1 Introduction

Within the wide range of people included under the Special Educational Needs umbrella, autism is a group that is increasing. This condition, according to De Luca et al. (2021), has an estimated worldwide incidence of approximately 1%, while other research has indicated that one of 44 of all 8-year-olds have autism (Maenner et al., 2021). The APA (2013) indicates that autistic individuals need support related to challenges in the social, communicative, and behavioral domains. According to Schmidt et al. (2021a, b), difficulties in the social, communicative, and behavioral domains can lead to issues with adaptive skills. These skills consist of the ability to respond to the demands of the environment. Similarly, Moon and Ke (2021) state that autistic students manifest some difficulties in the process of identifying variable social cues in the different contexts they face. In this regard, Newbutt et al. (2020) suggest that autistic people are impacted by the way people relate to the environment through perception, communication, and interaction. Moreover, Lorenzo et al. (2016) claim that autistic students are characterized as having visual preferences and extremely specific learning needs in addition to logical but somewhat abstract reasoning. It is this preference for visual learning that has fostered the use of new emerging technologies such as Virtual Reality by researchers and autistic groups (Lorenzo et al., 2016).

Virtual reality (VR) can be defined as a digital simulation of a three-dimensional environment (Mosher et al., 2021). For Mosher and colleagues, these environments allow a user to physically interact with the technology in a way that creates an atmosphere that resembles the real world. In addition to this, Mak and Zhao (2020) explain that VR allows the user to feel a sense of immersion with the environment. Moreover, users receive feedback from the environment through image, sound, and tactile manipulation. Consequently, VR allows autistic students to practice interactions and behaviors from a realistic environment avoiding overexposure and undesirable sensory and social inputs (Bradley & Newbutt, 2018; Didehbani et al., 2016). In addition, VR can provide an opportunity for autistic students to learning in a safe and predictable manner; without implications of real-world consequences. As such, Bozgeyikli et al. (2018) outline some of the features that enhance the use of VR for autistic people, including: predictability, customizable task complexity; control; realism; immersion; automation of feedback, assessment, reinforcement. Based on the potential of VR for use with/by autistic students, research projects have been developed in a range of areas and for a range of outcomes/benefits. For example, traveling by bus and participating in a coffee shop (Mitchell et al., 2007), emotion recognition (Lorenzo et al., 2016), practicing public speaking (Jarrold et al., 2013), learning social competence (Wang et al., 2017), calming a friend who had lost a pet, or a brief conversation with classmates (Didehbani et al., 2016), feeling of presence in a place (Wallace et al., 2010), and crossing a road safely (Strickland et al., 1996).

Previous systematic reviews and research have been conducted in this area (i.e. Mesa-Gresa et al., 2018). Mesa-Gresa and colleagues analyzed a total of 31 papers through an evidence-based systematic review including clinical and technical databases on the effectiveness of VR-based interventions when used with autistic groups. The results highlighted that the average sample size is 20 (autistic participants), and the average age of the participants is between 8–14 years old. In addition, social and emotional skills were the most researched/focused areas. One of the limitations of this study is the absence of an educational focus. In another study Bradley and Newbutt (2018) conducted a systematic review on the use of head-mounted display (HMD) devices with autistic students over the period 1990–2018. They only located six papers that met their criteria (autism, education, HMDs). The results indicated an absence of control groups in comparing the findings in addition to a lack of details related to ethical and safety practices. In the same sense, there is an absence of methodologies that enable follow-up in real environments such as schools in addition to including autistic people in the research. Along similar lines of enquiry, Glaser and Schmidt (2021) implemented a systematic review to better understand the present design characteristics of virtual reality (VR) systems designed as training tools for individuals with autism. Their results show that most of the interventions have been developed with desktop VR. This has led the authors to conclude that VR is not being fully exploited or used in ways researchers have previously proposed. Similarly, the environments analyzed were fictitious and did not match real situations. The average age of the participants was between 0–9 years. Most recently, in the work of Dechsling et al. (2021), 49 documents were located and analyzed. These focused on the use of virtual reality and augmented reality technology in social skills interventions for autistic individuals. The results indicate that the average number of participants was 13, with desktop VR being the most used. Social interaction and attention were areas that received the most research.

