for VR usage as a positive intervention system for individuals with autism, but due to the nascent nature of the field of study, only 12 articles were included, with only 49 initial articles included for screening. We hope to offer a more comprehensive view of the field as the number of studies within the field increases. Dechsling et al. (2022) [23] reported that HMDs had a high acceptance rate among the ASD community and that they were preferred among individuals with ASD. However, they reported that only 16% of studies from their review contained HMDs, making for a narrow view on the effectiveness of such equipment. Due to the review being several years old and great strides being made in accessibility and affordability in recent years for HMDs, our review includes a more up-to-date view on VR research, as our review included studies that had as high as 75% usage of HMDs and involved articles published throughout 2025. In addition to this, Dechsling et al. (2022) [23] reported only 7.4% female participants. Our study included 24.4% female participants, which is more in line with the current autism diagnoses ratio of 4.2 male to female (Zeidan et al. 2022) [25]. This increase in female participants suggests generalizability will be higher with the increase in recent research.

1.1. Rationale for the Study

The rationale for this research is founded in previous work carried out by researchers such as Parsons (2002) and Parsons and Cobb (2011), where they took the controllable environment of VR and decided to see if they can use that control to influence the development of social skills in individuals with autism [26,27]. Their research found that controlling what stimuli the individual is exposed to can allow for easier skill development in that controlled environment, and then over time the environment can be changed to allow for the generalization of the learned skills in the real world. This review can help further understand the ensuing research that has been performed based on these findings.

Based on that rationale, this systematic review seeks to address current gaps by synthesizing existing research on immersive VR interventions designed to teach social skills to individuals with ASD. As immersive VR becomes more accessible due to advances in commercial hardware, software, and less expensive products, understanding its effectiveness in this context is both timely and necessary. Unlike prior reviews that combine diverse technologies or focus broadly on educational or therapeutic outcomes, this review narrows its scope to immersive VR environments targeting social skills specifically. By doing so, it aims to provide actionable insights for clinicians and educators while also highlighting areas for future research and standardization of intervention design.

1.2. Study Objectives

In alignment with PRISMA guidelines, this systematic review aims to achieve the following research questions:

• What are the characteristics and formats of immersive VR interventions targeting social outcomes?
• What are the populations and settings in which these interventions have been implemented?
• What are the reported outcomes and effectiveness of these interventions?
• What are the methodological strengths and limitations of the included studies?
• What are the trends, gaps, and directions for future research?

2. Materials and Methods

This study used a systematic literature review approach using the recommendations of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) framework.

2.1. Eligibility Criteria

In order to be included in this review, the studies needed to meet the following criteria: (a) Peer-reviewed empirical studies; (b) at least one participant diagnosed with ASD; (c) the intervention includes immersive VR (e.g., head-mounted display or CAVE systems); (d) the study targets social or communication skills; (e) the study reports at least one pre-/post-outcome measure related to behavior, cognition, or perception; (e) the sample includes at least one child, adolescent, or young adult with ASD (≤21 years).

Studies were excluded for the following reasons: (a) non-immersive VR only (e.g., desktop games, 2D simulations); (b) studies that used VR only for assessment, not intervention; (c) literature reviews, editorials, or conceptual papers; (d) non-English language; (e) no ASD diagnosis. This exclusion and inclusion criteria were utilized to ensure that the most relevant literature was included in the review and to ensure that irrelevant articles were not included.

Table 1. Exclusion and inclusion criteria.

Exclusion Criteria	Inclusion Criteria
Non-immersive VR	Intervention includes immersive VR
Used VR for assessment but not for intervention	Reports at least one pre/post outcome measure related to behavior, cognition or perception
Literature reviews, editorials, or conceptual papers	Empirical studies
Non-English language	Targets social or communication skills
No autism diagnosis	At least one participant with an ASD diagnosis
	Sample includes at least one child, adolescent, or young adult)

provides a summary of the inclusion and exclusion criteria used in the eligibility process.

2.2. Information Sources and Search Strategy

To identify relevant literature, we systematically searched three electronic databases: ERIC, APA PsycInfo, and Scopus. These databases were selected for their broad coverage of peer-reviewed research in education, psychology, and interdisciplinary fields, and their relevance to topics related to ASD and VR. Additional databases were not included, as these three databases offer a wide scope of empirical evidence and peer-reviewed research that was deemed sufficient for the purpose of this review. Conferences such as IEEE VR were excluded due to limitations regarding pre-publication bias and limitations in data extraction; it is commonly not possible to perform an analysis of the risk of bias on conference presentations. The final search was completed on 10 June 2025. We tailored our search strings for the different databases to verify that the relevant studies were retrieved in each database. The Boolean operators ‘AND’ and ‘OR’ were used to ensure coverage of all relevant articles in the databases. The key terms are summarized for each database in

Table 2. Search terms for the databases.

Database:	Search Terms:
ERIC	(“virtual reality” OR “VR” OR “immersive technology” OR “head-mounted display” OR “immersive learning” OR “augmented reality” OR “AR”) AND (“autism” OR “autism spectrum disorder” OR “ASD”) AND (“social skills” OR “social communication” OR “interpersonal communication” OR “social interaction”)
APA PsycInfo	(“virtual reality” OR “VR” OR “immersive technology” OR “head-mounted display” OR “immersive learning” OR “augmented reality” OR “AR”) AND (“autism” OR “autism spectrum disorder” OR “ASD”) AND (“social skills” OR “social communication” OR “interpersonal communication” OR “social interaction”)
SCOPUS	TITLE-ABS-KEY (“virtual reality” OR “VR” OR “immersive technology” OR “head-mounted display” OR “immersive learning” OR “augmented reality” OR “AR”) AND TITLE-ABS-KEY (“autism” OR “autism spectrum disorder” OR “ASD”) AND TITLE-ABS-KEY (“social skills” OR “social communication” OR “interpersonal communication” OR “social interaction”)

. Additionally, the search format was limited to the years 2014–2025, the language was limited to English, and only articles were selected as the document type.

2.3. Selection Process

The selection process can be seen in Figure 1. Three databases were searched to find articles, with 531 articles identified. A total of 167 articles were from APA Psycinfo, 122 articles were from ERIC, and 242 articles were from SCOPUS. A total of 111 articles were then removed due to being duplicates. The remaining 420 articles were screened by a minimum of two and a maximum of four independent reviewers. The reviewers used the inclusion and exclusion criteria described in Table 1 to review the abstract of each study and determine eligibility. Using this process, 257 articles were excluded due to irrelevance to the study based on the abstract. A total of 163 reports remained, but only 161 full-text reports could be retrieved. These 161 reports were thoroughly assessed for eligibility through a review of the methodology sections of the articles. This process was conducted by a minimum of two independent reviewers. The research team that consisted of 4 independent reviewers met to discuss the ‘grey area’ of articles that were being assessed for eligibility; for example, studies that involved immersive-like VR that had projections on multiple walls of the room but were not specifically a CAVE or utilized a HMD. All reviewers came to a unanimous decision to determine whether these articles were eligible or not. Of the 161 assessed for eligibility, 132 were excluded. A total of 20 were excluded for not having a pre/post measure, 16 were excluded for being non-empirical or not being peer-reviewed, 25 were excluded for only having augmented reality and not virtual reality, 62 were excluded for using non-immersive virtual reality only, and 5 were excluded because there were no participants who were under 21 years old with a diagnosis of ASD. After this process, there were 28 studies to be included in the review and 29 reports of these studies. It is worth noting that “AR” and “Augmented reality” were included in the search criteria and then excluded when a review contained only augmented reality. This is because the application and implementation of AR interventions vary enough from that of VR that we determined it is deserving of its own review, to be carried out at a later time.

