Evaluating interaction methods in augmented reality training

As augmented reality (AR) adoption expands across training and operational domains, interaction design has served as a key determinant of user experience (UX). Although natural interaction techniques are believed to better support AR than traditional WIMP (windows, icons, menus, and pointing devices) interfaces, direct comparisons remain limited. This study compared natural and WIMP-based interaction paradigms in an individual, high-stakes procedural AR training task to evaluate their effects on usability, workload, presence, and task performance. Thirty participants completed a within-subjects experiment using both interaction paradigms in a Tactical Combat Casualty Care AR training scenario delivered on a Meta Quest 3. Condition order was counterbalanced. Subjective outcomes included the System Usability Scale, NASA Task Load Index (NASA-TLX), and Presence Questionnaire. Objective measures included task completion time, navigation errors, missed procedural steps, corrective actions, requests for clarification, and observed interaction smoothness. Paired-samples t -tests were used to compare conditions, with exploratory analyses examining order effects and relationships between immersive tendency and presence. Natural interaction produced significantly higher usability (p = 0.009) and presence (p = 0.001) than the WIMP interface. Participants also committed fewer navigation errors, missed fewer procedural steps, required fewer corrective actions, and demonstrated smoother interactions (p < 0.05) in the natural condition. No significant differences were observed for workload, task completion time, clarification requests, or observed efficiency. An order effect indicated lower perceived workload during participants’ second session, regardless of interaction condition, suggesting a learning effect. Natural interaction improves usability, presence, and interaction quality in individual, high-stakes procedural AR training while minimizing execution errors, although it does not necessarily improve speed or decrease cognitive workload during initial use. This evidence suggests that the principal benefits of natural interaction lie in improving the overall UX quality rather than maximizing efficiency, providing guidance for the design of future AR training systems.

The rapid growth of augmented reality (AR) applications across industry, retail, manufacturing, and education is well documented, reinforced by projections of accelerating adoption and market growth that position AR as a key supporting technology for the next-generation of digital experiences (Chen et al., 2022; Kazmi et al., 2021; Khamaisi et al., 2021; Picardi and Caruso, 2024). This expansion brings concurrent demands on designers and developers to address not solely technical feasibility but also the quality of user experience (UX) in real-world contexts, as AR integrates digital content with physical environments in both real-time and spatially accurate ways (Kabir and Kang, 2024; Nikolarakis and Koutsabasis, 2024; Sumithra et al., 2024). Empirical syntheses and systematic reviews emphasize that, despite the technical advances and market momentum, UX remains unevenly considered in AR development, with UX design commonly overshadowed by device capabilities and visualization techniques (Chen et al., 2022; Hughes and Karwowski, 2025; Khamaisi et al., 2021; Nikolarakis and Koutsabasis, 2024; Wafa, 2022). Accordingly, researchers have begun discussing interaction design as an essential element in usability, engagement, and value in AR systems, arguing for user-centered frameworks, pattern-based design guidance, and rigorous UX evaluation to realize the benefits of AR at scale (Hughes and Karwowski, 2025; Kabir and Kang, 2024; Law et al., 2021; Picardi and Caruso, 2024; Sumithra et al., 2024).

In this context, interaction design is a central driver of UX in AR, shaping how users perceive, learn from, and derive value from AR experiences. AR training applications vary considerably in structure and demand. They range from declarative, knowledge-based instruction to procedural, psychomotor skill training; from low-stakes onboarding tasks to high-stakes, time-critical scenarios; and from individual to collaborative formats. The present study focuses specifically on individual, high-stakes procedural/psychomotor skill training, exemplified by military and medical first-responder simulation, in which users must execute a sequence of physical actions correctly and efficiently under time pressure. Studies across AR contexts, including retail, cultural heritage, industrial maintenance, and education, emphasize that effective interaction design goes beyond ensuring functional accuracy; it requires formulating intuitive interaction paradigms, meaningful feedback, and compelling engagement that correspond with task goals as well as with the natural 3-dimensional (3D) environment (Chen et al., 2022; Hughes and Karwowski, 2025; Kim and Ha, 2021; Nikolarakis and Koutsabasis, 2024; Sumithra et al., 2024). If AR technologies are to thrive, the role of interaction design in forming UX becomes critical, guiding developers toward user-centered review frameworks, design patterns, and validated metrics that can support the long-term growth of AR ecosystems (Kabir and Kang, 2024; Law et al., 2021; Picardi and Caruso, 2024).