With respect to previous research, here we propose several advancements to what has gone before. Firstly, we have extended the study period from 1996–2021. This will result in a larger number of documents being obtained, and the representativeness of the results should be greater. Secondly, we also included conference papers, so that emerging topics in the area of our study can be detected. This aspect is not included, for example, in the works of Bradley and Newbutt (2018) or Glaser and Schmidt (2021). Thirdly, none of the previous research has analyzed the instrument that is most suitable for assessing the improvements that occur in autistic students after the application of VR. This question is fundamental if we are to determine and incorporate this tool in educational settings. Another of the strengths, is that in this study, unlike others such as those of (Mesa-Gresa et al., 2018), the subject matter is exclusive; to this end, the documents have been filtered by the categories Education and Educational Research and the category Education Science Discipline. In addition, an inclusion criterion has been included so that only documents that work on VR in education are analysed. In line with the contributions of this review, the search terminology used is exclusively VR, unlike the works of (Mesa-Gresa et al., 2018) or Dechsling et al. (2021), which use search terms related to augmented reality. Another study variable that has not been included in previous research has been the type of activity. In other words, what are the characteristics of the activity that autistic students are going to carry out. Only Glaser and Schmidt (2021) have some similarities since they analyse the objectives of the activities.

Despite the growth and rapid development of virtual reality applied to autistic children, very little is known about the type of technologies being used/applied in schools, what the focus of these technologies are, how these technologies are measured (instruments used), or the key findings from work in this area to date. As a result of these gaps in knowledge, the field remains unable to provide clear guidance to the autistic community, or a range of stakeholders (i.e., teachers, parents) on the best way(s) to utilize VR and HMD-based materials in educational settings. Based on the gaps in this field, the key aim of this study is to conduct a systematic and thematic review on the application of VR in teaching and learning environments of autistic students during the period 1996–2021. A series of research questions were established to guide our enquiry and systematic review.

  1. 1.

    What type of virtual reality has been used most frequently in research with autistic students?

  2. 2.

    What are the research objectives that have been most often worked on for implementing virtual reality for autistic students?

  3. 3.

    What type of instrument has been the most employed by researchers in their interventions for autistic students?

  4. 4.

    What are the characteristics of the activities designed for the intervention of autistic students using VR?

  5. 5.

    What have been the most frequent results achieved in the implementation of VR for autistic students?

2 Method

2.1 Approach

To address the research questions, it was decided to implement a systematic and thematic review (STR). There are several reasons that justify the use of this methodology. Firstly, it should be noted that STR makes it possible to synthesize a large amount of information and the identification of areas in which new research can be developed that incorporate methodologies to improve the field of study (Dakduk & González, 2018) Secondly, it allows a quantitative and qualitative approach (Crompton et al., 2021). These authors state that with this methodology it is possible to develop a study by thematic variables and by number of documents. Finally, Cook et al. (1995) suggest that the bias introduced by the researcher in a STR can be eliminated by developing a critical appraisal, synthesis of all relevant studies on a topic, and systematic collection. Beltran (2005) argues that the use of STR is justified because it reveals contradictions in the different studies in aspects, such as results or methodologies. This author concludes that one of the fundamental reasons for using STR is to offer an analysis of the existing evidence on a topic based on a methodical and systematic analysis so that a research question can be answered.

2.2 Instrument

Next, the variables selected for the study are justified and defined. Based on Elo and Kyngäs (2008), Krippendorff (1980), Weber (1990), it was decided to choose two types of variables for the thematic analysis: content and context. The aim of the content variables is to obtain a condensed and accurate description of the phenomenon to be studied, and the result of the analysis are concepts or categories that describe the phenomenon (Elo & Kyngäs, 2008).In this line, content variables seek to classify large amounts of text into an efficient number of categories representing similar meanings Whereas context variables are defined as those aspects that need to be considered to adapt the described phenomenon to different contexts (Elo & Kyngäs, 2008). Therefore, the context variables aims to provide knowledge, new insights, a representation of facts and a practical guide for action. The content and context variables used in the study are defined below.

  • Content

  • Objectives: the purpose of the research.

  • Results: The most frequent findings

  • Type of activity: The nature of the activity developed.

  • Context

  • Type of virtual reality: Type of device used in the support of the autistic person/group.

  • Type of instrument: Tool used to collect information on the evolution of the autistic student.

2.3 Procedure

The first phase of this systematic review involves the development of research questions. For this purpose, the target population of the study was autistic students currently attending an educational stage. Next, the type of support is addressed along with the length, location, and type. In our case we are interested in educational support that took place at the school. Subsequently, reference is made to the comparison where the authors choose the type of interventions to be compared.

