The review was conducted in accordance with PRISMA 2020 guidelines (Page et al., 2021) to provide transparency and replicability, and used the SPIDER framework (Sample, Phenomenon of Interest, Design, Evaluation, and Research type) to achieve a balance between methodological rigor and the different types of evidence that are typical of inclusive education (Cooke et al., 2012). The SPIDER framework was selected because it accommodates the interpretive, design-oriented, and experiential elements typical of inclusive-education research, offering greater sensitivity than PICO for multimodal interventions involving VR, AI, and SEN populations. This design prioritized teaching and human experience outcomes. The methodological position is founded on humanistic and participatory values, recognizing that research with SEN populations needs to be very alert to voice, dignity, and autonomy (O'Connor et al., 2023). Technology is not the primary focus of design; rather, the empowerment of learners and the socio-educational impact are given priority.

Three academic databases were systematically searched on 24 October 2025 to ensure comprehensive coverage of contemporary immersive-technology development: Scopus, Web of Science (WoS), and Education Resources Information Center (ERIC). Scopus and Web of Science were selected due to their extensive indexing of high-impact education, computer science, and interdisciplinary technology journals, while ERIC was included for its depth in pedagogical and special education scholarship. To capture the intersection of VR, Agentic AI, and inclusion, Boolean operators were applied to keywords relating to:

Search strings were adapted minimally for database-specific syntax while maintaining conceptual equivalence. The search spanned from January 2020 to January 2025, aligning with recent surges in immersive adoption following pandemic-driven digital transformation (Klimanov and Tretyak, 2024).

The inclusion criteria include the following:

The review deliberately restricted AI to agentic, adaptive, or affective systems because these forms of intelligence can respond in real time to learner behaviors, emotions, and cognitive load; an essential requirement for analyzing co-agency and equitable immersion.

Equally, the exclusion criteria include the following:

The initial search produced n = 144 records. After removing records for other reasons, including duplicates and language used (n = 3), a total of n = 141 remained for screening. The titles were screened for eligibility, and n = 94 records were excluded. Additionally, the abstracts were read, and n = 17 were eliminated. Out of the n = 30 remaining papers, n = 3 full-text papers could not be retrieved and were therefore eliminated. The remaining full texts of n = 27 papers were assessed on whether they answered the research questions posed, demonstrated sound methodological rigor, and had valid population samples. All screening stages: title, abstract, and full-text review were independently conducted by two reviewers, with disagreements resolved through discussion. Only 16 papers met the final inclusion criteria, and therefore, their study findings were extracted. This process is summarized in Figure 1.

To ensure inter-reliability, coding was performed independently by two reviewers, yielding 92% intercoder agreement. This high level of consensus arguably reflects shared familiarity with inclusive learning values and minimal interpretive drift in the appraisal of studies. Using an adapted MMAT (Hong et al., 2018a,b), a quality appraisal was performed to assess the methodology's integrity. Detailed itemized scoring for empirical studies was based on five specific criteria: research presentation, appropriateness of design, clarity of data collection and analysis, transparency, and applicability. Two independent reviewers assigned each study a quality rating of High, Moderate, or Low, and any discrepancies were resolved through a consensus process. In sum, nine studies received high-quality ratings, five were of mediocre quality, and eight were considered low quality. This was largely due to a small sample size or lack of methodological transparency. Cases receiving MMAT ratings were not excluded from the meta-synthesis; however, results were downstream with cautious weighting applied, especially when an effectiveness claim or broad generalizability was involved. The views of all studies, nonetheless, spurred further understanding of the context. This rigorous appraisal primarily enhanced credibility, dependability, and trustworthiness underpinning the collaborative integration of VR–AI into inclusive education.

Data extraction captured variables such as the country of origin, methodology adopted, type of special need or context, pedagogical applications, and learning outcomes. Importantly, ethical and implementation challenges were analyzed as critical variables, not optional reporting additions, because immersive technologies can both empower and unintentionally marginalize learners who are neurodivergent, sensory-sensitive, or mobility-challenged, especially if poorly designed (Parong and Mayer, 2021).