2.4. Data-Collection Process

There were three main phases to the selection process. First, we identified the articles by searching the databases, we screened the articles for those that were obviously irrelevant, and then we analyzed the articles for eligibility. Using the search terms outlined above, three databases were searched for articles and 531 articles were found. We used an online website called “Rayyan” that was accessible to all authors of the review to organize the references and identify duplicate records. We reviewed the 531 records independently and identified the records that were eligible and those that were not. There were 29 articles that were found eligible for the systematic review. The articles were assessed for eligibility by independent reviewers, who then had their decision verified by three other independent reviewers to help eliminate the risk of bias.

Afterward, the articles identified eligible were collected. The following variables were identified and analyzed: author, year of publication, country of publication, number of participants, age of the participants, gender of the participants, ethnicity of the participants, diagnosis of the participants, the type of VR device used, the target social skills addressed, the frequency and the duration of the intervention, the setting where the intervention was given, who the interventionist was, a general description of the intervention, the research design, a description of the control group if there was one, what measures were used to collect data, significant findings reported, effect sizes reported, if there was a maintenance or generalization phase, qualitative findings, and limitations reported. All data that was sought for was included in analyzation. Any unclear or missing data is discussed in Section 4.5. The data were coded by two independent reviewers. Interobserver agreement (IOA) was collected on the data collection process. An agreement was scored when entries for coded data were identical or substantively similar. The resulting IOA was 91.6%, indicating the reliability of the data-extraction process. To resolve the 8.4% of disagreements, after all articles were screened, all authors met and discussed discrepancies between reviewers until unanimous decisions were made regarding inclusion or exclusion. An assessment of certainty was not performed during this review.

3. Results

A total of twenty-eight studies, reported across 29 articles, with 2 articles reporting on the same study but not being duplicates, met the inclusion criteria for this review, with Beach & Wendt [28,29] publishing 2 articles based on one study. Together, these studies represent 591 participants with ASD, ranging in age from 2 to 43 years. The majority focused on children and adolescents under 21, with most samples consisting primarily of male participants.

The studies varied considerably in scope and design, including case studies, single-subject designs, quasi-experiments, and a small number of randomized controlled trials. Interventions targeted a wide range of social and communication outcomes, delivered through diverse immersive VR systems such as head-mounted displays (e.g., HTC Vive, Oculus Rift/Meta Quest) and, less commonly, CAVE or low-cost smartphone-based devices.

The following subsections and

Table 3. Summary data of included articles.