For the second and third phases of the systematic review, the following aspects are developed: first, it is necessary to establish the research protocol avoiding bias together with the inclusion and exclusion criteria. Subsequently, the databases are selected, and the search is carried out based on the established criteria and the elimination of those documents that have not been adjusted to the research questions. As for the research protocol, it was decided to extract all available information from the documents such as authors, abstract, title, institution, country, source, origin of the source and references included in the documents. All this knowledge was stored in excel files generated by Scopus and WOS. In addition, a visual review of the documents was carried out to avoid the appearance of publications that did not deal exactly with the subject matter of the study. In the research protocol phase, the following criteria for inclusion and exclusion of documents were established in Table 1. Inclusion and exclusion criteria can be according to Cardona-Arias et al. (2017) of two types: conceptual or related to the research topic, while the second type are linked to operational considerations. For these authors, within the first type, criteria such as temporal delimitation, population characteristics, type of study and area worked on are included. According to Cardona-Arias et al. (2017), the second type may consider the filters of the sources consulted. In this sense, Connelly (2020) assures that the inclusion criteria help researchers to determine whether or not an article should be included in the review sample. The importance of exclusion criteria lies in avoiding the influence of variables in the study. Also, for Connelly (2020), strict inclusion criteria can improve the internal validity of the study. However, too many exclusion criteria can lead to problems in the generalization of the studies obtained. Therefore, it was decided to extract all available information from the documents such as authors, abstract, title, institution, country, source, origin of the source and references included in the documents. All this knowledge was stored in excel files generated by Scopus and WOS. In addition, a visual review of the documents was carried out to avoid the appearance of publications that did not deal exactly with the subject matter of the study.

Table 1 Inclusion and exclusión criteria
Full size table

After establishing the inclusion and exclusion criteria in addition to the research protocol, the search for documents was executed using the keywords determined by the authors. The words were selected from the UNESCO and ERIC Thesauri. These tools collect the most used terms in the most prestigious articles in the field (Haas et al., 2020). We chose terms used in DSM-IV and DSM-V because this would allow us to obtain a larger number of documents. In this way, the terms were

("Autism" or "Autism Spectrum Disorders" or "High functioning autism" or "Asperger" or "ASD" or “Autistic”) AND ("Immersive virtual reality" or "Head Mounted display" or " IVR"or "virtual reality")

The filtering of the documents was carried out with the use of the PRISMA methodology (Page & Moher, 2017). As articulated in Fig. 1, the first phase is called identification, the indicated keywords were applied, and a total n = 1322 documents were identified. The second phase, known as "Screening," inclusion criteria IC1,2,3, and exclusion criteria EX1,2,3 were applied (details in Table 1). This part of the study was developed collaboratively by the different authors of the article. Therefore, each one of them shared the information of each article based on the data of the abstract and the deep reading of the complete article. The third phase of the study, named Eligibility, considers the thematic criteria for the study of the documents. For this reason, inclusion criteria 4 and 5 were applied, in addition to exclusion criteria 4 and 5. The last phase called "included" specifies the final sample size n = 38 documents representing 2.87% of the initial sample of documents. As a summary, Fig. 1 shows the document filtering flow in the various phases of the PRISMA methodology.

Fig. 1
figure 1

Flowchart according to the PRISMA statement

Full size image

To complete the fourth phase of the systematic review, the following points were extracted from the studies and added to an excel spreadsheet: Title of the article, type of virtual reality used and the device, objective of the research, type of instrument, type of activity, results. The spreadsheet was completed by two independent researchers of the team. At the end of the process, they checked the differences obtained and were able to eliminate bias. This also helped to increase internal validity.

In the fifth phase of the systematic review, the items of the CONSORT statement (Moher et al., 2001) were followed to assess the quality of the review documents. Although some of the variables were not targeted by the study, we considered some quality criteria such as "Interventions for each group with sufficient detail to allow replication including how and when they were actually administered" (Moher et al., 2001, p. 1193). Another item considered was "A table showing the baseline demographic and clinical characteristics for each group" (Moher et al., 2001, p, 1192). Therefore, all the items of the CONSORT statement (Moher et al., 2001) were analyzed and the documents were given quality scores. In the sixth phase of the systematic review, the data analysis was implemented as indicated in Sect. 2.3. By way of a summary, Table 2 synthesizes the 38 documents that make up the sample. Table 3 shows the documents that the previous research has in common with the one presented by the authors.

