Virtual Reality Interventions for Enhancing Executive Functions in Children and Adolescents with Autism Spectrum Disorder

Abstract

This study investigates the impact of a Virtual Reality (VR)-based intervention on the enhancement of executive functions—cognitive flexibility, inhibitory control, and working memory—in children diagnosed with Autism Spectrum Disorder (ASD). Employing a single-case experimental design with repeated measures, the research was conducted with two male participants, aged 9 and 10, both formally diagnosed with ASD. The intervention was structured into four phases: Baseline (no training), Intervention (targeted VR training), Generalization (skill transfer testing), and Follow-up (maintenance assessment). Each participant engaged in a total of 18 tasks (six per executive function), delivered through immersive VR environments featuring gamified elements, adaptive feedback, and increasing difficulty. Each task consisted of up to 15 sub-items, scored as correct or incorrect. Results indicate consistent improvements across executive function domains during the intervention phase, with partial maintenance at follow-up and evidence of task generalization. Given the single-case framework and limited sample size, findings should be interpreted as exploratory and hypothesis-generating rather than population-generalizable. The study provides proof-of-concept evidence supporting the feasibility and potential of immersive VR-based executive function training for ASD populations, warranting further validation through larger-scale controlled trials.

1. Introduction

Executive Functions (EF) constitute a set of higher-order cognitive processes, such as working memory, cognitive flexibility, and inhibitory control, which are critical for behavior regulation, learning, and individuals’ adaptation to the demands of everyday life [1]. These skills enable planning, impulse control, problem solving, and the management of complex cognitive and social situations.

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by difficulties in social communication and interaction, as well as restricted and repetitive patterns of behavior. A substantial body of research indicates that children and adolescents with ASD frequently exhibit significant impairments in executive functions [2]. These deficits do not merely represent secondary features of the disorder, but directly affect academic performance, self-regulation, skill generalization, and functional independence. For example, reduced cognitive flexibility is associated with difficulties adapting to change, while impairments in inhibitory control and working memory hinder rule-following, instruction compliance, and problem-solving.

Strengthening executive functions in individuals with ASD is therefore a central goal of educational and therapeutic interventions [3]. Although various approaches have been implemented, such as cognitive training, behavioral interventions, and structured learning programs, their effectiveness is often limited by low levels of engagement, reduced motivation and difficulties in transferring acquired skills. Particularly for individuals with ASD, VR enables repetition, gradual increases in task difficulty, and the reduction of external distractions—features that are critical for executive function training [4] the simulation of real-life situations in a safe and controlled manner, supporting experiential learning without real-world risks [5]. Based on these considerations, the present study focuses on investigating the effectiveness of a targeted Virtual Reality intervention for enhancing three core executive functions—cognitive flexibility, inhibitory control, and working memory—in children with ASD. In contrast to purely review-based studies, this work is grounded in primary empirical data derived from the experimental implementation of VR intervention.

Specifically, a single-case experimental design with repeated measures was employed to examine the impact of the VR intervention on participants’ performance before, during, and after the intervention, as well as their ability to generalize and maintain the acquired skills [6]. It was proved that immersion, interactivity, and game-based elements inherent in VR can lead to measurable improvements in executive functions and support the transfer of skills to new conditions.

The aim of the present paper is to present and analyze the results of this specific VR intervention, evaluate its effectiveness across individual executive functions, and discuss the findings in relation to their application in educational and therapeutic settings. The results of the study seek to contribute to a deeper understanding of the role of Virtual Reality as a cognitive enhancement tool in ASD, as well as to lay the groundwork for future research and optimization of VR-based interventions [7].

2. Literature Review

2.1. Executive Function Challenges in ASD and Importance of Intervention

Executive Functions refer to the brain’s self-regulatory processes that enable planning, flexible thinking, impulse control, and goal-directed behavior. In typical development, EF skills improve through childhood and adolescence, supporting academic achievement and adaptive life skills [8]. However, in Autism Spectrum Disorder, research has documented significant EF impairments, even in high-functioning individuals [8]. Common EF difficulties in ASD include poor cognitive flexibility (e.g., difficulty switching tasks or coping with change), inhibitory control deficits (impulsive responding or inability to suppress inappropriate behaviors), and working memory limitations (trouble holding information in mind and manipulating it, which affects problem-solving and language processing). These EF challenges can manifest in classroom settings as an inability to follow multi-step instructions, in daily life as trouble with organizing activities or inhibiting repetitive behaviors, and in social situations as difficulty taking turns or adapting to others’ behavior. Notably, EF skills are thought to underlie many adaptive behaviors; thus, improving EF in children with ASD could potentially enhance their independence and ability to learn new skills [9]. Traditional interventions for EF in ASD have included strategies like visual schedules and prompts (to compensate for planning deficits), behavioral inhibition training (e.g., “stop and think” techniques), and cognitive exercises or computer games aimed at memory and attention. While some factors have been observed, the engagement factor and ability to generalize these skills remain concerns. This is where interactive and immersive technologies have introduced new possibilities.

2.2. Virtual Reality as an Intervention Tool

Virtual Reality offers a novel medium for delivering interventions that are both immersive and controlled. VR systems typically consist of a head-mounted display or large screen that presents a 3D simulated environment, often combined with sensors or controllers that allow the user to navigate and interact within that environment. The high level of sensory immersion can create a feeling of presence in the virtual world, capturing the user’s attention more effectively than traditional two-dimensional games or exercises. For children with ASD—who often prefer visual/spatial learning and are drawn to technology—VR’s immersive qualities can make therapeutic exercises more engaging and motivating [10]. Crucially, VR allows repetition and consistency in a way that is hard to achieve in the real world: scenarios can be repeated as many times as needed, difficulty can be systematically varied, and distractions can be introduced or removed to suit the learner’s level. This aligns well with the needs of ASD interventions, which often require structured practice and gradual generalization.

Researchers have utilized VR in various ways to support individuals with ASD. A major category of applications is simulated real-life training, where VR is used to teach practical skills or safe behaviors. A classic example is the street crossing simulator developed by Josman et al., which placed children with ASD in a virtual city street and taught them how to safely cross the road [11]. This study found that children who practiced in the VR crossing scenario improved their street-crossing skills and were better able to demonstrate those skills outside of VR compared to untrained children [11]. Another example is virtual environments for grocery shopping or classroom routines, allowing students with ASD to rehearse those scenarios (like navigating aisles, waiting in a lunch line, etc.) with less anxiety [12]. What these applications share is the provision of a safe environment: the child can make mistakes in VR (e.g., step off a virtual curb at the wrong time) without real harm and can immediately retry the task. Continuous, real-time feedback is usually built into these systems—for instance, the program might alert the child if they attempt to cross when traffic is approaching or might praise them for successfully completing the task. The inclusion of immediate feedback and the ability to redo tasks aligns strongly with evidence-based teaching principles and reinforces learning through practice [12].