Background

Building on the importance of interaction design in shaping UX, a key distinction in AR systems lies in the interaction paradigm employed. In particular, the contrast between traditional WIMP (windows, icons, menus, and pointing devices)-based interfaces and more recent natural and spatial interaction approaches has been a focal point for understanding how AR systems may better align with human abilities. Here, naturalness refers to the degree to which an interaction technique’s input mechanics correspond to the sensorimotor actions people use to manipulate objects in the physical world, with WIMP interfaces representing a low-naturalness, symbolically mediated paradigm inherited from 2-dimensional (2D) desktop computing. Natural 3D interaction methods, including embodied actions, direct manipulation of virtual content, and use of effortless gestures, gaze, and voice, offer a more effective approach in AR environments than traditional WIMP interfaces because they more closely align with how people naturally perceive and act in physical space (Jin et al., 2022; Li, 2024). Rather than relying on abstract inputs like clicking icons or browsing menus, natural interfaces leverage gestures, gaze, voice, and movement, enabling users to interact directly with digital media in a more intuitive, embodied way. This directness enhances presence and engagement by reducing the cognitive effort required to translate symbolic commands into spatial actions, a limitation that often disrupts immersion in WIMP-based systems (Jin et al., 2022; Polvi et al., 2018).

Additionally, natural interactions are inherently more adaptable, enabling systems to operate dynamically in real-time response to user context, behavior, and environment, which is key to effective AR experiences. They also support richer, multisensory feedback, such as haptic and aural cues, improving usability and accessibility to diverse users. This facilitation of immediate, first-person interaction and synchronization with human motor and perceptual abilities makes certain that natural interfaces create smoother, more responsive experiences, whereas WIMP paradigms tend to feel rigid, visually cluttered, and disconnected from the 3D nature of AR (Jiboku and Obarayi, 2023; Jin et al., 2022; Li, 2024).

A systematic literature review by Hughes and Karwowski (2025) highlights a clear trend in AR research toward natural interaction paradigms, stressing their conformity with the spatial, embodied, and context-aware nature of AR environments. Across the reviewed studies, natural interactions consistently revealed stronger support for key UX dimensions, including immersion, presence, and intuitive control. In contrast, WIMP-based interfaces, derived from traditional 2D computing environments, have been shown to impose design constraints that can hamper smooth interaction and reduce experiential richness in AR contexts. Although integrated systems combining both approaches were most common, their observed benefits were largely driven by the inclusion of natural interaction elements rather than WIMP components (Hughes and Karwowski, 2025).

Despite this growing consensus, the literature also uncovers a critical gap in that relatively few studies conduct direct, controlled comparisons between natural and WIMP interaction paradigms (Hughes and Karwowski, 2025). Much of the existing evidence is based on evaluations of isolated systems, making it difficult to draw definitive conclusions about relative performance between conditions. This gap points to the need for empirical research that systematically compares these interaction styles across key UX metrics. For the purposes of this study, UX in AR training is operationalized as the composite of perceived usability, cognitive workload, and presence experienced during task performance. These three dimensions are most consistently identified as central to AR-specific interaction quality in prior systematic reviews (Hughes and Karwowski, 2025; Khamaisi et al., 2021). This operational definition guided the selection of outcome measures reported in Section 2.

To address these gaps, the present study conducts a controlled, within-subjects experimental comparison of natural and WIMP-based interaction paradigms in an AR environment. Participants perform equivalent, goal-oriented tasks in both interaction conditions, permitting direct evaluation of how each paradigm influences key UX dimensions. The study adopts a multidimensional assessment approach, integrating subjective measures of usability, cognitive load, and immersion with objective indicators of task performance and interaction efficiency. This combination of controlled comparison with both validated subjective and behavioral metrics allows this study to provide a rigorous and comprehensive examination of the relative strengths and limitations of natural versus WIMP-based interaction design in AR.

Hypotheses

Based on prior literature and the identified gaps in direct comparative evaluation, this study proposes a set of hypotheses to examine how the interaction paradigm influences key UX dimensions in AR. Given the demonstrated alignment of natural interaction modalities with embodied and spatial engagement, participants are expected to report significantly higher usability and satisfaction with natural interactions than with WIMP-based interfaces (H1). Similarly, because natural interactions reduce abstraction and more closely reflect real-world behaviors, they are expected to result in lower perceived cognitive load (H2). Building on evidence that immersive and context-aware interactions enhance experiential quality, the natural interaction condition is also expected to produce higher levels of immersion and presence (H3). Finally, due to increased intuitiveness and more direct manipulation capabilities, participants are anticipated to achieve better task performance and navigation efficiency under the natural interaction condition (H4). Together, these hypotheses provide an organized framework for empirically evaluating the relative effectiveness of natural versus WIMP-based interaction paradigms in AR environments.

This study supports the growing body of AR research by providing a controlled, empirical comparison of natural and WIMP-based interaction paradigms across multiple dimensions of UX. By combining subjective assessments and objective performance measures, it offers a more inclusive evaluation framework than is typically found in previous studies. The remainder of this paper is organized as follows: Section 2 describes the methods of the experiment; Section 3 presents the findings; Section 4 discusses the findings; and Section 5 concludes the paper.

Study design

The present study applied a within-subjects experimental design to compare the effects of natural and WIMP-based interaction paradigms on UX in an AR environment. Each participant completed tasks under both interaction conditions, with condition order randomly assigned for each participant (i.e., each participant had an equal probability of receiving either the natural-first or WIMP-first sequence) to reduce potential order and learning effects. This design allows direct comparison of interaction paradigms while controlling for individual differences in prior experience, cognitive ability, and interaction style.