Table 2 Description of the documents according to the variables
Full size table
Table 3 Articles in common between the different reviews
Full size table

2.4 Data analysis

To study of the documents obtained, two types of analysis were applied. On the one hand, a qualitative analysis was developed to qualitatively describe the phenomenon of interest based on the selected content and context variables. This is shown in Table 2. The process followed for the qualitative analysis was based on Cavanagh (1997), Elo and Kyngäs (2008), McCain (1988) and Polit and Beck (2004). First, the preparation phase was developed, in which the units of analysis or study variables were selected. Thus, it was decided what was to be analyzed and in what detail. This process was based on a theoretical review of the subject under study. Second, a categorization matrix was prepared, and the data were coded according to the established units of analysis. Third, the data obtained were reviewed and coded to establish correspondence with the units of analysis or variables contemplated. On the other hand, a quantitative analysis has been carried out to numerically study the chosen phenomenon by means of the frequency of occurrence of each of the variables previously studied in the qualitative analysis.

3 Results

3.1 The type of virtual reality

Our findings show that 52.63% (n = 20) of the included studies used desktop virtual reality. Research by Lorenzo et al. (2016) was noteworthy because it compares the use of VR and Immersive Virtual Reality (IVR) to determine how autistic students learn emotions. Immersive virtual reality was implemented in 42.10% (n = 16) of the studies, with the HMD device being the most widely used for IVR work (n = 10). Within this group we highlight the work of Schmidt et al. (2021a, b), the reason being that they employed spherical video in HMD devices. This allowed the research team and autistic groups to interact with the content and context of the environment via their head movements. Similarly, research by Lorenzo et al. (2013) pioneered the use of Cave Automatic Virtual Environment (CAVE) for VR interaction. There is also a study where both technologies are combined and another study in which the type of virtual reality technology used was not specified (5% of studies, n = 2).

3.2 The objective of the research.

The analysis of the objectives set shows great variability. In 18.42% (n = 7) of the publications, the objective was to develop/teach on social skills. Research by Ke et al. (2020) can be highlighted as one of those aimed at supporting social skills. This publication is characterized by working with everyday scenes such as the neighborhood, school, amusement park or fairytale scenes. In addition, the authors organize the tasks into four types: virtual teaching, social role-playing, environment design and social games. Ip et al. (2018) suggests the creation of a virtual reality programme to improve social and emotional skills. With this objective in mind, six learning scenarios are designed and developed, one of which focuses on the control of emotions and relaxation strategies, four of which work on social situations. The last of the scenarios consolidates generalization. In addition to this, the identification, interpretation, and emotional response was another of the objectives in 13.15% (n = 5) of the investigations. Within this topic, we should point out the work of Herrero and Lorenzo (2020). These authors analyze whether IVR produces improvements in learning and recognition of emotions. Studies aimed at the design, implementation, and evaluation of a VR system to help autistic students represented 10.52% (n = 4) of the studies. Tu et al. (2021) developed an online platform using desktop virtual reality to teach children with Asperger's adaptability and flexibility in addition to the integration of social and emotional development. After finalizing the design, they implemented a pilot study. In 7.89% (n = 3) of the included studies, symbolic play was proposed as one of the objectives. More specifically, Ke and Moon’s (2018) mixed-methods and multi-case research examine the association between game tasks, environmental features, and game-based social interaction using VR. To test whether VR is a facilitator of symbolic play. Finally, there are other objectives that have appeared in this study, such as communication skills (5.2%, n = 2), attention (5.2%, n = 2), social interaction (5.2%, n = 2) and motor skills. Jyoti and Lahiri (2020) work on attention by designing a desktop VR-based platform with a hierarchical queuing protocol (using eyes, head turns, finger pointing, etc.). This platform adapts to individualized performance and autonomously increases the level of cues on demand. Similarly, one of the research that aims to work on communication and social skills is Rutten et al. (2003). These authors use desktop virtual reality environments to foster social communication and interaction. More specifically, it focuses on various situations that can be encountered in a cafeteria.