Another broad area is using VR for social and communication skills training. Children with ASD often find live social interactions unpredictable and stressful. VR social skills programs create simulations of social scenarios (such as a conversation with a peer or interacting with a teacher) in a controlled manner. Kandalaft et al. introduced one of the early VR social training programs for young adults with autism, using an office environment simulation to practice job interview skills [13]. Participants could role-play an interview with a virtual human, receiving coaching and feedback as needed. More recent examples include “virtual playground” environments where children can practice social play and collaboration, or VR applications for recognizing facial expressions and body language. Yuan and Ip used a virtual platform to train children on understanding non-verbal cues by interacting with animated characters [14]. The outcomes from such studies have been encouraging—for instance, Didehbani et al. reported improved emotion recognition and social responsiveness in children with high-functioning ASD after they underwent a VR social cognition training program [15]. Similarly, Chen et al. found that VR-based social skills training led to better eye contact and increased initiation of conversation in a sample of school-aged children with ASD. These improvements highlight VR’s potential not only for cognitive practice but also for interactive social learning, a domain traditionally challenging to address in ASD [16].

Importantly, many VR interventions targeting social or practical skills inherently exercise executive functions as well. For example, a social scenario in VR might require a child to flexibly shift between talking about different topics (cognitive flexibility) or to remember a rule about taking turns (working memory), while inhibiting the urge to monopolize the conversation (inhibitory control). Thus, even when EF is not the explicit target, VR-based social simulations can engage EF processes and potentially strengthen them. Conversely, VR cognitive training exercises often have social components (like narrative contexts or cooperative game modes) that confer social understanding benefits. This overlap is advantageous because it means an intervention can address multiple needs simultaneously. Still, some programs have been developed with a primary focus on executive functions; these are discussed next.

2.3. VR-Based Programs for Executive Function Training

A growing number of studies have explicitly examined VR interventions for improving executive functions in ASD. One approach in these programs is to adapt standard EF training tasks into an immersive VR format. For instance, tasks like the Wisconsin Card Sorting Test (which assesses cognitive flexibility) or n-back tasks (for working memory) can be made into VR games where the child interacts with objects or navigates spaces to complete the tasks. The rationale is that the immersive context may increase engagement and the sense of challenge, thereby potentially leading to greater training effects. In a meta-analysis by Kim et al. that aggregated results from multiple studies, the authors found that VR-based executive function training produced significant improvements in overall EF scores for children with ASD, with some studies reporting effects comparable to traditional therapist-led interventions. However, the meta-analysis also noted variability between studies and methodological limitations such as small sample sizes, emphasizing the need for more rigorous trials [17].

Concrete examples of VR EF interventions include the work of Zhang et al., who conducted a randomized controlled trial specifically to test a VR program aimed at improving EF in children with autism. In their study, the VR group participated in a series of game-like executive function tasks within a virtual environment, over multiple training sessions, while a control group received standard care or no intervention [18]. Results demonstrated that the VR-trained group showed greater gains in executive function assessments (such as improved problem-solving and cognitive flexibility) compared to the control group at post-test. This suggests that immersive VR can be an effective medium for EF training when properly designed. Another study by Ji et al. compared VR cognitive training with physical exercise as two different interventions to boost EF in children with ASD [19]. Interestingly, they found both interventions led to EF improvements (reflecting the known benefits of aerobic exercise on cognition), but the VR group’s performance on tasks of working memory, inhibitory control, and cognitive flexibility improved to a similar extent as the exercise group [19]. Moreover, the VR training had the advantage of being easily standardized and monitored—all participant actions and responses in the virtual tasks were recorded, allowing fine-grained analysis of their progress. This study also included a follow-up phase: after a three-week break with no interventions, children were tested again and it was noted that while some of the gains slightly diminished, a portion of the improvements persisted, indicating a lasting benefit from the VR intervention. Such findings underscore that VR interventions can produce meaningful cognitive changes, though like any training, ongoing practice may be needed to maintain peak benefits.

Several VR interventions for EF incorporate adaptive difficulty and personalized feedback, which are key for maintaining an optimal challenge level. For example, a VR game might start with simple tasks (e.g., remembering two items) and then progressively increase the difficulty (e.g., four items, then six) as the child demonstrates mastery—a principle known as graduated challenge. Some systems adjust in real time: if a user is struggling, the game might provide hints or lower the difficulty to prevent frustration, whereas if the user is succeeding easily, the game can introduce a harder level to keep them engaged. Adaptive feedback mechanisms ensure that the intervention remains neither too hard nor too easy, thereby maximizing the training effect. Many studies attribute at least part of their success to these adaptive features. For instance, Mouga et al. employed a virtual supermarket task combined with eye-tracking to train and assess EF; the task difficulty would escalate only once the child consistently performed well at the current level, which helped in pinpointing specific EF deficits while also training them (e.g., increasing memory load in a shopping list task to stretch working memory) [20]. The integration of eye-tracking also provided insight into the child’s attention strategies (an executive aspect) during the task.

Moreover, many EF-focused VR programs explicitly include game reward systems to leverage motivation. This might involve awarding points, unlocking new levels or virtual rewards (like badges or avatar accessories), when the child demonstrates improvement or uses a specific executive skill appropriately [20]. Such gamification elements have been shown to sustain participant interest and encourage repeated practice—critical factors for neurocognitive training. In a review of design considerations for VR in ASD, Bozgeyikli et al. emphasized that playful features and rewards can turn training exercises into a fun experience rather than a tedious drill [21]. Indeed, participants in several studies reported enjoying the VR games so much that they were more willing to practice longer than they would in a non-VR setting, a qualitative outcome that bodes well for skill acquisition.