Spatially, the 16 empirical studies reflect a multi-regional but uneven distribution of immersive inclusion innovation, as depicted in Figure 2. As confirmed by the geographic mapping, the United States and Canada lead adoption (31.25%) (Yeganeh et al., 2025; Hamed et al., 2024; Smith et al., 2024; Shu and Gu, 2023; Lowell and Yan, 2024), followed by China (18.75%) (Strielkowski et al., 2025; Hutson and McGinley, 2023; Zahid Iqbal and Campbell, 2023), with the remaining contributions dispersed individually across the United Kingdom (Tracy and Spantidi, 2024), Spain (Song et al., 2023), India (Piki et al., 2022), Ireland (Strousopoulos et al., 2024), Greece (Lowell and Yan, 2024), Taiwan (Liao et al., 2024), Malaysia (Mokmin and Ridzuan, 2023), and the United Arab Emirates (Lorenzo Lledó, 2025) (each 6.25%). The heavy bias toward North America and East Asia in the number of studies not only points to the differences in access to immersive technologies but also to the structural research bias that makes the findings less representative. This pattern suggests hopeful global adoption; however, the lack of empirical studies from Sub-Saharan Africa and Latin America points to the continued existence of hidden structural inequalities in access to immersive technologies. This creates a geographical knowledge gap that limits generalizability and raises concerns over the fairness of the global distribution of immersive learning innovations. If the imbalances are not addressed, the benefits of VR–AI will be monopolized by the already affluent education systems, which is precisely the situation that contradicts the equity commitments of SDG 4 and the UNCRPD. Therefore, immersive education research is no longer just a Global North activity. However, still depends on regional privilege, which means that generalizing the evidence should be contextually cautious and accompanied by targeted capacity building.

The studies included in the review presented a variety of educational levels, thereby capturing both the school and the higher education sectors. Several studies have been conducted in primary and secondary school settings, particularly those examining the development of motor coordination in children with specific learning disabilities (SLD) and the effectiveness of social communication interventions in various schools (Mokmin and Ridzuan, 2023; Song et al., 2023; Shu and Gu, 2023). On the other hand, a significant amount of research has come from university or postgraduate education cases, for instance, occupational therapy training, English language learning, and testing experimental prototypes in university labs or immersive CAVE-based learning environments (Yeganeh et al., 2025; Strielkowski et al., 2025; Tracy and Spantidi, 2024; Liao et al., 2024; Hutson and McGinley, 2023; Zahid Iqbal and Campbell, 2023; Strousopoulos et al., 2024; Lowell and Yan, 2024). Studies conducted in higher education demonstrated more mature and technically sophisticated VR–AI implementations compared to school-level interventions, which were generally smaller in scale and more exploratory. This stratification suggests that infrastructure and research capacity strongly influence the types of immersive inclusion models that can be deployed across sectors. This distribution suggests that the early adoption of integrated VR–AI technologies has been most pronounced within higher education institutions, where research capacity and technical infrastructure are more readily available. Nonetheless, there is growing, albeit smaller-scale, application in school-age SEN contexts, demonstrating expanding inclusivity across educational tiers.

The study also analyzed the various methodologies used within the included studies, examining how the convergence of VR and Agentic AI can reimagine the possibilities of inclusive education, as depicted in Figure 3.