3.3 The instruments used in the research/support program

About the evaluation instruments, 42.10% (n = 16) of the studies used questionnaires. In the case of Jyoti and Lahiri (2020) a series of standardized questionnaires are used such as the social responsive scale (SRS) and the Social Communication Questionnaire (SCQ). Next, the combination of video recording and questionnaires are found in 15.78% (n = 6) of the studies. In research by Herrero and Lorenzo (2020), the sessions are recorded for subsequent analysis, while before and after the intervention, the study used a questionnaire they developed to determine the improvements of autistic students after working with VR. In relation to this, automatic evaluation systems appear in 13.15% (n = 5) of the studies. For example, Bozgeyikli et al. (2017), utilized an algorithm to automatically evaluate student performance across tasks based on a set of parameters. These included the completion time, the number of prompts and the number of incorrect actions. In doing so, a more personalized response to the students was achieved. Another element of evaluation is qualitative analysis of videos. 7.89% (n = 3) of the publications included this form of evaluation. Of note is the research of Ke and Moon (2018) who developed a behavioral analysis of children during interaction games taking verbal initiation, nonverbal initiation, and interpersonal negotiation to resolve a conflict as parameters. This highlights that there are also other types of combinations of instruments such as interview and questionnaire or video recording, interview, and questionnaire. In both cases they appear in 2.1% (n = 1) of the studies. Within this extensive group of papers is the research of McGowan et al. (2021), where video recording is used during the sessions. An interview with parents is also conducted to find out what changes their children have experienced. Throughout the intervention it is necessary to fill in the CRASS (communicative responses/acts score sheet) questionnaire.

3.4 According to the type of activity

In the studies analyzed, the results were as follows. In 5.26% of the studies (n = 2) the desktop VR user identified a series of images associated with an emotion. However, they did not have to apply it to a social situation. In this case, the work of Tu et al. (2021) stands out, which uses desktop virtual reality to design an online platform for working on social skills in autistic students. In 13.15% of the publications (n = 5), identification and recognition activities are proposed, as in the previous case, but associated with social interaction. In this way, the child can learn what would be the consequence of expressing that emotion in a given situation. The work of Lorenzo et al. (2016) stands out, where, by means of a script, the child can learn how to act when expressing a certain emotion, both for him or herself and for the interlocutor. Within this group of work, it is worth highlighting that more than 70% of the publications have used the IVR in its HMD or CAVE modality. The use of desktop virtual reality was a minority.

There are other types of activities with non-social content that have also been developed. For example, presenting the child with a series of bubbles to be popped. Another case is the possibility of guiding a ball along a path. It is also possible to use a driving simulator (Bian et al., 2019), answer questions about a country (Bossavit & Parsons, 2018) or put a series of geometric figures in a certain place (Lu et al., 2018). Thus, there are n = 5 studies that have focused their work on non-social tasks. Only one of the studies has used IVR, the rest have worked with desktop virtual reality. As for the rest of the activities that have been studied, it has been possible to observe that there is a great diversity of context. In this sense, there are a total of 7 studies where activities related to the school context are implemented. For example, in the work of Halabi et al. (2017) using IVR, the user starts by navigating through the school until the arrival in the classroom. This classroom consists of two students and a teacher. The teacher explains the instructions to introduce himself by addressing all participants including the avatar of the autistic child. In this way we try to work on presentation skills. In this line is also the work of Ke et al. (2020) who among the various scenarios have a school café where the user has to sit and negotiate with other peers. Similarly, it is also included within the classroom where he carries out a series of mathematics activities to learn flexibility. All these activities are carried out with desktop virtual reality. Within this group, the activities developed by Ip et al. (2018) also stand out. In the first of the user activities, the child practises the routines for going to school. The scenario starts in the bedroom when the alarm clock has been set. The child is encouraged to go to the alarm clock and stop it by ringing. A checklist will appear to indicate that the task has been completed and to direct the child to the next step, which is to use the toilet. In each of the situations the therapist may introduce an alternative situation. The second scenario designed by Ip et al. (2018) consists of a series of routines that children can practice and experience during school days in the setting such as greeting the teacher, handing in homework, following the teacher's instructions, and joining in learning activities, etc. These activities were developed with the IVR. It is also the case that there were n = 4 studios that have designed their activities for school-related settings but do not take place within the school, such as the cafeteria and getting on transport. The activities carried out in the cafeteria take place with desktop virtual reality, with the work of Ke and Im (2013) standing out, while the transport activities, such as those of Schmidt et al. (2021a, b), are carried out with immersive virtual reality. In both cases the user has to choose a place to sit. Activities have also been developed in non-school settings such as money management, cleaning the house, a restaurant, or a shop. However, in all of them activities were designed in which the user was supported by a therapist through virtual environments. These were all interactive and realistic. To summarize, it can be concluded that the activities have been planned in the form of games while mostly using desktop virtual reality. This is in addition to all the activities needing the user to provide some feedback to make the scenario resemble a real social interaction using immersive virtual reality. Interestingly, the more complex tasks used IVR, while the more basic activities tended to use desktop VR.