2.4. Integrating ABA and Behavioral Strategies in VR

The fusion of VR with Applied Behavior Analysis (ABA) principles is a notable theme across literature. ABA’s emphasis on reinforcement of desired behaviors, repetition, and data-driven adjustments can complement VR technology’s strengths. Some researchers have effectively translated ABA techniques into VR by structuring virtual tasks as a series of trials and providing systematic reinforcement [22]. For instance, an ABA-based VR intervention might present a child with a virtual scenario requiring a target behavior (say, responding to a social question or performing a step in a task). If the child responds correctly, the system immediately delivers a reinforcer—which in VR could be as simple as a cheerful animation or as elaborate as unlocking a mini-game or receiving a token that the child collects [23,24]. If the child responds incorrectly or not at all, the system might employ an error-correction strategy (like a prompt or hint) and later give another opportunity, much like a therapist would in a discrete trial training (DTT) session. Epifânio and Da Silva have discussed how a Serious Game Design Document (SGDD) model can explicitly incorporate ABA elements, ensuring that game-based interventions maintain the fidelity of behavioral intervention techniques [25]. In practice, VR programs that were developed with ABA consultants on board often include features like customizable reinforcement schedules (so that after a certain number of successful behaviors the child gets a bigger reward), data logging of each trial’s outcome, and the ability for a practitioner to review performance data and adjust the program for next sessions [26].

The benefits of combining VR with ABA are two-fold. First, ABA provides a proven framework for teaching new skills and reducing problematic ones; by embedding this framework in VR, one can create an intervention that is both fun and rigorously structured. Second, VR can enhance ABA by providing consistency and by collecting precise data. The VR system can ensure every learning trial is presented the same way, with the only variance being the child’s response, which yields cleaner data on performance. It can also automatically track metrics such as reaction times, error rates, and hints used, which are valuable for assessing progress in executive function tasks [27]. Some preliminary results suggest that VR interventions with ABA-like reinforcement and repetition achieved higher improvements in target skills than those without such structure, and importantly, that children were able to generalize the skills learned in VR to real-world settings more effectively [28]. For example, a child who learned impulse control in a VR game (where they had to wait their turn or resist clicking on a non-target object) not only improved in the game but also showed better impulse control in a classroom setting afterward—likely because the VR training reinforced the behavior strongly and the experience was memorable [29].

It is worth noting that beyond ABA, other therapeutic approaches like Cognitive Behavioral Therapy (CBT) have also been augmented with VR in ASD populations. One case study (De Luca et al.) used a VR application called “BTS-Nirvana” as part of a cognitive–behavioral intervention for an adolescent with ASD and anxiety, finding reductions in anxiety symptoms and improved executive regulation when VR exposure therapy was alternated with traditional therapy sessions [30]. While this review’s focus is on executive functions, these examples illustrate the versatility of VR as a platform that can host various intervention paradigms.

2.5. Design Principles for Effective VR Interventions

From the collective findings of the literature, several design principles have emerged as critical for the success of VR interventions in ASD. Recent systematic reviews (e.g., Li et al. [2], Zhang et al. [18]) have pointed out common features of effective programs. Key principles include:

Clearly defined goals and skills for each VR activity, ensuring that the purpose (e.g., improving working memory or practicing social greeting) is explicit and can be measured.

Repetitive practice and mastery learning, allowing the child to repeat tasks or levels until the skill is acquired. VR enables this by keeping scenarios available on-demand, which is especially useful for EF training where frequent practice is needed.

Gradual increase in difficulty (graduated challenge), so that tasks become more challenging only as the child is ready. This might mean increasing the amount of information to remember, the level of distraction to ignore, or the complexity of a problem to solve—aligning with the child’s improving capability. The child progresses to a next level or new scenario only when mastery of the prior level is achieved, preventing frustration and building confidence stepwise [31].

Personalization and flexibility, whereby the VR content or gameplay is tailored to the individual’s profile. For instance, the themes of the game can be adjusted to the child’s interests (to enhance motivation), or specific supports can be enabled/disabled (like visual prompts or simplified interfaces for children who need them). This customization is crucial in ASD, given the heterogeneity in abilities and sensitivities. Many VR platforms allow facilitators to choose settings appropriate for each child [32].

Additionally, attention to sensory design is crucial when creating VR environments for ASD users. Children on the autism spectrum may be prone to sensory overload or may focus on irrelevant details if the environment is too cluttered. Best practices call for maintaining visual simplicity and clarity in virtual scenes [33]. That means avoiding excessive background movement or noise, using a clean layout with only task-relevant stimuli, and preferring softer colors and contrasts over very bright, flashing elements (which could be distracting or aversive) [34,35,36]. Some programs even include “low sensory stimulation” modes that can be toggled on for users who are easily overwhelmed. Another design consideration is the type of avatars or characters used: interestingly, studies have found that using non-human or fantastical avatars (such as friendly robots, animals, or fantasy creatures) can sometimes keep children more engaged than realistic human avatars [37]. One reason is that human-like avatars might inadvertently trigger social anxiety or uncanny valley effects, whereas a cartoonish robot is perceived as fun and non-judgmental. For example, a shy child with ASD might feel more comfortable practicing conversation with a cute alien avatar than with a realistic child avatar. This strategy helps maintain high engagement without the pressure of real human interaction [38].

Moreover, effective VR design often grants the user a degree of control over the experience. Allowing children to make choices—such as selecting which activity to do next, controlling the pace of the scenario, or deciding how to customize their avatar—can increase their sense of agency. This is important because many individuals with ASD experience anxiety when they feel a loss of control; giving them some autonomy in the virtual environment can reduce stress and foster a positive learning state. For instance, a VR training program might offer a menu of mini-games targeting different skills, and the child can pick the order in which to play them or take breaks by exploring a calm virtual space in between challenging tasks. Such features align with person-centered intervention approaches and make the experience more user-friendly for neurodiverse learners [18,39].

Finally, continuous monitoring and data collection are embedded in most VR interventions, which is a design strength. Every action the child takes in VR can be logged: their response times, errors, strategies (if telemetry is sophisticated, e.g., tracking eye gaze or movement patterns), and progress over time. This data not only allow researchers to evaluate outcomes rigorously but can also inform on-the-fly adjustments. For example, if a child is consistently timing out on a particular task stage, the system might flag this for the therapist, indicating that a certain executive sub-skill (like response inhibition under time pressure) may need additional support or a different approach. In some advanced setups, neurophysiological data (like electrocardiogram, pulse plethesymograph, skin temperature, and galvanic skin response) have even been incorporated to gauge the user’s stress or engagement level, although such uses are still experimental [40]. Overall, the consensus design approach is to capitalize on VR’s adaptability and feedback capabilities to create an intervention that is tailored, engaging, and efficacious for each user.