The 16 empirical studies reviewed demonstrated a wide range of methodological diversity, reflecting the interdisciplinary nature of immersive educational research. Several studies employed quasi-experimental or experimental designs to examine SLD motor learning (Mokmin and Ridzuan, 2023), ASD social skills training (Chalkiadakis et al., 2024), smartphone-based VR prototypes (Sonawane et al., 2024), and GPT-assisted virtual classrooms (Tracy and Spantidi, 2024). Experimental and quasi-experimental studies have consistently produced stronger evidence of effectiveness, particularly for motor skills training and cognitive load reduction. Others used mixed methods, integrating quantitative measures and qualitative reflections, as in Yeganeh et al. (2025) and Mokmin and Ridzuan (2023), which evaluated inclusive language learning in metaverse settings. Qualitative multi-site studies (collaborative), including (Piki et al., 2022), explored social-emotional and pedagogical engagement. These qualitative and prototype studies contributed richer insights into emotional safety, learner experience, and co-agency dynamics. Complementing these were design-based usability investigations that iteratively refined immersive systems through user testing, including the CAVE teaching assistant with agentic support (Jia et al., 2024). Notably, none of the included studies employed longitudinal designs, and very few provided follow-up assessments, which limits understanding of the durability of VR–AI benefits once novelty effects diminish. While methodological innovation is clearly progressing, longitudinal validation remains scarce. Thus, methodological heterogeneity strengthens the field but also complicates direct comparison across outcome domains.

Importantly, although all included studies involved some form of intelligent support, not all systems operated with the same degree of autonomy. Many applications relied on assistive or adaptive mechanisms that guided learners through predefined pathways or rule-based adjustments (Mokmin and Ridzuan, 2023; Chalkiadakis et al., 2024; Piki et al., 2022), whereas fewer studies described movement toward genuine pedagogical co-agency, where the system interpreted learner states, initiated instructional adjustments, and engaged learners through more autonomous interaction (Yeganeh et al., 2025; Strielkowski et al., 2025; Sonawane et al., 2024; Tracy and Spantidi, 2024). Figure 4 summarizes the reported learning outcome domains.

The analysis of the 16 empirical papers shows that 12 studies explicitly measured engagement and motivation (Yeganeh et al., 2025; Mokmin and Ridzuan, 2023; Tracy and Spantidi, 2024; Piki et al., 2022; Liao et al., 2024; Lorenzo Lledó, 2025; Song et al., 2023; Hutson and McGinley, 2023; Hamed et al., 2024; Smith et al., 2024; Shu and Gu, 2023; Zahid Iqbal and Campbell, 2023), often linking these outcomes to immersive presence, learner satisfaction, or enjoyment. Moreover, eight studies examined cognitive performance and accuracy (Yeganeh et al., 2025; Mokmin and Ridzuan, 2023; Tracy and Spantidi, 2024; Liao et al., 2024; Lorenzo Lledó, 2025; Hutson and McGinley, 2023; Zahid Iqbal and Campbell, 2023; Strousopoulos et al., 2024) through the lenses of memory retention, problem-solving, and concept understanding in immersive contexts. In addition, six studies have focused on social-communication skills (Yeganeh et al., 2025; Mokmin and Ridzuan, 2023; Tracy and Spantidi, 2024; Liao et al., 2024; Lorenzo Lledó, 2025; Hutson and McGinley, 2023; Zahid Iqbal and Campbell, 2023; Strousopoulos et al., 2024), specifically among learners with autism or communication difficulties. Motor-skill development was the subject of two studies (Mokmin and Ridzuan, 2023; Song et al., 2023), which focused primarily on primary-level interventions for students with SLDs, thus indicating an under-explored domain. Finally, six papers addressed confidence and affective safety (Yeganeh et al., 2025; Mokmin and Ridzuan, 2023; Tracy and Spantidi, 2024; Lorenzo Lledó, 2025; Smith et al., 2024; Shu and Gu, 2023), showing that agentic AI can reduce anxiety and increase self-efficacy in virtual environments. All the findings presented above indicate that the integration of VR and AI has a strong positive impact on engagement and motivation. However, evidence for deeper cognitive or psychomotor outcomes is relatively limited, thus calling for longitudinal, multi-domain validations across all educational levels. However, pedagogical benefits were highly dependent on the AI agent's sophistication. Studies using basic rule-based agents or non-adaptive scripts showed significantly weaker outcomes, demonstrating that ‘VR alone' is insufficient without meaningful agentic intelligence.