Ref. No	Author and Year	Country	Sample Size and Diagnosis	Type of VR Device Used	Target Social Skills	Setting	Research Design	Measures
[30]	Artiran et al., 2024	USA	16; ASD without intellectual disability (ID)	HTC Vive Pro Eye headset	Eye contact and gaze during job interviews	Research lab at a local university	Mixed-methods feasibility study with a within-subjects, pre/post design	(1) Average eye contact duration; (2) Average period of time without eye contact; (3) Percentage of eye contact while listening; (4) Percentage of eye contact while speaking
[28]	Beach & Wendt, 2014	USA	2; ASD	Unspecified HMD	Dealing with conflict, applying for jobs, asking for directions, asking for help	Summer camp and immersive virtual environment at a university	Case study approach using an ethnographic perspective	(1) Interviews with the participants and with their teachers; (2) The Technological, Pedagogical, and Content Knowledge Framework; (3) Social Skill Menu for 14 through Adulthood
[29]	Beach & Wendt, 2016	USA	2; ASD	Oculus Rift	Recognizing when conversation begins to move in an aggressive direction, maintaining eye contact, and initiating conversation	University and Summer camp	Case study approach using an ethnographic perspective	(1) The Technological, Pedagogical, and Content Knowledge framework; (2) Social Skill Menu for age 14 through Adulthood; (3) Performance Improvement/HPT Model. (4) Interviews, reflective journals, and file notes.
[31]	Bozgeyikli et al., 2017	USA	18: High- functioning ASD	VR2200 HMD	Vocational social skills	University lab	Mixed-methods, quasi-experimental design with between-subjects comparison	(1) Custom scoring algorithm per task (0–100 points); (2) Survey on immersion, dizziness, and satisfaction; (3) Job trainer evaluations; (4) Follow-up ratings after 1 month
[32]	Carreon et al., 2024	USA	22; ASD. Additionally, ADHD, DMDD, CP, OCD, ODD, TD, SLD, Anxiety.	Oculus/Meta Quest 2	Expressive communication social skills: instrumental performance, affective expression, paralinguistic signals.	Classroom setting or home	Group experimental design measured with a repeated-measures ANOVA	(1) Clinical Assessment of Pragmatics; (2) Presence scale developed by Witmer and Singer (1998) [33]
[34]	Cheng et al., 2015	Taiwan	3; ASD	HMD- I-Glasses PC 3D Pro	Social understanding and social skills (non-verbal communication, social initiations, social cognition)	A quiet room of the special education school	Single-subject design was combined with a multiple-probe design for all participants.	(1) Social Behavior Scale (SBS); (2) Social Event Cards (SEC)
[35]	Crowell et al., 2018	Spain	25; ASD	Full-body interaction collaborative system and a 6 m in diameter floor-projected digital world	Joint attention, asking for help, collaboration	Controlled lab	Within-subject counterbalanced crossover with free play control and repeated-measures design across 3 sessions	(1) Social initiations; (2) Requesting help and responding to requests; (3) Comments directed toward game characters, parents, the therapist, researchers, and peers.
[36]	Gabrielli et al., 2023	Italy	12; ASD	Oculus/Meta HMD	Sustained attention, selective attention, and inhibition, turn-taking, and collaboration	Cooperative Albero Blu Center	Mixed-methods observational study	(1) Social Skills Questionnaire (SSQ); (2) Teacher form, filled out by therapist; (3) Post-session observations on the targeted behaviors exhibited by participants; (4) Ad hoc user experience questionnaire and semi-structured interview.
[37]	Gayle et al., 2024	USA	3; ASD	Floreo platform used with Google Cardboard VR goggles	Social, communication, and safety skills: safely crossing the street, listener identification of animals given two features, and joining a conversation.	Site where ABA services were provided and the park, the playground and a shopping center.	Multiple baseline across behaviors design	(1) Percentage of correct responses to steps within the task analysis; (2) Indices of happiness; (3) Indices of unhappiness (4) Equipment acceptance; (5) VR session engagement
[38]	Herrero & Lorenzo, 2020	Spain	14; Low or medium severity ASD	Oculus Rift HMD	Communication problems and social communication, including theory of mind, empathy, and emotional regulation.	N/A	Pre/post intervention with randomized control group.	(1) General questionnaire testing social and emotional reciprocity, nonverbal communication, and flexibility to changes; (2) Visual contact data from the VR
[39]	Ip, H.H.S. et al., 2024	China	107; High- functioning ASD	Oculus Rift HMD with touch controllers	Social norms and interaction skills. Executive functioning and daily life skills. Emotional skills	School	A quasi-experimental design and waitlist control design. Pre/post intervention with control group.	(1) Raven’s Progressive Matrices Test; (2) Two subtests, Affective Expression and Social Reciprocity, of the Chinese version of the Psychoeducational Profile Third Edition (PEP-3)
[40]	Jialiang et al., 2021	China	12; ASD	HTC Vive HMD	More demanding social skills, hands-on tidying skills, and physical coordination skills	Classroom	Pre/post intervention with a matched control group	(1) Greeting task scores (out of 20 points)
[41]	Kim et al., 2024	Korea	14; ASD	VIVE Pro Eye VR HMD	Work related social skills- active listening, initiating conversations, not interrupting, informing customers about options, verifying orders, problem solving; perceived self-efficacy	N/A	Within-subjects design, mixed methods approach	(1) Perceived Self-Efficacy for VR Social Skill Training Scale; (2) iGroup Presence Questionnaire; (3) Physiological data including electrodermal activity, heart rate, changes in voice volume, and detection of eye contact
[42]	Kouhbanani et al., 2021	Iran	43; ASD without ADHD or other comorbidities	Unspecified “VR glasses”	Meeting new people, joining a group of friends, consulting with a friend, emotional cognition, social authority, executive performance, and deductive reasoning	Participant’s home	Pre/post intervention with control group	(1) Children Autism Rating Scale II; (2) Vineland Adaptive Behavior Scale
[43]	Kuper et al., 2020	USA	10; ASD	Cardboard VR viewer	Electrical wiring, self-efficacy	Homes or conference rooms at local universities	Within-subjects, pre-post, mixed methods design	(1) Modified New General Self-Efficacy Scale (NGSE); (2) Thematic analyses on responses to open-ended questions regarding their training experiences
[44]	Lee & Chen, 2025	Taiwan	4; High-functioning ASD	HTC VIVE headset	Empathy, perspective-taking, cooperation, verbal communication, symbolic thinking, social storylines	N/A	Multiple-baseline design	(1) Social Cognition test; (2) Behavioral observation and social interaction assessment (based on the Social Responsiveness Scale (SRS))
[45]	Lee & Wang, 2025	Taiwan	4; High-functioning ASD	HTC VIVE headset	Empathy, imagination, symbolic play, social cognition, imagination	School classroom	Multiple baseline across subjects design	(1) Social Cognition Assessment; (2) Subjective behavior observation forms (including items from the Social Responsiveness Scale (SRS)); (3) Parent observations
[46]	Lee & Yang, 2025	Taiwan	8; ASD	HTC VIVE headset	Perspective taking and empathy	Lab setting	Multiple baseline across subjects design	(1) Social Cognitive Assessment (based on Social Stories); (2) Behavioral and Psychological counseling assessment; (3) parent and teacher daily questionnaires and interviews
[47]	Lorenzo et al., 2016	Spain	40; ASD	“Semi cave” and desktop VR setup.	Identification of emotions and correct social responses, empathy	School	Quasi-experimental, pre/post intervention with a randomized control group.	(1) Number of incorrect facial expressions; (2) Interviews with teachers (for generalization)
[48]	Meng & Yeh, 2021	Taiwan	10; High-functioning ASD	HTC VIVE headset	Reduction in inappropriate social behavior, increase in environmental adaptability, self-regulation, communication with others.	Special education counseling room	Pre/post intervention	(1) Elementary and Junior High-School Students Social Skills Behavior scale; (2) Social Skills Effectiveness Survey; (3) Interviews with teachers and parents after the intervention
[49]	Miller et al., 2020	USA	5; ASD and co-occurring language impairments	Google cardboard and iPhone X	Air travel training, functional communication, behavioral readiness for air travel	Research lab setting and San Diego Airport	Within-subjects pre/post design	Air travel questionnaire on a 5-point Likert scale (for the parents)
[50]	Sanku et al., 2023	USA	50; ASD	HTC VIVE Pro headset and E4 empatica wristband	Attention and sustained focus in a classroom setting	Lab setting	Between-subjects pre/post design with control group	(1) Heart rate; (2) electrodermal activity; (3) Eye movements; (4) Pre/post questionnaire about feedback on the game and their overall experience.
[51]	Simões et al., 2018	Portugal	20; ASD and some additional Intellectual Disability (ID)	Oculus Rift HMD	Using the bus system, executive functioning, planning, adaptive skills, stress/anxiety regulation.	Autism center	Mixed-methods quasi-experimental design that combined between-groups comparisons and repeated-measures analysis	(1) Number of steps performed correctly (automatically measured by the game); (2) Accuracy of the debrief describing the process; (3) Task Duration; (4) Anxiety level measured by skin conductance
[52]	Tsai et al., 2020	Taiwan	3; ASD without comorbidities	CAVE automatic virtual environment (CAVE)	Social greeting skills	N/A	Multiple baseline design across single subjects	(1) Social Story Tests; (2) Number of correct responses to facial expressions
[53]	Yeh & Meng, 2025	Taiwan	11; ASD without Intellectual Disability (ID)	HTC VIVE headset	Empathy, how to handle emotional changes in others, how to think calmly and resolve sudden situations.	Special education counseling room and EEG research lab	A quasi-experimental mixed design method with pre- and post-tests for unequal groups	(1) EEG; (2) Elementary and Junior High School Students Social Skills Behavior Scale; (3) Descriptive Statistics for Self-Developed Social Skills Performance Rating Scale; (4) Social Skills Effectiveness survey; (5) Response speed in conversation; (6) Conversation etiquette
[54]	Yu et al., 2024	China	36; ASD	Unspecified HMD	Gaze fixation, eye contact, visual attention to facial cues	Controlled clinical setting at a hospital	Between-subjects design with 3 groups	(1) Subjective questionnaire filled out by parents; (2) Game performance; (3) Gaze fixation analysis algorithm
[55]	Yuan & Ip, 2018	Hong Kong	72; ASD without Intellectual Disability (ID)	CAVE automatic virtual environment (CAVE), also used stereotypic VR goggles	Emotion expression and regulation, social interaction and reciprocity	University	Mixed repeated-measures design with a waitlist control group.	(1) Psycho-educational profile, 3rd edition (PEP-3): Subtests of affective expression and social reciprocity; (2) Qualitative logs from parents and teachers
[56]	Zhao et al., 2020	China	120; ASD	Unspecified VR HMD	Cognitive skills, social communication, decrease in restricted interests, decrease in rigid behavior	N/A	Between-subjects with repeated measures (RCT)	(1) Autism Behavior Checklist; (2) Childhood Autism Rating Scale; (3) Clancy Autism Behavior Scale
[57]	Zhao et al., 2022	China	44; ASD	Unspecified VR HMD	Cognitive skills, social imitation, emotional expression, language understanding, listening to instructions	N/A	Pre/post intervention (RCT)	(1) Psycho-educational profile, 3rd edition (PEP-3)

summarize study characteristics, participant demographics, intervention features, and outcomes, before synthesizing overarching trends and limitations.