3. Materials and Methods

MetaQuest 2, Meta Platforms, Inc., Menlo Park, CA, USA was selected as the VR device for the VR intervention since it combines a high degree of immersion with portability and ease of use. It is a standalone VR headset that does not require a connection to a computer during use, making it ideal for use in schools or treatment centers (easy to transport and install).

The VR device also features motion controllers that are tracked by cameras in it (inside-out tracking), allowing the user’s hands to be detected in space. This way the child can interact with virtual objects (e.g., “grab” an object, point, press buttons) with physical movements, which increases the sense of embodiment. This interaction is crucial for our cognitive goals as it motivates the children to participate actively, giving them the leading role in performing tasks, rather than being passive recipients. Finally, Quest 2 supports the development of custom applications (via Unity 3D or Unreal Engine), offering great flexibility to the intervention designer to create exactly the scenarios required. The computer used was an Micro-Star International Co., Ltd. (MSI), New Taipei, Taiwan, Cyborg 15 A13U gaming laptop, Intel Core, with an NVIDIA GeForce RTX 4060 graphics card. All the equipment used is shown in Figure 1. The VR software (including all the appropriate tasks) was developed in Unity, version 2022.3.12f1, using C#.

Table 1. VR software details.

Feature	Description
Game Engine	Unity (XR Interaction Toolkit)
SDK	Meta XR SDK
Programming Language	C#
3D Assets and World Design	Unity Asset Store
Locomotion	No scene or camera movement; the user remains seated. Objects are displayed in front of the user (object-centric VR).
Interaction	Controller-based interaction; object-centric actions (inspect, select, grab, place) with multimodal feedback (visual and audio).
Performance Optimization (PO)	Not applied, due to the use of simple graphics.
VR Acceptance Testing	Approved by experts to ensure that the VR intervention is clinically safe and functionally acceptable.

offers the central characteristics of our VR software.

As highlighted in Table 1, the virtual reality application is a lightweight, controlled, and clinically oriented VR software designed for early intervention in children with autism, which is deployed on the Meta Quest 2 platform. The application is developed using the Unity game engine in C#, with the Extended Reality (XR) Interaction Toolkit and Meta XR SDK, while its 3D environments and objects are sourced from the Unity Asset Store. The experience follows an object-centric VR design with no scene or camera locomotion; users remain seated and interact with virtual objects presented directly in front of them, minimizing sensory overload and reducing the risk of cybersickness. Interaction is controller-based and supports simple, intuitive actions such as inspecting, selecting, grabbing, and placing objects, enhanced through multimodal visual and auditory feedback. Due to the use of simple graphics, no explicit performance optimization techniques were required. Finally, the application has undergone VR acceptance testing and has been approved by domain experts, ensuring that it is both clinically safe and functionally appropriate for use in early autism intervention contexts.

3.1. Methodology and VR Intervention

3.1.1. Participants and Design

The study employed a single-case experimental design with repeated measures across multiple phases. Two children with Autism Spectrum Disorder (ASD), referred to here as Alexandros and Andreas, participated in the intervention. Both completed a series of 18 VR-based tasks (6 tasks targeting each executive function skill) over the course of the study, as presented in Appendix A. This single-subject design allowed each child to serve as their own control, with performance tracked over sequential phases to determine the intervention’s impact. No control group was used; instead, each participant’s baseline performance was compared to their performance during and after the intervention.

The VR intervention phase consisted of two sessions per week, with each session lasting 60 min. The total duration of the intervention phase was two consecutive weeks, resulting in four intervention sessions per participant. Each participant followed an individual single-case experimental timeline consisting of four structured phases (Baseline, Intervention, Generalization and Follow-up) as indicated analytically in

Table 2. Participant Timeline—Alexandros.

Phase	Week	Sessions	Description
Baseline	Week 1	2 sessions	Behavioral assessment without VR intervention
Intervention	Weeks 2–3	4 sessions (2/week × 60 min)	VR executive function training
Generalization	Week 4	1 session	Transfer tasks in modified VR environment
Follow-up	Week 5	1 session	Maintenance assessment

and

Table 3. Participant Timeline—Andreas.

Phase	Week	Sessions	Description
Baseline	Week 2	2 sessions	Behavioral assessment without VR intervention
Intervention	Weeks 3–4	4 sessions (2/week × 60 min)	VR executive function training
Generalization	Week 5	1 session	Transfer tasks in modified VR environment
Follow-up	Week 6	1 session	Maintenance assessment

. All sessions across all phases had a standardized duration of 60 min.

3.1.2. Experimental Phases

The intervention was structured into four distinct phases in sequence (see Figure 2), following a typical A–B single-case format with generalization and follow-up extensions:

Baseline: An initial phase with no specialized intervention. Each child’s performance on the target cognitive tasks was measured in its natural state to establish a baseline for comparison. This provided a reference point for each executive function before training.

Intervention: The training phase in which the VR-based program was introduced to improve the targeted skills. During this phase, the child engaged in the VR tasks designed to enhance cognitive flexibility, inhibitory control, and working memory. Performance data were collected under active intervention to capture immediate improvements relative to baseline.

Generalization: A post-intervention phase to assess skill transfer to new contexts. Here the tasks were deliberately varied (e.g., different stimuli or scenarios) to determine if improvements gained during intervention would carry over beyond the specific trained tasks. This phase evaluated each child’s ability to apply the acquired skills in environments or with stimuli not identical to those used in training.

Follow-up: A maintenance phase conducted after a delay, with no ongoing intervention. In follow-up sessions, each child was re-tested on tasks similar to those in training, but without any support or reinforcement, to examine whether the previously observed gains were retained over time. A sustained high performance in this phase would indicate that the improvements had been maintained (i.e., not fleeting or dependent on immediate intervention), whereas any decline would suggest skill regression once the intervention was removed.

3.1.3. Target Skills and VR Tasks

The intervention focused on three core executive functions, namely Cognitive Flexibility, Inhibitory Control and Working Memory (see Figure 3). Each of these skills were assessed through custom-designed VR tasks in a game-like format. The tasks were immersive and interactive, providing multisensory engagement through a head-mounted display and 3D virtual environments. Each executive function was exercised via distinct mini-games or challenges:

Cognitive Flexibility: Tasks required the children to switch between different rules or criteria and adapt to changing task demands (for example, sorting virtual objects by one rule, then suddenly having to switch to a new rule). Successful performance demanded the ability to shift attention and strategy when prompted by the game scenario (see Figure 4).