The analysis of the 16 papers also revealed the pedagogical applications of integrated VR and Agentic AI in inclusive learning environments. These were summarized in Table 1.

The integration of VR and Agentic AI for pedagogical purposes demonstrated a complex ecosystem of adaptive, affective, and multisensory learning affordances. Most of the research relied on personalization and scaffolding that were sensitive to learners' emotions, demonstrating how the AI agents would adjust task difficulty and emotional tone to meet the learners' needs. Immersive simulations enabled mastery, while metaverse classrooms offered opportunities for collaboration, social acceptance, and co-agency. Several studies have placed a UDL-aligned multisensory design at the forefront, thereby catering to a wide range of cognitive and physical abilities. Across studies, adaptive personalization and emotion-aware support emerged as the most effective mechanisms. Applications that integrated real-time feedback loops, such as adaptive pacing or frustration detection, produced higher reported gains in learning outcomes than static VR simulations.

The most common barriers are infrastructure disparities and teachers' lack of preparedness, which have a greater impact on integration success than the sophistication of the hardware. The persistent accessibility gap indicates that many immersive systems remain far from compliance with UDL principles, while ethical and algorithmic dilemmas highlight the bigger risks of bias, surveillance, and exclusion. Moreover, misalignment between finances and policies undermines sustainability, leaving the promising prototypes on the pilot stage. Therefore, the learners' struggle has become evident, necessitating adaptive pacing and ergonomic design. Challenges were not evenly distributed across contexts: resource-constrained settings primarily reported hardware, cost, and connectivity issues as the main obstacles, whereas more affluent areas primarily referred to ethical issues, data governance, and teacher preparedness. Of the 40 studies that mentioned data governance, four studies (Mokmin and Ridzuan, 2023; Chalkiadakis et al., 2024; Tracy and Spantidi, 2024; Piki et al., 2022) discussed the governance of biometric data. No equity or bias checks were performed on AI decision-making, indicating a significant empirical blind spot given the vulnerable position of SEN populations. It seems that the aspects of the implementation process that hinder progress are more closely linked to the context than to neutrality. Furthermore, the results suggest that genuine equity comprises infrastructural support, inclusive teaching, open and fair governance, and ethical responsibility. It is not just a highly desirable goal of sophisticated technology. These were summarized in Table 2.

Taken holistically, the findings indicate strong pedagogical potential for the collaboration between VR and AI as co-agents; they also show the major flaws in the quality of the evidence, accessibility practices, geographic representation, and ethical evaluation. Such patterns across different areas of study form the basis for the ensuing discussion, which examines the role of VR–AI integration in enhancing or limiting inclusiveness and equity of learning in varied educational settings.

Across the reviewed studies, VR combined with Agentic AI offered strong leverage for engaging learners and strengthening motivation and emotional investment. Learners reported more memorable experiences when virtual environments responded dynamically to their actions and affective states (Tracy and Spantidi, 2024; Yeganeh et al., 2025). This responsiveness reflected co-agency, whereby the system functioned as an adaptive partner, shaping personalized timing, difficulty, and context-sensitive feedback. Such interaction resonates with constructivist principles, which emphasize meaning-making through active experience (Vygotsky, 1978).

Engagement and motivation emerged as the most frequently reported outcome, reinforcing the argument that agency can serve as a precursor to presence and resonance (Mokmin and Ridzuan, 2023; Piki et al., 2022). Nevertheless, the studies also indicate that fascination alone is insufficient: durable educational gains depend on structured scaffolding and reflective debriefing (Makransky and Petersen, 2021). Without these pedagogic anchors, immersion risks remaining superficial.