3.5 According to the findings

The results obtained in the different studies have been distributed as follows. The results of 18.42% (n = 7) of the studies show an increase in the identification, understanding and interpretation of emotions due to VR. As an example, the research results of Ip et al. (2018) may be analyzed. These authors observed a substantial improvement in emotion expression, emotion regulation and social interaction. Although most of the training took place in virtual environments, generalization of skills was achieved in real environments. Another of the results to highlight is that in 15.78% (n = 6) of the studies it is observed that VR is very well accepted by autistic students. In this sense, the system designed by Jyoti and Lahiri (2020) was accepted in a good way by these students. Similarly, it was a system enabled to estimate the level of group attention skills of autistic child in terms of task performance. In addition, it was able to identify this level individually and adapt to the needs of the students. Following this line, improvements in social skills and attention (10.52%, n = 4) have been experienced due to VR. In this way, Tu et al. (2021) suggest that the designed system has led to an improvement in social skills (affect recognition skills, analytic reasoning skills and social attribution skills).

Regarding attention, Banire et al. (2021) report in their results that there has been an improvement in the time that autistic child are engaged in a social interaction developed with virtual reality. In this sense, 7.8% (n = 3) of the publications show a development of symbolic play. In one example, Halabi et al. (2017) provide a suggestion that VR has led to an improvement in the symbolic play of autistic students after developing the activities in interactive scenarios with VR. Also, some aspects such as communicative behavior (5.2%, n = 2), improved eye contact (2.6%, n = 1) are considered important. In the case of communicative behavior, McGowan et al. (2021) suggest from their results that the tool they used assisted with the development of communicative behaviors such as the initiation of social greetings or interest in starting a conversation. For eye contact, Herrero et al. (2020) state in their research that eye contact had increased in students after working with virtual reality; although it’s not clear the benefits this finding provides.

4 Discussions

In this section we will discuss the results of our study focused around the five guiding research questions. The first research question referred to the type of virtual reality. Desktop virtual reality was employed in 52.63% (n = 20) of the studies. These results are slightly less than those of Mosher et al. (2021) who located 72.2% of studies using desktop virtual reality. In addition, Mak et al. (2020) located 75% of studies making use of desktop VR. However, in our study the number of studies employing immersive virtual reality are equal to 42% (n = 16) whereas previous reviews do not find anything more than 20%. The limited application of immersive virtual reality in previous studies is in line with research by Newbutt et al. (2016), who argue that research around the impact and evidence of CAVE or HMD devices on autistic users is scant. It is therefore not entirely clear how widely accepted this tool is for these individuals. Furthermore, Newbutt et al. (2016) claim that despite the accessibility and low cost of these relatively new tools (i.e., HMDs), there is insufficient evidence, or knowledge, on negative effects. Therefore, the case for using desktop virtual reality is higher, as the equipment used is common (i.e., computer monitor connected to a personal computer) (Bellani et al., 2011). As a result of these simple interfaces, desktop VR is accessible for implementation in educational environments.

The second research question focused on the most studied objectives (in terms of targeted outcomes). These were social skills and emotions (31.42%, n = 12). These results are slightly less that found in the study by Mesa-Gresa et al. (2018) who located 65% of studies focused on social skills and emotions. In addition, our findings are similar to those of Thai and Nathan-Roberts (2018) who located 42% of publications focused on emotions. The third most common objective, in 10.52% (n = 4) of the studies is that of design and implementation of VR systems. No other previous research identified this as a key objective of research in this field. The importance of designing immersive systems is due to the fact that there are still no conclusive studies that allow the generalization of activities learned in virtual environments into real environments (Glaser & Schmidt, 2021). For Dalgarno and Lee (2010) it is assumed that the greater veracity of virtual environments the greater the behaviors will be produced in real environments. However, Stokes and Osnes (2016) assert that VR system designs must consider the interaction of tasks, environments, participants, and technology; as well as the way in which performance supports are introduced. This will influence the generalization of learning in different contexts. Another of the most important objectives is that of symbolic play with 7.82% of the studies focusing on this. This is in line with the results of Dechsling et al. (2021) where 2% of the research has targeted this area. According to Wetherby et al. (2004) and Thiemann-Bourque et al. (2012) one reason that some work has focused on symbolic play is because this is considered an early indicator for the diagnosis of autism and its evaluation (APA, 2013).