3.1. Risk of Bias Analysis

For randomized and randomized-control comparison studies, the risk of bias was assessed using the RoB 2 assessment tool. Using this method, three studies were found to be of low concern for bias with the biggest concern being that sample sizes were small, and blinding was not carried out. For non-randomized tests, the ROBINS-I tool was used. Using this method, 2 showed some concern, 17 showed moderate concern, 8 showed serious concern, and 2 showed critical concern. In this process, two reviewers evaluated each study independently and disagreements were resolved by consensus. When reports were incomplete, “no information” was recorded, and conservative judgments were made.

Analyzing the results, the randomized comparisons received “some concerns” due to the incomplete reporting of allocation concealment and assessor blinding. Most non-randomized comparisons were “moderate” due to bias from non-random allocation and measurement bias from the use of objective indicators such as gaze data. Single-case experimental designs were also mostly “moderate”; however, this was due to small sizes and non-blinded observations. “Serious” and “critical” studies were rated as such due to absent comparators and reliance on self-reported or clinician-reported outcomes. All studies were retained in order to provide a comprehensive synthesis of the available literature, with the understanding that the findings should be interpreted in light of some methodological constraints.

3.2. Basic Information

Selected articles were published from 2014 to 2025, with a large number of articles being published after 2020. This may be due to the increase in the availability of virtual reality technology to the general public. Figure 2 gives a summary of the years the articles were published in. Studies from nine different countries were included in the review. The majority of studies were conducted in the United States (n = 9), followed by Taiwan (n = 7), China (n = 5), and Spain (n = 3). Studies were also conducted in Hong Kong, Iran, Italy, Korea, and Portugal. The studies were relatively well dispersed across the different regions of the world.

3.3. Participant Information

Participants in the studies ranged from 2 to 43 years of age, with the majority of studies focusing solely on participants under the age of 21. The average age of participants was approximately 12. Every study included had at least one participant under the age of 21. Nineteen of the twenty-nine included articles had less than 20 participants, with about half of those articles including less than 10 participants. Two articles contained 21–30 participants, and two articles contained 21–40 participants. Three articles contained 41–50 participants, and three of the included articles contained 51 or more participants. The total number of participants included was 591.

An overwhelming number of the articles included had mostly male participants, as shown in

Table 4. Summary of participant gender.

Gender	# of Articles	Article Reference Number
More male than female	21	[23,24,25,28,29,30,32,34,36,37,38,41,42,44,46,47,48,49,50,51,52]
More female than male	1	[31]
Not specified	7	[26,27,35,39,40,43,45]

. Many of the articles specified that this was reflective of the overall trend in autism diagnoses. Interestingly, one study included a non-binary participant, which may also be reflective of the overall trend in autism diagnoses. Seven of the included articles did not specify the gender of their participants. The total number of participants whose gender was specified was 470 males, 120 females, and 1 non-binary individual.

Twenty-five of the included articles did not specify the ethnicity of their participants. Four articles did specify the ethnicity, and out of the articles that included the ethnicity, the majority of participants were Caucasian (n = 24). Other participant ethnicities included Asian (n = 2), American Indian/Alaska Native (n = 1), and Hispanic/Latino (n = 1). One participant self-reported their ethnicity as “Other/Decline to Answer”.

3.4. Research Design

A variety of different research designs were used in the included studies, including case-study approaches, mixed-methods designs, quasi-experimental designs, single-subject designs, and repeated-measures designs. Most studies incorporated a within-subjects, pre/post-test design, while others used a multiple-baseline design or incorporated a control group. The variety of different research designs reflects the exploratory nature of immersive VR interventions for individuals with ASD. Fifteen of the studies included control or comparison groups, such as waitlist controls, non-immersive technology conditions, or standard care. Only 15 of the 29 articles included a control group, highlighting the need for more rigorous experimental control.

3.5. Intervention Characteristics

The majority of studies took place in a classroom setting (n = 7). However, some studies took place in a controlled lab setting (n = 6), an autism treatment center (n = 3), or the participant’s home (n = 1). One study took place in a hospital setting, and six studies did not specify the setting where the intervention took place. Occasionally, the intervention took place in multiple locations (n = 5). This often included a primary location such as the participant’s house, with supplementary locations such as a research lab, or a generalization location such as the airport or a local public park. The majority of interventions in the selected articles were conducted by researchers, followed by teachers, therapists, medical staff, and parents. Many of the studies included multiple interventionists (n = 12). The most common combination of interventionists was researchers and teachers (n = 7), closely followed by a combination of researchers and therapists (n = 4). This often included the researcher creating and monitoring the training, and the therapist or teacher delivering the intervention with the researcher present. One study mentioned a “professional job trainer”, which was placed into the therapist category. Three studies did not specify who delivered the intervention.

3.5.1. Social Skills Concepts Addressed in the Intervention

There were a wide variety of social skills targeted in the interventions conducted in these articles. Some interventions targeted foundational social behaviors such as eye contact and requesting help, but others focused on more complex social abilities such as empathy, collaboration, and symbolic communication. It is important to note that the majority of studies targeted multiple related social skill behaviors within a single intervention. Emotional regulation was the most frequently targeted skill, appearing in seven studies. This reflects the growing interest in promoting emotional understanding among individuals with ASD using an immersive environment. Several studies also incorporated role-playing, narrative scenarios, or perspective switching to foster the idea of responding with empathy. Other skills that were commonly targeted include maintaining eye contact (n = 3), asking for help (n = 3), initiating conversations (n = 6), dealing with conflicts (n = 3), collaboration (n = 5), and empathy (n = 5). In some studies, the social skills focused on were very specific to the situational skills tested in the intervention. For example, some studies focused on gaze behavior during job interviews, managing conflict situations among peers, vocationally relevant tasks such as cleaning, travel skills, or dealing with bullying. The wide range of social skills targeted reflects the potential of immersive VR to address many social deficits among individuals with ASD.

3.5.2. Type of VR Software/Hardware Used in the Intervention

There was a diverse range of immersive VR systems and software platforms that were used across the including studies, reflecting the rapid increase in VR technology in the last decade. Devices ranged from high-end, fully immersive HMDs to CAVE-based systems and lower-cost smartphone-integrated devices. The most commonly used platform was the HTC Vive, which appeared in nine studies. The Oculus/Meta line of HMDs (e.g., Oculus Rift, Meta Quest) was used in five studies. These devices typically offer features such as motion tracking, eye tracking, or haptic feedback. A CAVE system was used in three studies, and Google Cardboard was used in three studies. A Google Cardboard is a lower-cost alternative to other HMDs because it uses the user’s smartphone as the main display. One study utilized i-Glasses HMDs, five studies used HMDs but did not specify what kind of HMD was used, and one study used mixed systems by combining a CAVE and a desktop VR setup into one fully immersive environment. The vast majority of the studies (n = 21) used some sort of HMD, highlighting the user-friendly, fully immersive design that has become increasingly more accessible as VR advances.