Inhibitory Control: Tasks involved situations where an automatic or prepotent response had to be suppressed. For instance, the children might see a frequent “go” signal in the VR game but must withhold responding when an infrequent “stop” cue appears. Such tasks exercised impulse control by rewarding delayed or rule-governed responses instead of immediate reactions (see Figure 5).

Working Memory: All tasks challenged the children to temporarily hold and manipulate information. Examples included remembering and recalling a sequence of virtual items or following multi-step instructions in the VR environment (see Figure 6). These activities require the children to maintain information (e.g., a sequence of lights or numbers) in mind while performing actions, reflecting real-time working memory use.

The VR tasks were developed with engaging, game-based elements and built-in feedback mechanisms to motivate the participants. The virtual scenarios were immersive and age-appropriate, featuring attractive graphics and narratives to captivate the children’s attention. Immediate performance feedback and positive reinforcement were provided within the VR environment—for example, the software would give auditory/visual rewards (cheers, points, badges) for correct responses and gentle corrective cues for incorrect responses. These reinforcement elements helped sustain motivation and clarify the link between the child’s behavior and outcomes, consistent with best practices in serious game design for learning. The tasks also progressively adjusted in difficulty (e.g., increasing working memory load or decreasing response time windows) as the child’s performance improved, ensuring an optimal challenge level throughout the intervention. Overall, the VR platform provided a realistic yet controlled space for the children to practice executive skills with continuous feedback and reward, a feature known to enhance engagement and skill acquisition in ASD interventions.

3.1.4. Trial Structure and Procedure

Each child completed a total of 18 discrete task trials during each phase of the study—specifically, 6 trials per executive function skill (cognitive flexibility, inhibitory control, working memory). In practical terms, for each of the three target variables there were six game-like tasks tailored to that skill, and each task could be administered in each phase. Within a given trial, the task consisted of multiple sub-components or levels (up to 15 per trial) that required a response. For example, a working memory trial might involve 15 recall attempts, or a cognitive flexibility trial might present 15 sorting decisions. Each sub-task was scored in binary fashion as either correct or incorrect (successful completion vs. failure). This yielded fine-grained performance data for every trial. A trial’s accuracy was quantified as the percentage of correct responses out of 15.

During Baseline, the children attempted the tasks without any instructional help or VR-based guidance, and their initial performances were recorded to capture natural ability levels. In the Intervention phase, the same tasks were performed but now within the full VR training context—the children received the VR-mediated prompts, practice, and feedback as designed, and their performance (with the support of the intervention) was logged for each trial. In the Generalization phase, the tasks were modified or presented with novel stimuli while still targeting the same underlying skill domains; this required the children to apply what they learned to new scenarios (for instance, a cognitive flexibility task might use different sorting criteria or stimuli than those used in training). Performance in this phase indicated whether improvements would carry over when conditions changed. Finally, in the Follow-up phase, conducted after a delay (e.g., several weeks’ post-training), the children were tested again on tasks analogous to those in training but without any assistive cues or reinforcement from the VR system. This follow-up served to examine skill maintenance over time. Across all phases, each child’s responses in every trial were recorded for subsequent analysis.

3.2. Data Collection

All performance data were captured and organized using a structured Excel spreadsheet for each participant. Each child had a dedicated worksheet labeled with their name (e.g., “Alexandros” and “Andreas”), containing all trial results across phases and variables. Within each sheet, data were grouped by cognitive variable: there were separate sections (with clear headers) for Cognitive Flexibility, Inhibitory Control, and Working Memory, each followed by the rows of data corresponding to that skill.

Each section’s data was arranged in a tabular format. Rows corresponded to individual trials, and columns corresponded to the sequence of item attempts (up to 15) within that trial. The first column of each data row indicated the trial identifier, composed of an index number and a phase code. Following the trial ID, the next 15 columns recorded the outcome of each attempt within that trial (columns were numbered 1 to 15, representing the maximum number of sub-items per task). In each cell of this grid, the child’s response was coded as “v” (a check mark indicating a correct response) or “x” (indicating an incorrect response). An empty cell (or a placeholder like “NA”) indicated that the trial did not include that many items (for instance, if a particular task only had 10 items, columns 11–15 would remain blank). This binary coding of success (“v”) vs. failure (“x”) for each item enabled precise calculation of performance percentages. The Excel organization thus provided a comprehensive record: each row captured a single trial’s outcomes (for a specific child, phase, and skill), and by scanning across the row one could see exactly which items were answered correctly or not. This detailed logging was crucial for later computing accuracy metrics and allowed examination of patterns (e.g., whether performance improved in later items of a task, or if certain task types had more errors). For the purposes of analysis, however, the primary focus was on aggregate performance per trial and per phase rather than item-by-item responses.

3.3. Data Analysis

After data collection, a custom Python-based algorithmic analysis pipeline was used to process the data and compute accuracy metrics for each participant and phase. The analysis was automated via a specially developed Python class called Analyzer, which was designed to read the Excel sheets and calculate performance statistics systematically (see Figure 7). The Analyzer class included methods to: (i) parse the Excel input into a workable data structure, (ii) compute the percentage of correct responses for each trial and phase and (iii) generate visualizations (e.g., bar charts) to summarize the results. Using this automated approach ensured that calculations were done consistently and reduced the potential for human error in data handling.

3.3.1. Data Parsing and Structure

The first step in analysis involved loading the Excel files and organizing the raw data into an internal Python data structure. The Analyzer class utilized the Pandas library (pd.read_excel) to import all sheets of the Excel workbook at once. Each child’s worksheet was read into a Pandas DataFrame, and then the data were transformed into a nested dictionary (nested) for convenient access in code. In this nested dictionary, the hierarchy was structured as follows:

• Top level: Child name (e.g., “Alexandros” or “Andreas”).
• Second level: Cognitive variable, inhibitory control, or working memory.
• Third level: Phase and trial identifier as extracted from the first column in the Excel, indicating which trial number and phase the data correspond to.
• Bottom level: A list of trial outcome dictionaries, each mapping item indices to the child’s response on that item (e.g., {“1”: “√”, “2”: “x”, …, “15”: “√”} for a given trial). Typically, each phase-trial combination had one such dictionary (since most trials were single instances per phase), but the data structure allowed multiple dictionaries in a list in case a phase had more than one trial entry.