Beyond engagement, several studies have reported improvements in cognitive performance, accuracy, and effectiveness when adaptive agents regulated cognitive load in real time (Chalkiadakis et al., 2024; Piki et al., 2022). By monitoring user interaction, agents adjusted task difficulty and stimulus intensity to sustain concentration and support retention. This aligns with refined CLT accounts for immersive contexts, in which the agent assumes a hybrid role between challenge and comfort, reducing overload while still sustaining productive effort. Notably, this adaptivity appeared particularly valuable for neurodivergent learners who often experience cognitive fatigue in traditional classrooms.

Social-communication gains also emerged, with VR–AI environments functioning as preparatory spaces for learners with ASD and social anxiety. Studies reported that avatar-based interaction reduced fear of judgment and enabled learners to practice dialogue and empathy in psychologically safer conditions (Yeganeh et al., 2025; Mokmin and Ridzuan, 2023). Similarly, increased confidence and affective safety were reported when socially aware AI tutors supported emotional regulation and reduced anxiety during difficult tasks (Tracy and Spantidi, 2024).

However, immersive co-agency also carries limits. Although less common, (Parong and Mayer, 2021) suggested that fine motor coordination could be disrupted for learners with physical or perceptual difficulties, indicating that embodied interaction may not be universally enabling without careful calibration. Moreover, benefits depended heavily on the sophistication of the AI: scripted or rule-based agents tended to yield weaker and shorter-lived gains than adaptive, emotion-aware systems. This reinforces a central inference across the evidence base; immersion alone is insufficient without real-time pedagogical intelligence.

The studies analyzed have shown that VR coupled with Agentic AI provides strong pedagogical leverage for engaging learners and fostering their motivation and emotional investment. Students reported more memorable experiences when virtual environments dynamically responded to their actions and emotional states (Tracy and Spantidi, 2024; Yeganeh et al., 2025). Such responsiveness paralleled agentic co-agency and thus an aura of companionship between the learner and the system, which, in turn, facilitated personalized timing, adaptive difficulty, and context-sensitive feedback. It enabled interactive, self-directed learning, thus adhering to constructivist principles, which emphasize the creation of understanding through active experiences (Vygotsky, 1978).

The findings indicate that agentic AI in immersive learning should be interpreted as an emergent capability rather than a consistently realized condition across the evidence base. While some studies suggest more autonomous, emotion-aware pedagogical agents capable of co-regulating learner experience (Yeganeh et al., 2025; Strielkowski et al., 2025; Sonawane et al., 2024; Tracy and Spantidi, 2024), a substantial portion of the literature reflects transitional systems that remain primarily assistive or adaptively responsive (Mokmin and Ridzuan, 2023; Chalkiadakis et al., 2024; Piki et al., 2022). Consequently, immersive co-agency is better framed as a developmental trajectory, in which adaptive support can serve as a pragmatic precursor to more advanced learner-AI partnership when guided by pedagogical intentionality and reflective scaffolding (Makransky and Petersen, 2021; Vygotsky, 1978).

The second theme concerns the pedagogical shifts underpinning immersive inclusion. The results identified 10 broad categories of pedagogical applications, reflecting an emerging move toward holistic, personalized learning design. Dominant applications included adaptive learning, experiential simulation, and emotion-aware tutoring, collectively forming a triadic relationship among the learner, the agent, and the environment. In this configuration, AI does not merely deliver content; it increasingly acts as a responsive learning partner. Consequently, teacher-centered models are challenged by new questions of shared agency, autonomy, and trust.

Adaptive learning and personalized instruction have emerged as foundational components across many interventions, with AI agents adjusting difficulty, response timing, and sensory input dynamically (Tracy and Spantidi, 2024; Yeganeh et al., 2025). Such adaptivity, historically the domain of teachers, can now scale across learning contexts, yet only if agentic responsiveness is transparent and ethically sustained.

Immersive simulation similarly strengthened experiential learning, with studies showing that metaverse classrooms and scenario-based tasks supported the practice of abstract concepts, vocational skills, and social behaviors in realistic yet controlled environments (Yeganeh et al., 2025; Piki et al., 2022). This aligns with experiential learning traditions associated with Kolb, where reflective engagement with simulated experience consolidates understanding.