3.5.3. Generalization and Maintenance Phase

Relatively few of the included studies assessed the durability of transferability of treatment effects. A total of 10 out of 28 of the included studies included a maintenance phase to reassess the performance of the targeted social skill following the intervention after a period of time. The follow-up assessments generally occurred 1 to 6 weeks post-intervention and often involved repeated administration of the same outcome measures used during baseline and post-testing (n = 8). Two of the studies also used a follow-up survey to assess whether the participant was still using the skills learned in the intervention. Most studies reported that gains in social communication were sustained (n = 6), but one noted that there was a regression of skills or inconsistency in the skill retention over time.

Seven of the included studies included a generalization phase to assess whether the targeted social skills were generalized to a real-life situation. Generalization was assessed in a variety of different ways, including parent or teacher questionnaires (n = 2), direct observation in classroom setting (n = 1), or structured real-life role-play scenarios (n = 4). Of the studies that measured generalizations, it was addressed as a secondary outcome and was not measured systematically. While all studies reported positive outcomes in the generalization phase, only one study had statistically significant results regarding the generalization of targeted social skills to a real-life environment.

3.5.4. Measures

The articles included in this review included many measures, ranging from widely accepted and validated measures such as the Psychoeducational Profile 3rd edition (PEP-3), to more subjective measures such as the percentage of eye contact given by the participant, indices of happiness such as laughing and smiling, and interviews with teachers and parents. These measures can be sorted into measures that quantify the social skills that the intervention was trying to improve, and measures asking about the qualitative feelings that participants or their parents and teachers had about the VR. The most common measure was observational measures within the VR such as average eye contact duration, number of social initiations, percentage of correct responses to certain VR tasks, and number of correct responses to facial expressions (n = 11). Other measures included subjective interviews with teachers, parents, or therapists (n = 10), subjective interviews with the participants about their experiences in the VR (n = 5), presence questionnaires (n = 2), the Childhood Autism Rating Scale (n = 2), the PEP-3 (n = 2), and the Social Responsiveness Scale (n = 2). Some studies also used physiological measures such as EEG data, heart-rate variability, and electrodermal activity (n = 4). It is important to note that most articles incorporated multiple measures, and that the vast majority of these measures were empirical. The measures that were not empirical were seeking to gather subjective data from participants and their parents and teachers. For a full list of measures, please see Table 3.

3.6. Conclusions of the Studies

There were many significant findings in the selected articles. The vast majority of the articles found that VR training helped their participants improve on the selected social skills at a significance value of p < 0.05 or greater (n = 20). A list of the articles that included p-values can be found in

Table 5. Summary of articles with included p-values.

Article	p-Values
[26]	Vocational skill modules, p = 0.01. Effect of distracters, p > 0.05.
[27]	Instrumental Performance subtest p = 0.75. Total Pragmatic Performance score, p = 0.01. Relationships between the social skills learning and the levels of presence, p > 0.05.
[29]	Pico’s Adventure: Overall social initiations (1) p < 0.05, and integrated requests (1) p < 0.05. Total social acts (2), p = 0.64, and total social acts (3), p = 0.39. Social behaviors between peers (4), p < 0.05. Lands of Fog: Social initiations—Sessions 1 vs. 3, p < 0.05; Sessions 1 vs. 2, p = 0.12; Sessions 2 vs. 3, p = 0.058; Total social acts—Sessions 1 vs. 3, p < 0.05; Session 2 vs. 3, p < 0.05; responses to partner, p < 0.05; collaborative actions, p < 0.05; parent-reported interaction, p < 0.05.
[30]	Positive social interaction and communication, p = 0.008. Social initiations: spontaneous, p < 0.001; prompted, p < 0.001. Social responses: spontaneous, p < 0.001; prompted, p = 0.003.
[34]	Intervention group, p < 0.001. Control group, p = 0.226.
[36]	Perceived self-efficacy, p = 0.02.
[37]	Social skill scores, p < 0.001; between pre-test and 3-month follow-up p < 0.001.
[38]	Perceived self-efficacy, p < 0.001.
[39]	Accuracy rate, p < 0.05. Maintenance phase, p < 0.05. Behavioral effectiveness: average scores, p < 0.05; Maintenance phase, p < 0.05.
[40]	Performance, p < 0.05. Maintenance stage, p < 0.05.
[41]	After the intervention, p < 0.05. Maintenance phase, p < 0.05.
[42]	Experimental group, p = 0.00001. Control group, p = 0.37.
[43]	Social skills, p = 0.039.
[44]	p = 0.0625.
[46]	Game, p = 0.01. Debrief, p = 0.02. Task duration, p = 0.02. Number of steps, p = 0.1. Anxiety levels, p = 0.11.
[47]	All three scores, p < 0.05.
[48]	Inappropriate behaviors, p = 0.022. Social skills, p = 0.624. Help-seeking behaviors, p = 0.044.
[49]	After the intervention, p = 0.010. Game performance, p = 0.001. Gaze fixation, p < 0.001. Body fixation, p = 0.141.
[50]	Children scored higher on emotional expression and regulation after the VR training (p = 0.037), and higher on social interaction and adaptation after the VR training (p < 0.0005).
[51]	Autism Behavior Checklist, p < 0.05. Childhood Autism Rating Scale, p < 0.05. Clancy Autism Behavior scale, p < 0.05.
[52]	After the intervention, p < 0.02.

. Five studies provided objective findings but did not run formal statistical tests on them. Two studies did not report any objective findings, instead relying on subjective interviews with the participants and their teachers. Additionally, the selected articles used many different types of statistical tests including t-tests (n = 14), ANOVA (n = 8), ANCOVA (n = 2), one-way ANOVA (n = 1), chi-square tests (n = 2), the Mann–Whitney test (n = 2), Kolmogorov–Smirnov tests (n = 2), the Wilcoxon signed-rank test (n = 1), the Kruskal–Wallis test (n = 1), Cronbach’s alpha (n = 1), Fisher’s exact test (1), and a nonparametric sign test (n = 1).

Many of the selected articles did not include effect sizes (n = 23). The studies that included effect sizes typically used one effect-size measure such as partial eta squared (partial η²; n = 3), Cohen’s d (n = 1), and Pearson’s correlation coefficient (r; n = 1). One study included both partial eta squared and Cohen’s d, both of which indicated a moderate effect size. Four of the studies reported a large effect size (d = 0.96, r = 0.846, partial η² = 0.86 to 1.99). Two of the studies reported a moderate effect size (partial η² = 0.047 to 0.05).

3.6.1. Qualitative Findings

Each of the 28 studies included in this review reported qualitative findings that provide meaningful insights in addition to the quantitative findings of the studies. A consistent theme across studies was high engagement and motivation. A total of 21 of the 29 articles described the participants as attentive, curious, or eager to participate. The immersive and interactive features of VR, especially with the game-like formats, seemed to play a big role in sustaining attention and encouraging involvement, even among participants who typically struggle to engage in more traditional interventions.