In essence, this data structure captured every recorded response in a machine-readable form: one could query, for example, nested [“Alexandros”] to retrieve the detailed pattern of correct/incorrect answers that Alexandros gave in his first working memory trial at baseline. Organizing the data in this hierarchical dictionary greatly facilitated the next steps of analysis, as it enabled iterating through each child, each cognitive skill, and each phase in a coherent manner.

3.3.2. Computation of Accuracy Metrics

Using the organized data, the Analyzer computed performance metrics focusing on accuracy ratios—specifically, the proportion of correct responses out of total responses for each trial and each phase. For every trial entry, the number of “√” (correct) symbols was counted and divided by the total number of attempts in that trial, yielding a decimal between 0 and 1 representing the accuracy for that trial. For example, if a trial had 15 sub-items and a child got 9 correct, the trial’s accuracy would be 9/15 = 0.60 (60%). The script automated this calculation across all trials. An illustrative output of the set of labeled scores (or feature–value pairs) from this procedure for child might be:

• {
• “1 Β”: 0.20,
• “2 Β”: 0.40,
• “1 I”: 0.67,
• “1 G”: 0.80,
• “1 F”: 0.75
• }

The above set indicates that the child answered 20% of items correctly in the first baseline trial, 40% in the second baseline trial, 67% in the first intervention trial, 80% in the generalization trial, and 75% in the follow-up assessment. These per-trial accuracy ratios captured the trajectory of performance within each phase for that individual. The analysis code also included functionality to plot these individual trial performances over time for each child (yielding a learning curve), though for the purposes of summarizing results, attention was directed more toward phase-level aggregates.

To obtain a clearer summary of intervention effects, the Analyzer next computed mean accuracy per phase for each cognitive skill. This was done in two stages: first, for each child, the accuracy percentages of all trials within a given phase were averaged to produce that child’s overall performance in that phase for a given variable. For instance, if Alexandros had two baseline trials for cognitive flexibility (say 20% and 40% accuracy) and one intervention trial (say 67% accuracy), his mean baseline accuracy for cognitive flexibility would be 30%, and his intervention accuracy 67%. Likewise, Andreas’ data would be averaged within phase. In the second stage, the two children’s phase means were averaged together to yield a group mean for each phase and variable. By averaging at the child level first, each participant was given equal weight in the group statistics (this avoids one child skewing the average if they happened to have more trials or items than the other).

The outcome of this computation was, for each of the three executive functions, a set of four mean accuracy values—one for Baseline, one for Intervention, one for Generalization, and one for Follow-up. These values represent the percentage of correct responses achieved on average by the participants in each phase, for each targeted skill.

3.3.3. Visualization

Finally, the analysis produced graphical summaries to visualize the results. Bar charts were generated to compare the mean accuracy across phases, with separate bars for each cognitive skill. In these plots, the x-axis listed the four phases (Baseline, Intervention, Generalization, Follow-up) and the y-axis indicated the average percentage of correct responses. Each executive function was depicted in a distinct color (e.g., blue for cognitive flexibility, orange for inhibitory control, green for working memory). Error bars were not applicable given the single-subject design and very small sample, but numeric labels could be overlaid on each bar to show the exact percentage. Legends were included to identify which bar color corresponded to which skill. This visual format made it easy to see overall improvement trends and to compare the magnitude of gains between the different executive functions. For instance, one could readily observe if one skill (say, inhibitory control) showed a larger improvement from baseline to intervention than another. The plotted results support the quantitative findings, which are described in detail in the Section 4 below.

4. Results

4.1. Baseline Performance

At baseline, both participants demonstrated low levels of correct responding on tasks across all three targeted executive functions. The group mean accuracy (averaging the two children) in the Baseline phase was approximately 27% for Cognitive Flexibility, 31% for Inhibitory Control, and only 15% for Working Memory. In other words, on average less than one-third of the responses were correct for cognitive flexibility and impulse control tasks, and only about one-seventh for working memory tasks without any intervention. These low baseline scores are indicative of the challenges the children had in these cognitive domains prior to training—particularly in working memory, which emerged as the weakest of the three skills at the outset. The minimal differences between the cognitive flexibility (27%) and inhibitory control (31%) baseline levels suggest that both of those skills were similarly underdeveloped, whereas working memory lagged further behind. This provided a clear need and opportunity for improvement through VR intervention.

4.2. Intervention Phase Performance

During the Intervention phase, substantial improvements were observed in all three executive functions for both children. With the support of the VR training, the participants’ accuracy rates more than doubled relative to baseline. The average accuracy in Cognitive Flexibility rose to about 62%, in Inhibitory Control to about 70%, and in Working Memory to about 64%. Thus, the children were able to answer roughly two-thirds of the task items correctly on average when engaging with the VR intervention, a marked improvement from the roughly one-quarter correct at baseline. Notably, the largest relative gain during intervention was seen in working memory—jumping from ~15% to ~64% correct (over a four-fold increase in success rate). Inhibitory control and cognitive flexibility also showed robust gains (approximately doubling their percentages). These findings indicate that VR-based training had an immediate and positive effect on all targeted skills. The fact that working memory, initially the weakest area, improved to a level comparable with the other skills by the end of the intervention phase suggests that the adaptive memory exercises in VR were particularly effective. Overall, performance during intervention stabilized at a moderate-to-high accuracy range (roughly 60–70%), confirming the effectiveness of the VR program in improving executive function task performance in the short term.

4.3. Generalization Phase Performance

In the Generalization phase, when the children were tested on variations of the tasks (new contexts and stimuli), their high performance not only persisted but, in some cases, further improved. The group mean accuracy for Cognitive Flexibility increased to approximately 77% in the generalization trials, indicating that the participants could flexibly apply rule-switching skills even with new task elements—in fact, slightly better than they did during the intervention itself. The most striking result was for Inhibitory Control, which climbed to an average of about 95–96% correct in the generalization phase. This means that nearly all inhibitory control task responses were correct when the children were faced with novel scenarios requiring impulse regulation. Such a near-ceiling performance suggests that the skill of response inhibition was not only learned but mastered to a degree that it transferred robustly to a different context. Working Memory performance in the generalization phase was around 74%, representing a modest additional improvement over the intervention phase (which was ~64%). The working memory tasks with new stimuli still saw the children getting roughly three-quarters of the items correct on average. These generalization results demonstrate that skills learned in VR showed strong transfer to untrained contexts: both participants were able to carry over their improved cognitive flexibility and impulse control almost seamlessly, and their working memory gains also held up well in new situations. The impressive performance in inhibitory control—approaching 100% accuracy in a novel setting—underscores how effective the intervention was for that domain. In summary, the generalization phase provided evidence that the VR intervention’s benefits were not limited to the specific training scenarios, but generalized to tasks differing from the training, a crucial finding for real-world applicability.