Emotion-aware tutoring was also prominent in enabling inclusion (Chalkiadakis et al., 2024; Tracy and Spantidi, 2024). Through affective computing, including behavioral tracking and facial recognition, agents detected signs of frustration or disengagement and adjusted pacing accordingly. This emotional synchrony shifts AI from purely cognitive machinery to socio-emotional companionship, particularly relevant for SEN learners who rely on predictable, empathetic feedback.

Finally, multisensory and cross-modal design was repeatedly framed as a practical pathway to inclusion. Haptics, audio design, and flexible visual affordances expanded access for learners with sensory impairments and supported embodied cognition (Piki et al., 2022). Autonomous mastery learning further strengthened inclusion by enabling learners to regulate progression through metacognitive feedback loops (Piki et al., 2022). In combination, these pedagogical adaptations suggest a substantive reframing of inclusion: from standardization and placement toward personalized participation.

A consistent pattern across studies is that barriers to immersive innovation are not evenly distributed. In resource-rich contexts, the dominant concerns involved ethical governance, data protection, and teacher preparedness; in resource-constrained contexts, challenges begin with access itself. Studies from Malaysia, India, and the UAE pointed to equipment costs, network instability, and maintenance burdens that limit implementation, particularly in public or rural schooling contexts (Mokmin and Ridzuan, 2023; Yeganeh et al., 2025). Consequently, much of the empirical literature reflects privileged environments, raising epistemic equity concerns about whose realities are represented in this emerging evidence base. These disparities indicate that immersive equity is not only a technical question but a structural justice issue shaped by unequal infrastructure, funding, and policy support.

In addition, pedagogical readiness often lagged behind technical capability. Several studies reported that teachers lacked the professional understanding and pedagogical frameworks needed to integrate immersive systems meaningfully (Chalkiadakis et al., 2024). Without structured implementation and intentional instructional design, VR–AI risks becoming novelty rather than pedagogy. Accessibility gaps further complicate this reality: while many systems are described as inclusive, few demonstrate compliance with UDL in ways that substantively include learners with visual, auditory, or mobility impairments, who remain underrepresented (Mokmin and Ridzuan, 2023).

Ethical risks associated with affective and biometric data also remain unresolved. Studies have flagged unclear data management practices, opaque algorithmic decision making, and potential biases that could amplify exclusion (Chalkiadakis et al., 2024; Tracy and Spantidi, 2024). However, despite frequent mention of affective AI, formal bias audits and transparent reporting of biometric retention, processing, or consent mechanisms were largely absent, an omission that is particularly concerning given the vulnerability of SEN populations and the sensitivity of behavioral-emotion data.

Finally, learner fatigue and sensory overload emerged as subtle but important constraints. While most studies reported gains in engagement and cognitive accuracy, two studies documented increased sensory overload and cognitive fatigue under prolonged immersion, indicating that co-agency can either mitigate or intensify cognitive load depending on pacing and design quality (Piki et al., 2022). These patterns align with CAMIL, which emphasizes that deep learning arises when immersive presence is supported by emotional regulation; agentic AI potentially operationalizes this by modulating cognitive load, suggesting convergence between CAMIL and adaptive tutoring models.

Relatedly, multiple studies reported that learners struggled when interfaces were fragmented or agent behavior was inconsistent, highlighting that co-agency requires stability and predictability to support inclusion. In particular, (Piki et al., 2022) noted that disruptions in interface continuity undermined peer collaboration, (Strielkowski et al., 2025) reported that inconsistent cues from the CAVE-based NivTA assistant confused learners relying on routine, and (Yeganeh et al., 2025) described interaction delays that disproportionately affected learners who needed structured predictability for emotional regulation. Taken together, these findings suggest that immersive learning is most effective when intelligent interaction is coherent and rhythmically paced, especially for neurodivergent learners. They also extend socio-constructivist theory by illustrating that meaning can be co-constructed not only among learners and peers, but also between learners and intelligent virtual agents.