Several studies (n = 5) described the VR environments as realistic or lifelike, while two of the studies specifically mentioned that participants experienced a strong sense of immersion or presence. These qualities were interpreted by researchers as contributing to more natural reactions during virtual reality and helped to make the experience more authentic and meaningful.

In four studies, the VR environment was described as a safe, low-stress space for practicing socially demanding skills. Participants were observed as being more relaxed and comfortable in the VR environment compared to real-life situations. The structure and predictability of the VR setting may have helped to reduce anxiety and create a space where participants are more likely to take social risks. Similarly, two studies reported increased confidence or self-efficacy among participants, particularly with tasks like initiating interactions, having conversations, or problem-solving. These changes were observed instead of directly reported.

Caregiver and teacher reports made up a significant portion of the qualitative findings reported. Eleven studies included feedback from observers who noted positive changes outside of the VR environments. The changes mentioned include improved eye contact, emotional expression, responsiveness, and interaction with peers, often in natural settings like classroom or homes. These observations offer promising signs of generalization outside of the VR intervention.

Nine studies described improved social or emotional responsiveness during the intervention itself. Participants were observed using more eye contact, responding to emotional cues and engaging more naturally with virtual peers or avatars. Two studies also mentioned that participants enjoyed the in-system feedback or prompts. This was specifically noticed in younger participants who benefited from more structured guidance.

While most feedback was positive, four studies reported negative experiences. This included boredom, frustration, or confusion, specifically in response to repetitive content, technical issues, or limited interactivity. These challenges highlight the need for creating well-designed, responsive VR content that is tailored to the user. While mixed, the qualitative findings suggest that participants not only generally enjoyed the experience but often appeared to benefit in ways that extended beyond the VR environment.

3.6.2. Limitations Reported by the Studies

All 29 included studies self-reported limitations in their studies. The most commonly cited issue was small sample sizes, mentioned in 26 studies. Most samples ranged from 2 to 30 participants, which significantly limits statistical power and generalizability. Authors frequently acknowledged that larger and more diverse samples are needed to confirm the findings observed in these studies. Limited generalizability was another frequently reported limitation, reported in 13 studies. In many cases, the studies only tested individuals with high-functioning ASD, within narrow age ranges, or in controlled settings. These studies pointed out that their findings may not apply to individuals with limited verbal abilities, lower-functioning ASD, or those from different cultural backgrounds.

Additionally, a lack of follow-up data or maintenance assessments was reported in nine studies. Without long-term data, it is unclear whether social gains achieved through intervention are retained, making it difficult to know the real-world utility of the interventions. Three studies also acknowledged the absence of a control group, and three studies also reported a lack of random assignment. This introduces potential bias and a lack of causal conclusions. These limitations are consistent with the field’s development stage, where experimental control tends to be overlooked in exploratory research.

Short session durations or brief interventions were reported as a limitation in three studies. Several authors suggested that extended or repeated sessions might yield stronger and more durable results. Another common limitation was the use of non-standardized or researcher-developed outcome measures. Three studies reported that their data measures lacked validation and were created specifically for their intervention. Two studies also described technical challenges, such as discomfort with the HMDs or software bugs.

Other limitations included limited representation, such as samples limited to high-functioning ASD (n = 4), indirect or subjective assessment methods (n = 3), and potential novelty effects, where the initial enthusiasm about VR may have temporarily inflated participant engagement or performance (n = 3). A few studies (n = 2) also noted interface design or feedback issues such as unclear prompts or disengaging repetition. One study reported the use of snowball sampling as a limitation, and another study lacked behavioral validation for outcomes. Two articles also reported that demographic information such as gender data was missing. These limitations highlight the need for more rigorous and scalable research in the future.

4. Discussion

Drawing on 28 empirical studies, this systematic review aims to synthesize and evaluate the use of immersive VR interventions designed to support social skills in individuals with ASD. This review focuses specifically on five research questions that address trends in intervention design, participant characteristics, reported outcomes, methodological rigor, and emerging directions for future research.

4.1. RQ1: What Are the Characteristics and Formats of Immersive VR Interventions Targeting Social Outcomes?

The first question assesses the characteristics and formats of immersive VR interventions for targeting social outcomes. The results of the review show that immersive VR interventions for individuals with ASD vary greatly in terms of technology use, instructional structure, and intervention strategy. Most studies used HMDs, and most commonly the Oculus Rift and the HTC Vive, to deliver immersive 3D social scenarios. A small subset of studies used CAVE to deliver interventions [52,55]. The variations in delivery format may significantly improve the way social content is structured and engaged with. Similarly, intervention lengths and frequency also ranged greatly across the sample. Some studies such as Sanku et al. (2023) used few intervention sessions, while others used multi-week interventions [50]. For example, Zhao et al. [57] implemented a 12-week training protocol with repeated exposures. Few studies justified their intervention length and frequency choices with theoretical rationale, highlighting a gap in design standardization in the studies. This has implications on both the outcome strength and generalizability of the studies, as the different lengths of the studies may not be comparable in effect or feasibility.

The instructional modality of the interventions also varied greatly. Some studies used automated systems with embedded prompts and gamified feedback [41,50], while others, such as Cheng et al. [34], incorporated teacher facilitation. Notably, Crowell et al. [35] introduced collaborative multiplayer VR, an innovative format that encouraged real-time peer interaction. Some studies, like Gabrielli et al. [36] and Gayle et al. [37], embedded reward structures within the environment to sustain engagement.

Targeted social skills spanned from foundation behaviors such as eye contact [30] to higher-order capacities like perspective-taking [46] and empathy [44]. While this range demonstrated the flexibility of VR technology, it limits consistency across designs and only a few studies [34,52] grounded skill selection in developmental theories or clinical frameworks.

The wide range of intervention characteristics shows a clear strength in the adaptability of VR but also leads to significant discrepancies in implementation. Studies like those by Zhao et al. [57] and Gabrielli et al. [36] offer promising models that include integrated, feedback-rich, multi-session interventions, but the overall scope of studies still lacks agreement on which instructional components are the most essential for achieving meaningful and generalizable gains.

4.2. RQ2: What Populations and Settings Have Been Used in Immersive VR Interventions Targeting Social Outcomes?

The vast majority of the interventions included in this review were developed and tested with children and adolescents with high-functioning ASD [31,44,45]. While this focus is understandable given the need for verbal comprehension and the ability to use the VR hardware, it significantly narrows the population represented. Inclusion criteria throughout studies often excluded participants with co-occurring intellectual disability, limited verbal communication, or sensory processing differences [48]. As a result, the interventions are mostly reflective of a subset of the ASD population and are not reflective of all children and adolescents with ASD.

Age was another notable constraint. Most studies focused on participants between the ages of 6 and 18, with very few interventions targeting toddlers, older adults, or individuals in early adulthood. One exception was Kuper et al. [43], which targeted adults with ASD in a university setting and examined change in social self-efficacy following VR-based interview practice.