4.4. Follow-Up Performance (Retention)

In the Follow-up phase, conducted after an interval without practice, the children’s performance remained high, indicating a significant degree of skill retention. In fact, Cognitive Flexibility showed a further slight improvement: the average accuracy rose to about 87% correct in follow-up, even higher than in the generalization phase. This suggests that the children may continue to consolidate their cognitive flexibility skills post-intervention or at least retained it so well that natural maturation or continued informal practice led to ongoing improvement. Inhibitory Control was maintained at approximately 95% accuracy in the follow-up (virtually unchanged from the generalization phase). The near-perfect impulse control observed after training was essentially sustained over time, which is a strong indication that this skill was firmly learned; the participants continued to perform almost flawlessly on inhibition tasks even without ongoing intervention or support.

Working Memory was the one area that exhibited a slight decline at follow-up. The average working memory accuracy decreased from ~74% in the generalization phase to about 65% in the follow-up phase. This drop represents a partial loss of the gains that had been achieved: some forgetting or skill decay seems to have occurred in the absence of continued practice, bringing working memory performance down to roughly the level it was during the intervention phase (which was ~64%). Importantly, however, the follow-up working memory score of ~65% was still much higher than the baseline level of 15%. In other words, despite the regression from the peak performance, the children retained a substantial improvement in working memory relative to where they started. The modest decline suggests that some booster sessions or extended practice might be beneficial to fully solidify working memory improvements over the long term. It is not uncommon in skill training studies to see the most cognitively demanding skill (here, working memory) show a bit of fading without reinforcement, and this outcome aligns with that pattern. Aside from this slight backslide in one area, the follow-up results overall indicate that the gains from the VR intervention were largely preserved: both participants continued to perform far above baseline levels on all tasks even after a period with no intervention. The diagrams displaying the analytical results for both children per session are presented in Figure 8, while Figure 9 portrays the average box plots per phases and executive functions.

4.5. Key Findings

Taken together, the results show clear evidence that VR intervention led to meaningful improvements in executive function task performance for these children with ASD. The largest gains in absolute terms were observed in Inhibitory Control, which went from ~31% at baseline to ~95% in the post-training assessments—an improvement of over 60 percentage points to near-perfect accuracy. Cognitive Flexibility also improved dramatically (from ~27% to ~87% by follow-up), representing a steady upward trajectory through each phase. Working Memory saw the most pronounced improvement during the intervention itself (15% to 64%), and although its retention was not complete (falling to 65% from a high of 74%), the end result was still a roughly 50 percentage-point net gain over baseline. Notably, the generalization of skills was successful across the board: all three executive functions, especially inhibitory control, transferred effectively to novel tasks outside the exact training regimen. The sustained high performance in follow-up for cognitive flexibility and inhibitory control, and the partial sustenance for working memory, suggest that the children truly learned these cognitive skills and were not just briefly performing well due to immediate practice effects.

From a broader perspective, these outcomes underscore the efficacy of immersive VR-based interventions for enhancing executive functions in ASD. The improvements observed in this two-participant case series are in line with patterns reported in the literature. For instance, controlled studies have found that children with ASD who undergo VR cognitive training exhibit significantly greater gains on attention and memory tests compared to those using traditional interventions. In one example, VR-trained groups improved about 25–30% on working memory tasks and showed ~20% better attention (error reduction in continuous performance tests), versus minimal changes in control groups. The present study’s findings—such as the dramatic rise in working memory accuracy and the near-elimination of impulsive errors—echo these results, suggesting that VR provides a potent and engaging medium for cognitive skill development. The combination of realistic yet controlled environments, game-based engagement, and real-time feedback likely contributed to the strong outcomes observed. In summary, the Methodology and Results of this intervention confirm that a well-designed VR program can lead to substantial improvements in executive functioning for children and adolescents on the autism spectrum, with improvements that generalize beyond the training context and largely persist over time. These results offer promising evidence in support of Virtual Reality as an effective tool for cognitive enhancement in ASD, warranting further research with larger samples and controlled designs to fully establish its benefits.

5. Discussion

5.1. Discussion on the Results

The present study investigated the effectiveness of a targeted VR intervention for enhancing executive functions in children with ASD. In contrast to review-based studies, the discussion presented is grounded in primary empirical data derived from the experimental implementation of the VR intervention.

The results of the study indicate that the VR intervention led to measurable improvements in the core executive functions that were targeted, specifically working memory, cognitive flexibility, and inhibitory control. Compared to the baseline phase, participants demonstrated improved performance during the intervention, suggesting that systematic engagement with immersive and interactive VR activities can enhance critical cognitive processes in children with ASD. In particular, increases were observed in accuracy and information retention in working memory tasks, reductions in impulsive responses in inhibitory control tasks, and improved ability to adapt to changing demands in cognitive flexibility tasks.

Furthermore, follow-up retention measurements conducted after the completion of the intervention revealed that some of the cognitive gains were maintained over time, supporting the sustainability of VR-based interventions.

The findings of the present study complement the existing literature, which has primarily focused on social skills or the general feasibility of VR applications, by providing experimental data that document the targeted enhancement of executive functions through VR. This research reinforces the view that the effectiveness of VR lies not solely in the technology itself, but in the pedagogical and therapeutic design of the intervention, including clear structure, gradual increases in task difficulty, immediate feedback, and the use of reinforcement.

From a methodological perspective, the single-case experimental design allowed for in-depth monitoring of each participant’s individual progress and highlighted differences in responsiveness to the intervention, reflecting the heterogeneity of the autism spectrum. The findings suggest that VR interventions are more likely to be effective when they are tailored to the individual needs and learning pace of each child.