In relation to the use of instruments for the evaluation of outcomes for autistic students, 42.10% (n = 16) have used a questionnaire. According to Chakraborty et al. (2021), the vast majority of questionnaires are used for diagnosis and not for the evaluation of improvements. The combination of questionnaires with video recording was present in 15.78% (n = 6) of the studies. The inclusion of video as a measurement is a consequence, according to Grossi et al. (2021), of the fact that this instrument can help to describe the intricate pattern of restricted and repetitive behaviors of people with autism. In this way, a better understanding of their behavior is achieved (Melo et al., 2020). This is not possible with the questionnaire, which is restricted mostly to diagnosis. One more of the evaluation systems is automatic systems, which appear in 13.15% (n = 5) of the investigations. There are two reasons for the use of automatic systems. Firstly, Alnajjar et al. (2021) claims that automatic systems collect relevant information on various parameters of social interaction. This will be used for the creation of retrievable databases of user evaluations and history. The second reason according to Rudovic et al. (2018) is that these instruments, unlike traditional ones, enable better monitoring of a larger number of learners with different needs and in varying situations.

The following research question is focused on the characteristics of the activity. Firstly, activities that have been developed to work on emotions, have mostly been carried out with HMD devices and in social situations where the identified emotion can be applied. According to Garon et al. (2018), the reason for designing these activities in this way is due to the fact that children with autism find it easier to interpret synthetic and computer-generated devices than natural stimuli. Similarly, Parsons (2016) justifies those interactions in virtual environments that are more naturalistic, more direct and allow for more interaction will facilitate the generalization of learning. In the same way, Smidth et al. (2021) add another reason justifying that the learning of autistic students will be favored when the user has interaction devices that are as realistic as possible. This contrasts with desktop VR where the user interacts with a joystick or keyboard. Following the analysis, it has been observed that in 5 of the publications, activities have been proposed in which knowledge of VR has been promoted rather than the improvement of the skills of the child with autism. According to Parsons and Cobb (2011), the reason that could justify the work of non-social activities could be that there is still no conclusive research on the characteristics that VR technology should have to support the learning of autistic students. In this way, a better understanding of this technology is achieved to subsequently implement it in social situations. It is important to note that the context in which most of the work has been carried out is the school, with 7 studies. It is important to note that the context in which most of the work has been carried out is the school, with 7 studies. According to Nikula et al. (2021), this may be a consequence of the fact that the school plays a fundamental role for students with special educational needs according to the inclusive policies that are being implemented in many countries. It is therefore necessary to ensure that barriers to access, participation and learning that limit their inclusion in the school along with the rest of their peers are removed. In this way, virtual reality will be the element that will provide the educational response for these students. Finally, according to Dalgarno and Lee (2010), it can be said that the use of IVR in activities that work on complex social situations is due to the fact that VR systems that allow for greater involvement and interaction of the students will be those that facilitate better learning. To this end, these authors suggest that VR systems should provide several features such as immersion, fidelity, and presence. For these authors immersion is based on the technical capabilities of the VR technology to produce sensory stimuli, while presence is context dependent and is based on the individual's subjective psychological response to VR. Whereas fidelity is that which allows the user to feel included in the environment in such a way that he/she considers every action to be as if he/she is performing it for real.

The last of the research questions focuses on what the most frequent results of the studies are analyzed. First, it was found that in 18.42% of the publications (n = 7), there was an improvement in emotion identification and recognition. In the same line, Mesa-Gresa et al. (2018) identified 22% of publications that confirmed improvements in the identification, recognition, and expression of emotions. These results may be a consequence of the fact that VR allows the generation of instructions to guide autistic students on how to respond to an identified emotion (Russo-Ponsaran et al., 2016). Moreover, in autistic learners the ability to recognize and identify emotions from non-human stimuli, such as cartoons, caricatures or schematic faces, or stimuli via VR, remains intact, unlike with the identification of human stimuli (Brosnan et al., 2015).

Second, ASD students' acceptance of VR as a learning tool was found in 15.78% (n = 6) of the studies. Bradley and Newbutt (2018) found that in 50% of the investigations there was no acceptance of VR, with side effects such as dizziness, anxiety, or cybersickness identified. This variability in the acceptance of VR environments may be due to several reasons. On the one hand, ASD is mostly accepted as a complex, pervasive, and heterogeneous condition with a wide range of etiologies, subtypes, participant age, and developmental trajectories (Glaser & Schmidt, 2021; Masi et al., 2017). On the other hand, according to Parsons (2016), the acceptance of VR is conditioned by context, goal, and task type. Moreover, the lack of standardized VR technologies (Parsons, 2016) forces to combine different types of technology, with the problems that this can entail (Skarbez et al. 2017).