The studies also lacked geographic diversity. They were conducted primarily in high-income countries with access to advanced VR hardware such as the United States, Taiwan, China, and select European nations. This lack of geographic and socioeconomic diversity has implications for the scalability and accessibility of VR interventions. High-end VR devices like those represented in these studies are often costly and countries with lower incomes may struggle to gain access to these devices.

One of the most important omissions across studies was a lack of diversity in participant demographics. Very few studies reported on race, ethnicity, or socioeconomic status, and when gender was reported, most samples were over 80% male [28]. This imbalance is expected given ASD’s diagnostic trends but still underscores the need to explore how gendered socialization or cultural communication norms might interact with social VR training. Furthermore, none of the studies disaggregated outcomes by demographic subgroup, which limits conclusions about intervention equity or responsiveness to individual needs. Taken together, these findings suggest that while immersive VR interventions have been tested in a range of formal settings, they have not yet reached the breadth of participants or contexts needed to support widespread adoption. The strong emphasis on high-functioning, verbal children in high-resource environments reflects a pattern of limited inclusivity. As immersive VR becomes more portable and affordable, researchers should prioritize extending access to underrepresented groups, including those with higher support needs, non-English-speaking populations, and diverse learning profiles.

4.3. RQ3: What Are the Reported Outcomes and Effectiveness of Immersive VR Interventions for Social and Communication Skills in ASD?

Nearly every one of the twenty-nine studies reviewed reported positive outcomes following immersive VR interventions, particularly in domains such as emotion recognition, perspective-taking, gaze behavior, and conversation skills. These improvements were generally measured using a combination of standardized assessments (e.g., CARS, ABC, CABS), task performance scores, and qualitative feedback. However, the depth and quality of these outcomes varied widely, and a closer examination reveals both promising trends and substantial methodological gaps that affect how confidently these results can be interpreted.

In total, 27 studies reported statistically significant gains on at least one outcome. These gains were typically seen immediately post-intervention and often focused on skill acquisition rather than long-term application. For instance, Lorenzo et al. [47] found improvements in emotion identification and affective expression after participants engaged with a symbolic play-based VR environment. Sanku et al. [50] demonstrated increased attention to social stimuli through biometric data—including heart rate and eye tracking—after a series of VR classroom sessions. Similarly, Yu et al. [54] found that children showed enhanced gaze fixation during a hide-and-seek VR task aimed at improving joint attention.

A smaller subset of studies used more comprehensive data sources to validate gains. For example, Kim et al. [41] evaluated workplace social readiness using physiological monitoring, self-reported confidence, and third-party behavioral observations. Other studies, like Miller et al. [49] and Yeh & Meng [53], also used multiple forms of data to evaluate social outcomes and user experience. These studies reflect a trend toward more multi-dimensional outcome assessments that go beyond pre/post comparisons.

However, the field still struggles with consistency in measurement. Only six studies reported formal effect sizes, and fewer than half used a control or comparison group. Many studies relied on simple pre/post comparisons without accounting for maturation, novelty effects, or external influences. For example, Gabrielli et al. [36] found increased social engagement after a single VR session but did not include a control group or conduct any follow-up. These kinds of designs, while useful for pilot testing, make it difficult to draw strong conclusions about effectiveness.

One of the most methodologically rigorous studies was conducted by Kouhbanani et al. [42]. This randomized controlled trial included a waitlist control group and measured pre/post outcomes on social reciprocity and verbal responsiveness, with statistically significant gains reported in the intervention group. Importantly, the gains were retained at a 4-week follow-up, making this one of the few studies to demonstrate both short-term and sustained effects.

Unfortunately, few studies examined whether skills generalized beyond the VR context. Only seven attempted to assess real-world applications using parent or teacher feedback. For example, Miller et al. [49] reported that participants were better able to navigate an airport environment after VR-based travel training, but this was based solely on caregiver interviews. Lorenzo et al. [47] noted increased peer interaction at school but did not use structured follow-up to validate these claims. The lack of standardized generalization probes leaves open questions about whether skills practiced in VR transfer to daily life.

Similarly, only 10 studies addressed maintenance of gains over time, and most only assessed outcomes 1–6 weeks post-intervention. Among them, Kouhbanani et al. [42] again stood out as one of the few to show statistically significant skill retention at follow-up. The absence of longer-term data is a notable gap, especially given the time and repetition typically required for social skills to become generalized and automatic.

On the qualitative side, many studies described increased confidence, motivation, and emotional comfort among participants. In 11 studies, caregivers or educators observed improved behaviors such as verbal initiation, emotional awareness, and peer engagement. While these findings are subjective, they provide important insights into the perceived effectiveness and acceptability of immersive VR—especially when these reports are consistent with quantitative measures.

That said, most studies did not closely examine implementation fidelity or isolate which components of the VR experience (e.g., interactivity, repetition, feedback) were most impactful. Without this information, it is difficult to determine exactly what is driving positive outcomes.

In summary, while the evidence suggests that immersive VR can support improvements in key social skills for individuals with ASD, many questions remain about the durability, generalizability, and underlying mechanisms of these effects. Studies like Kouhbanani et al. [42] and Kim et al. [41] provide promising examples of multi-informant, rigorously designed interventions, but the field as a whole still lacks a foundation of replicated, high-quality trials with consistent outcome reporting. More standardized and longitudinal research will be essential to move from early promise to real-world application.

4.4. RQ4: What Are the Methodological Strengths and Limitations of the Included Studies?

While most studies reported positive short-term outcomes and were grounded in creative and well-intentioned designs, significant inconsistencies in research rigor, transparency, and evaluation practices were evident. One of the most consistent methodological limitations across the reviewed studies was small sample size. Of the 29 studies, 26 included fewer than 30 participants, and many used single-case, case series, or small-group designs. While small samples are often acceptable in pilot research or early-phase design studies, they reduce statistical power, increase the risk of Type I and II errors, and limit the generalizability of results. A related concern was the frequent absence of control or comparison groups. Just 15 studies incorporated any form of control, and only a few [50,57] used randomization. The lack of experimental controls makes it difficult to isolate the effect of the VR intervention from other variables, such as developmental maturation, novelty effects, or general therapeutic exposure.

Another methodological challenge was the limited use of standardized, validated outcome measures. While some studies employed well-established tools like the Autism Behavior Checklist or Childhood Autism Rating Scale, many relied on researcher-developed rubrics or game-based performance metrics that were not norm-referenced or externally validated. For example, Bozgeyikli et al. [31] reported gains using task-based social scores within VR job skills modules, but did not provide any standardized comparison or external corroboration. Without common outcome metrics, cross-study synthesis becomes difficult, and the robustness of findings is harder to assess.

The follow-up and generalization data were also sparse. As noted in the previous section, only 10 studies assessed maintenance, and only 7 reported on generalization to real-world settings. Even among those, the methods were inconsistent, ranging from unstructured caregiver interviews to basic teacher checklists. Few studies included direct behavioral probes, momentary assessments, or structured follow-up tasks. This makes it difficult to assess the functional value of VR-acquired skills beyond the digital context.