The review identified 10 frameworks and models that propose ethical and pedagogical pathways for immersive intelligence. UDL remains foundational, foregrounding sensory access and flexibility of representation (Mohamed et al., 2025). Studies applying UDL principles were positioned as moving beyond access toward agency and participation (Boot and Hughes, 2025; Mokmin and Ridzuan, 2023). In parallel, the Meta-MILE model (Yeganeh et al., 2025) conceptualizes the metaverse as a multi-layered inclusive environment for active learning, flexible AI, and co-presence. The Agentic Presence and Co-Agency Model (Sonawane et al., 2024) and Emotionally Intelligent Pedagogical Agent (EIPA) (Chalkiadakis et al., 2024) position AI as an active social actor that mediates interaction, shifting the conceptual emphasis from automation toward partnership and ethical companionship; an especially consequential move for inclusive education.

Learning theory also informed design choices. Refined CLT foregrounded task sequencing and flow, whereas Embodied Learning Theory legitimized bodily cognition for SEN learners requiring physical interaction (Mokmin and Ridzuan, 2023). Socio-constructivist assumptions were similarly extended in AI-mediated environments, where knowledge is co-created through interaction and collaboration (Piki et al., 2022). The Hybrid Immersive Pedagogy (HIP) (Sripan and Jeerapattanatorn, 2025) and Adaptive Metacognitive Scaffolding (AMS) (Tracy and Spantidi, 2024) models further emphasized reflective learning and self-regulated mastery as outcomes of AI-supported autonomy.

Overall, the review demonstrates that immersive inclusion requires integrating sensory, cognitive, emotional, and ethical dimensions into a multi-layered foundation synthesized from UDL, CAMIL, socio-constructivism, and agentic-presence models. While many frameworks treat these dimensions separately, the reviewed evidence suggests their interdependence, thereby motivating the development of a more comprehensive theory of immersive equity.

It is important to clarify that the Integrated VR–AI Equity Framework proposed in this study is inductively grounded in the empirical patterns identified across the sixteen reviewed studies, rather than imposed a priori. Core dimensions of the framework, namely pedagogical co-agency, learning outcome domains, ethical and implementation challenges, and integration frameworks, emerged consistently from a cross-study synthesis of immersive interventions employing varying degrees of adaptive and agentic AI (Yeganeh et al., 2025; Strielkowski et al., 2025; Mokmin and Ridzuan, 2023; Chalkiadakis et al., 2024; Sonawane et al., 2024; Tracy and Spantidi, 2024; Piki et al., 2022). At the same time, the framework represents a conceptual consolidation that situates these empirical insights within broader theoretical traditions, including UDL, the CAMIL, socio-constructivism, and agentic presence models (Mohamed et al., 2025; Vygotsky, 1978; Sonawane et al., 2024). In this sense, the framework functions both as an evidence-informed explanatory model and as a theory-building contribution that articulates how immersive equity is produced through the interaction of intelligence, embodiment, emotion, and governance.

In synthesis, VR and Agentic AI appear increasingly positioned as inclusion mediators that shape how learning is designed, delivered, and experienced. However, technological sophistication does not automatically translate into educational equity. Persistent disparities in infrastructure, pedagogy, and ethical governance mean that immersive inclusion can support equity or inadvertently reproduce exclusion. Consequently, equitable scaling depends on a tighter alignment of pedagogy, readiness, and governance alongside inclusive, design-based research approaches that embed accessibility and equity from conception through implementation. Furthermore, frameworks such as Meta-MILE and EIPA require broader empirical testing across diverse geographic contexts to establish robustness beyond resource-rich contexts. Ultimately, it is through the deliberate co-location of technology and ethics, immersion and empathy, that VR–AI can more credibly fulfill the promise of immersive equity in which learners, regardless of difference, can meaningfully participate, persist, and flourish within digitally mediated classrooms.