Thirdly, it is worth noting that 10.52% of the papers (n = 4) confirmed the improvement of social skills due to VR. Similar are the results of Mesa-Gresa et al. (2018) and Dechsling et al. (2021), who identified 9.6% (n = 3) and 12% (n = 5) of the publications, respectively. In these improvements it is timely to take into consideration that autistic users are more likely to adopt the self-identities of their avatars, which facilitates the perception of simulated social contexts in VR (Wang et al., 2016). Furthermore, Wang et al. (2016) demonstrated that autistic students can be proactive to social situations presented to them through VR. In addition, the acquisition of social skills is faster in social situations that the child experiences on a day-to-day basis and that he/she will be able to work with VR to develop his/her learning (Frolli et al., 2022).

5 Conclusions

This review has highlighted that the application of virtual reality between 1996 and 2021 with and for autistic students has helped to enable and support individuals in the development of autonomy and inclusion in educational contexts. Based on the results, we highlight the following conclusions:

  • The most widely used type of virtual reality to date is desktop virtual reality.

  • In the research analyzed, the two most frequent research purposes have been: firstly, to develop and teach social skills and secondly, to work on the identification, interpretation, and emotional response of autistic students

  • The instrument most widely used in the interventions was a questionnaire, followed by a questionnaire-video combination.

  • The most applied activities are of a realistic nature and with elements of interaction that allow autistic users to be included in the environment.

  • The most frequent findings have been the improvement in the identification and interpretation of emotions and the students’ acceptance of VR as a support tool.

Our study has highlighted that the use of virtual reality for students with autism has been applied schools (in various situations and in non-social contexts. These are spaces where social situations do not arise due to the absence of another interlocutor. However, the there are some limitations that should be outlined. For example, there are publications that do not include the necessary information for a more complete analysis; they do not include a detailed description of the activity being developed. Also, publications were found that did not define the instrument to evaluate the student’s progress or learning. More specifically, some papers mention that they used an instrument developed by them do not indicate its main features. Therefore, to achieve greater transparency in the scientific process and to help the performance of systematic reviews, it would be advisable to agree on minimum indicators that should necessarily be described in all studies.

Regarding the practical implications that can be derived from the study, firstly, the knowledge of the objectives on the use of VR in autistic students provides the educational community information about the applications that VR can have. Teachers, based on this, can consider the use of VR to respond to the behavior of these students. It is recommended to adjust the objectives to the characteristics and needs of the students; placing autistic learners more centrally to the design and development of VR. Likewise, the results allow the educational community to know the most relevant advances that VR has brought about. It is recommended that future research considers work in the areas in which VR has been found to be most effective; ensuring an evidence-based approach. However, it is also advisable to progressively expand the use of VR to other areas, trying to implement methodological and technological improvements based on the limitations identified in the previous studies in the field. The characteristics of the activities applied with VR provide information to the educational community on how to develop the didactic strategies to be applied with autistic students. It is recommended to design activities that solve situations in contexts that and autistic child encounters in everyday life. Moreover, it is suggested that the activities be manipulative in nature and include a virtual instructor who can help the child when he/she does not know how to act in the environment.

Considering the type of virtual reality, our analysis has enabled us to identify what type of hardware has been used the most in studies with autistic students to generate VR. Furthermore, information has been provided on the nature of the virtual reality used. The use of immersive virtual reality and HMD devices is recommended, due to the ease of interaction they offer to autistic students and the realism of the environments that can be created. Likewise, the analysis of the assessment instruments used provides the educational community the possibility of knowing which instruments are suitable for assessing variables in a VR intervention. In this sense, it is recommended to use instruments already validated and successfully used in previous research. However, it is suggested to add an automatic data collection system, which can counteract the possible bias generated by the researchers. For example, a camera system to be able to determine the position and orientation to which the child is looking within the VR environment.

In sum, this study aimed to contribute to educational communities and to better understand how to organize educational responses to students with autism via VR. Thus, if VR is intended to be used for learning emotions, it is recommended to use an avatar to ask the child what his or her mood is. The child could respond with a series of YES or NO cubes if he/she knows the expression that the avatar is expressing. The cubes are used so that this activity can be navigated by autistic students with different levels and input of communication. In addition, VR can be used for learning classroom rules. In this case, a child would have a pictogram agenda to know how to behave in a classroom. Then, the teacher's virtual avatar will tell him/her to arrange a series of pictograms on a sheet classifying the rule as what can and cannot be done in the classroom. Each of the pictograms will have a color associated with it so that the child can relate it to whether the rule is done.