Despite these limitations, there were notable methodological strengths in a handful of studies. For example, Kim et al. [41] used a mixed-methods design that triangulated biometric data (e.g., heart rate variability) with performance metrics and participant self-reports. Kouhbanani et al. [42] conducted a well-powered randomized controlled trial with pre-registration, a control group, and a 3 month follow-up—an example of methodological rigor that is still rare in this field. These studies demonstrate that high-quality, immersive VR research is feasible and scalable, especially when paired with strong theoretical frameworks and interdisciplinary design teams.

Another strength found in many studies was the integration of social validity measures. Participants, caregivers, or educators were often asked to comment on the relevance, comfort, and usability of the VR tools. While often informal, these data provide critical insight into intervention acceptability, which is a necessary component for future implementation in schools, clinics, or homes. In sum, while early-stage research on immersive VR for ASD has shown creative and promising approaches, methodological inconsistencies remain a major barrier to broader impact. The field would benefit from larger, more representative samples, the use of standardized outcomes, and stronger experimental designs.

4.5. Time-Related Analysis

While 11 years of research does not necessarily seem like much in many fields of study, in VR-based research it is almost the entire history of this field. With so much development in such a short time, it is worth discussing the studies included in this review from the perspective of a time-related analysis. Over the 11 years, 8 studies were performed before 2019, with the remaining 21 studies carried out in 2020 or later. Prior to 2019, the most common equipment used was first-generation commercial equipment which was often very expensive, or lab equipment. These devices had limitations such as not offering eye-tracking technology or being difficult to gather anything more than proof-of-concept results. Studies performed after 2019 started using more advanced commercially available technology such as HTC Vive and Meta Quest devices, which offered eye-tracking technology, more realism, and more programmability for researchers to utilize the advances in technology. These devices also came with a significant price reduction, from USD 499 down to USD 199 for first-generation to second-generation Quest products, allowing for less of a financial barrier for research.

Beyond the technological aspect, studies also indicated a change in intention and design over the 11-year span. Pre-2019 research was composed of studies designed around immediate skill gain and involved the smallest population sizes of any study included in this review. Post 2019, standardization was started, with reviews starting to use SSQs and SRSs to determine social skill improvements along with subtests such as the Affective Expression and Social Reciprocity subtests found in the Chinese version of the Psychoeducational Profile Third Edition. This standardization in the collection of data allows for less bias within data analysis and results and allows for easier generalization of the results as well. This trend gives a positive outlook on the future of educational VR studies in relation to individuals with autism.

4.6. Limitations

While this review offers valuable information on how VR can support social communication in individuals with autism, it is not without its limitations. First, only English articles were searched for and included in this article. While some articles from non-English speaking nations were included, if they were not translated into English, they were not included. Considering only 9 of the 29 articles included were completed in a nation with its national language as English, it is safe to assume that had the search included other languages, more than 29 articles would have been able to be included.

Second, this review only focused on VR research with individuals with ASD. In excluding other disabilities and disorders from the search, information regarding VR’s effectiveness in supporting the development of those with disabilities could have been left out of this review. Additionally, the individuals with ASD were most often reported as having high-functioning autism, which adds further difficulty to generalization in regard to the effectiveness of VR interventions with individuals along the entire autism spectrum.

Third, many of the studies in this review did not account for potential differences in results across gender or ethnicity, which might decrease the generalizability of the findings. Only 29 of the 591 participants had their ethnicity noted and, additionally, 420 of the 591 participants were male, which gives limited demographic perspective and may require future research to be to be, specifically noting potential changes in results with changes in demographic diversity. In regard to the male and female participants, the review did properly portray the current accepted ratio of male-to-female individuals with autism (4:1) [25], but due to the large variance within this accepted number, further research focusing more on female-specific generalization is recommended.

Fourth, studies that focused on the potential benefits of augmented reality (AR) were excluded from this review. Augmented reality can be defined as a technology that can superimpose computer-generated graphics onto a user’s view of the real world via an electronic device such as a handheld device or a head-mounted display. This additional immersive technology could additionally provide numerous benefits for individuals with ASD to develop delayed skills such as communication or social skills. Additionally, non-immersive forms of VR were also excluded, highlighting the same potential limitations as augmented reality being excluded.

Fifth, it is possible that eligible studies were missed due to potentially narrow search criteria. While every effort was made to ensure that all relevant studies were collected, there is a chance that some were missed. Similarly, only published peer-reviewed articles were included, which may have excluded some studies that were relevant to the review.

Sixth, there were sometimes significantly varied research methods used that resulted in serious or critical bias. Due to this, generalizing the findings with future research could be difficult, and standardization of research methods and models is recommended.

Another limitation to consider is possible reporting bias due to the mostly positive results. It is also important to consider clinical versus practical significance when looking at the selected articles. The evidence base likely overestimates benefits of immersive VR for social communication due to publication and reporting practices common in emerging fields. Exclusion of non-English and gray literature may further amplify positive results. When taking these factors into account, the uncertainty around the true magnitude of effect is increased and the precision in estimating population-level impact is limited.

4.7. Practical Implications

Virtual reality can be a useful tool to teach social skills in classroom, home, and clinical settings to children, youth, and young adults with ASD. These social skills include making eye contact, asking for help, initiating conversations, environmental awareness, showing joint attention, being empathetic, maintaining social norms, and taking the perspective of others. Virtual reality may be used by teachers, parents, or therapists as an integrated way to teach these social skills in their lesson plans, home life, or sessions. Individuals with ASD tend to enjoy using VR, which helps them stay engaged and focused as they learn, and makes it more likely that they will absorb and use the skills in their daily lives. In the future, it may become a normal part of therapy or treatment for ASD individuals.

4.8. Suggestions for Future Research

Research studying the usage of virtual reality to improve social skills in adolescents with ASD has been detailed, but there are still ways in which future researchers could fill gaps and provide more information to this field. One way would be to study the long-term effects of VR on individuals with ASD. Using longitudinal study designs and measuring the maintenance of VR would be useful information to provide. Another area where research in this field could use more support is in the size of the studies. Most studies include group sizes of less than 15, so researchers should consider including larger groups of participants. A large number of studies include only participants with high-functioning ASD. For more generalizability, future researchers should include participants with low-functioning ASD that are non-verbal. This, in many cases, would require researchers to find new ways to measure social cues, but would be of great benefit to those who are often not included in these studies. Finally, future research would benefit from focusing on the effects using VR in different settings. From being in educational settings to being at home, studies in the future should include how being physically or virtually in different settings may change results for individuals with ASD.

5. Conclusions

More and more researchers are beginning to explore immersive virtual reality as a way to develop and maintain the social skills of individuals with ASD. This systematic review used PRISMA guidelines to evaluate existing research on immersive VR interventions and described the effectiveness of VR interventions. Immersive VR interventions were mostly found to be safe, effective, and engaging for individuals with ASD and will likely continue to be used in the future to improve the social skills of individuals with ASD, though due to the lack of standardization within VR implementation, future research to confirm these findings is recommended.