Moreover, from a theoretical standpoint, the results support the view that the effectiveness of VR lies not merely in the technology itself but in the pedagogical and therapeutic design of the intervention. The integration of repetition, graduated difficulty, clear rules, and immediate feedback appears to have supported meaningful executive function training. Additionally, the object-centered and low-sensory-load design of the virtual environments likely reduced extraneous cognitive load, facilitating focused engagement with executive processes.

A closer examination of the follow-up data reveals a partial decline in working memory performance relative to the generalization phase, although performance remained substantially elevated compared to baseline. This pattern warrants careful interpretation. Working memory is widely recognized as one of the most cognitively demanding executive functions, requiring continuous updating, temporary storage, manipulation of information, and resistance to interference. In children with ASD, these processes are often particularly vulnerable to cognitive fatigue and to the discontinuation of structured support. The slight attenuation observed during follow-up is therefore consistent with the known sensitivity of working memory gains to the absence of continued rehearsal. Importantly, the intervention protocol was designed as a structured, time-bound training model within a single-case experimental framework aimed at establishing functional relationships rather than optimizing maximal long-term dosage effects. The sustained improvement above baseline levels indicates that the intervention was sufficient to induce meaningful cognitive change. Nevertheless, future studies could explore extended duration protocols or the incorporation of periodic booster sessions in order to further consolidate and stabilize working memory outcomes over time.

Regarding the adaptive difficulty mechanisms embedded in the VR intervention, task progression followed a structured, performance-contingent escalation model grounded in behavioral thresholds, rather than opaque algorithmic recalibration. Difficulty adjustments were implemented through systematic increases in rule complexity, memory load, sequencing demands, and conditional response requirements. For example, in cognitive flexibility tasks, children initially responded to simple spatial or categorical discrimination (e.g., identifying whether an object was positioned above/below or left/right relative to a highlighted stimulus). Upon achieving high accuracy within a trial block (≥80% correct responses), the task progressed to multi-rule switching conditions or more complex stimulus discrimination demands. In inhibitory control tasks, progression involved increasing rule conflict and introducing conditional motor sequences (e.g., responding only when a specific character appeared while suppressing responses to competing cues). In working memory tasks, difficulty was scaled by increasing the number of items to be retained, the complexity of item combinations, and the requirement to recall ordered sequences rather than single elements. This structured performance-based threshold ensured both clinical transparency and reproducibility, while maintaining an optimal challenge level tailored to each child’s evolving cognitive capacity.

5.2. Discussion on Methodological Scope and Generalizability

The present study was conducted using a single-case experimental design (SCED) with two participants, where each child served as their own control across repeated phases. While this design allows for strong within-subject comparisons and detection of functional relationships between intervention and performance shifts, it inherently limits external validity and statistical generalization.

The improvements observed across executive function domains should therefore be interpreted as preliminary and exploratory rather than representative of the broader ASD population. The purpose of this investigation was not to establish population-level efficacy but to evaluate the feasibility, behavioral responsiveness, and mechanistic potential of the proposed VR-based intervention within a structured experimental framework.

Moreover, given the design of the study and the limited sample size, implementing a full component-level ablation (e.g., systematically removing adaptive difficulty, reinforcement, or immersion across counterbalanced conditions) was not methodologically feasible without compromising the internal validity and stability of the intervention protocol.

Future research should expand this work using larger sample sizes, randomized controlled designs, and multi-site validation in order to statistically confirm generalizability and establish effect robustness across heterogeneous ASD profiles. From a methodological perspective, one could argue that integrating physiological measures in the future (e.g., EEG recordings, eye-tracking) would help explain and validate the underlying neural mechanisms, although this would require the redesigning of the scientific protocol.

5.3. Discussion on Limitations

Despite the encouraging results, the study has certain limitations, such as the small sample size and the absence of a control group, which restrict the generalizability of the conclusions. The findings of the present study should be interpreted in light of several methodological constraints. The use of a single-case experimental design with two participants limits statistical generalizability beyond the examined cases. Although repeated within-subject measurements strengthen internal validity and allow detection of functional relationships across phases, the results cannot be considered representative of the broader ASD population.

The absence of a control or comparison group further restricts causal inference. While phase-based performance shifts suggest an association between the VR intervention and executive function improvements, alternative explanations such as maturation, repeated task exposure, or practice effects cannot be fully ruled out.

In addition, outcome measures were primarily based on task-specific accuracy within the VR environment. Although generalization and follow-up phases were incorporated to examine transfer and maintenance, broader ecological validity was not systematically assessed through standardized neuropsychological instruments or real-world executive functioning evaluations. Furthermore, the heterogeneity inherent to Autism Spectrum Disorder suggests that responsiveness to immersive VR interventions may vary depending on cognitive profile, sensory sensitivity, and motivational factors.

6. Conclusions

The present findings should be interpreted in light of several restrictions. First, the study was designed as a pilot investigation employing a single-case experimental design, aiming primarily to explore feasibility and within-participant performance changes rather than population-level efficacy. The limited sample size restricts statistical generalization to the broader population of children with autism spectrum disorder. However, the repeated-measures structure across clearly defined experimental phases allowed each participant to serve as their own control, supporting the identification of functional relations between the intervention and behavioral outcomes. Consequently, the results should be considered preliminary and hypothesis-generating, providing initial evidence to inform future large-scale studies employing randomized controlled methodologies and larger multisite samples.

Future research could extend the present findings through studies with larger samples, randomized experimental designs, and long-term follow-up, in order to examine the maintenance and generalization of cognitive gains in natural educational and family environments. In addition, the combined use of VR with established intervention approaches, such as Applied Behavior Analysis (ABA), is of particular interest, as previous research suggests that such combinations may further enhance intervention effectiveness. Practical implementation issues, including professional training, cost, and the integration of VR into school and therapeutic settings, also constitute critical directions for future investigation.

In summary, the present study demonstrates that a structured Virtual Reality intervention can lead to meaningful improvements in the executive functions of children with ASD. By providing empirically grounded results, this work complements existing research and reinforces the role of VR as an effective cognitive enhancement tool in ASD, with significant implications for special education and clinical practice. The findings should be viewed as proof-of-concept evidence supporting the feasibility of immersive VR-based EF training in ASD, forming the foundation for future large-scale validation studies. With larger samples and randomized controlled designs future studies could systematically implement feature-level ablation to isolate the contribution of adaptive difficulty, reinforcement mechanisms, and immersive embodiment.