Accordingly, the Integrated VR–AI Equity Framework proposed in this review should be understood as an inductive synthesis grounded in recurring cross-study patterns, while also capturing the direction of travel toward stronger forms of agentic co-agency suggested by the most advanced implementations (Yeganeh et al., 2025; Strielkowski et al., 2025; Sonawane et al., 2024; Tracy and Spantidi, 2024). The framework does not assume that full agentic maturity is present across all current systems; rather, it integrates both present adaptive practices and emerging agentic capacities into a coherent model aligned with inclusive design logics, particularly UDL's emphasis on flexible access and participation (Mohamed et al., 2025; Mokmin and Ridzuan, 2023).

In sum, the hybrid of AI and VR has shown promise as a path toward inclusive education. However, its advantages would need to be supported by design consistency, contextual appropriateness, and the presence of ethical safeguards, among others. The integrated nature of immersive experience, smart adaptability, and accessibility could either support equity or, even worse, unwittingly undermine it. The knowledge the study provides calls for the creation of a holistic framework that will not only influence the direction of technological progress but also the workflows, policies, and future research prioritization of education.

Theoretically, this study reframes the concept of inclusion by integrating the constructivist, socio-cultural, and embodied learning paradigms with agentic intelligence theory, positioning immersive education as an interactive state of co-agency in which the learners and the AI systems jointly negotiate meaning, emotion, and control. The author reviews the existing literature by presenting emotionally intelligent pedagogical agents that range from mere cognitive support to those that can discuss sharing and ethical companionship, helping with anxiety regulation, sensory load modulation, and overall psychological safety. The Integrated VR–AI Equity Framework further contributes to the development of the learning theory by uncovering that the immersive equity comes to be only when the embodied presence, emotional regulation, cognitive load protection and accessibility infrastructures are working together; a joining that has been rarely seen in the earlier frameworks like CAMIL, UDL, socio-constructivism and cognitive load theory which have been isolating and thus addressing these elements. By foregrounding co-agency as a new pedagogical construct, the framework reconceptualizes AI not as a passive instructional tool but as an active learning partner that collaborates with learners to shape pacing, depth, and affective support. Additionally, the framework positions emotional regulation as a core mechanism of knowledge construction in immersive environments, making affect central rather than peripheral to inclusive pedagogy. Collectively, these insights transform educational technology from a tool of efficiency into a tool of empowerment, insisting that theoretical conversations about immersion must be simultaneously pedagogical, ethical, and social, and aligned with a broader pursuit of educational justice.

The framework is a policy-oriented call for national and institutional governance to regulate immersive VR–AI technologies in education in an ethical, equitable, and sustainable manner. The protection of emotional, biometric, and behavioral data is central to such governance, which necessitates establishing transparent standards for data collection, imposing restrictions on the misuse of surveillance, and conducting mandatory fairness audits to ensure that decisions made by intelligent agents are not biased. Furthermore, the framework argues that the teaching of immersive ethics and accessibility must be incorporated into teacher-training curricula and accreditation programs, rather than treated as optional add-ons, to ensure that teachers are equipped to address the complex responsibilities of emotion-sensitive and adaptive AI. Governments, NGOs, and donors must prioritize funding for low-cost VR devices, bandwidth optimization, and context-appropriate AI models. At the same time, current access is heavily tilted toward affluent regions and private institutions. This way, schools with limited resources can still participate at an affordable cost, such as through smartphone headsets or existing collaborative metaverse platforms. In addition, institutions should create strategic digital-transformation plans that align with budgeting for sustainable hardware, cross-departmental collaboration as the norm, and the embedding of ethical procurement standards. Ultimately, the framework shifts the narrative on immersive equity from a technological luxury to a basic educational right, compelling the policy landscape to change to ensure safe, accessible, and just participation for all learners, especially those most at risk of exclusion.