The natural interaction condition utilized embodied, multimodal inputs aligned with human spatial behavior, including gesture-based manipulation, gaze-based selection, and voice commands. These input mappings were selected because each preserves a direct, continuous correspondence between the user’s physical action and its digital effect (e.g., reaching toward an object to select it), operationalizing naturalness as defined above. These interactions were designed to be context-aware and adaptive, enabling users to directly engage with virtual objects in ways consistent with real-world actions (see Figure 1). In contrast, the WIMP-based condition implemented a traditional graphical user interface (GUI) adapted for AR, consisting of floating windows, icons, menus, and pointer-based interactions, but lacked more natural interactions. These interface elements were rendered spatially but retained a 2D symbolic interaction structure, thereby requiring indirect manipulation of virtual content (see Figure 2).

Participants

A total of 30 participants were recruited from university and community populations. Eligibility criteria included being at least 18 years of age, having normal or corrected-to-normal vision (including depth perception), and having no history of neurological, vestibular, or musculoskeletal conditions that could interfere with safe participation in AR-based tasks. Individuals with a history of seizures, vertigo, or implanted electronic devices were excluded.

Sample size was determined using a priori power analysis conducted with G*Power (Faul et al., 2007) for a paired-samples t- test. Assuming a moderate effect size (Cohen’s d = 0.5), an alpha level of 0.05, and a desired power of 0.80, a minimum of 30 participants was required. This sample size provides sufficient statistical sensitivity while remaining feasible.

Apparatus and materials

The study was conducted using the Meta Quest 3, a head-mounted AR device that supports 6-degree-of-freedom (DoF) tracking and multimodal interaction. Both interaction conditions were implemented within the same hardware environment to ensure consistency.

The experimental tasks were delivered using Design Interactive, LLC’s AUGMED® platform, a Tactical Combat Casualty Care (TCCC) training simulation focused on the identification and treatment of massive hemorrhage in a battlefield scenario. Tasks were identical across interaction conditions in terms of objectives, sequence, and content, assuring that any differences in observed outcomes could be attributed to interaction modality rather than task variation.

Measures

To capture a comprehensive assessment of UX and performance, the study used a combination of subjective self-report instruments and objective behavioral metrics (see Table 1). Standardized, domain-general instruments (SUS, NASA-TLX, PQ) were selected over emerging AR-specific evaluation approaches (e.g., Law et al., 2021; Sumithra et al., 2024) for two reasons. First, these instruments have established psychometric properties and normative benchmarks accumulated across decades of human-computer interaction (HCI) and virtual reality (VR) research, allowing the present findings to be situated within, and compared against, a broad existing evidence base. Second, AR-specific alternatives are still in early stages of validation and have not yet been used consistently enough across studies to support the kind of direct, controlled paradigm comparison this study aims to conduct; several remain heuristic or framework-based rather than psychometrically validated (e.g., Picardi and Caruso, 2024; Sumithra et al., 2024). Together, these measures provide a multidimensional evaluation of usability, cognitive load, immersion, and performance, permitting robust comparison across interaction conditions.

Procedure

Participants first completed informed consent in accordance with institutional review board (IRB) protocols, followed by a brief orientation to the Meta Quest device. This included practice interactions unrelated to the experimental tasks to ensure baseline familiarity with the system. Before completing the experimental tasks, participants completed a demographic questionnaire and the Immersive Tendencies Questionnaire (ITQ) to characterize the sample and assess baseline immersion propensity. Each participant then completed the AR scenario twice, once under the natural interaction condition and once under the WIMP-based condition, with condition order randomly determined for each participant to achieve a balanced counterbalancing across the sample. During task execution, the experimenter observed and noted errors, requests for clarification, and navigation efficiency.

Following each condition, participants completed the set of self-report measures assessing usability, workload, and immersion. A short break was provided between conditions to reduce fatigue. After completing both conditions, participants took part in a brief debriefing session to provide qualitative feedback on their experience and interaction preferences.

The results are presented in relation to the study hypotheses. Preliminary analyses first describe the participant sample and evaluate the assumptions underlying the planned statistical analyses, including assessment of normality and potential order effects. The remaining sections present the effects of the interaction paradigm on usability, cognitive workload, presence, task performance, and navigation efficiency.

Preliminary analyses

The final sample included 30 participants. Participants had a mean age of 32.07 years (SD = 12.21). The sample included 18 male and 12 female participants. Most participants were right-handed (n = 28), while two participants were left-handed. Participants represented a range of ethnic backgrounds, including White (n = 17), Asian (n = 5), Hispanic/Latino (n = 5), Middle Eastern (n = 1), multiple ethnicities (n = 1), and prefer not to say (n = 1). Educational backgrounds also varied, with participants reporting high school diplomas (n = 2), some college (n = 3), associate degrees (n = 5), bachelor’s degrees (n = 8), master’s degrees (n = 11), and doctoral degrees (n = 1). Participants reported a mean of 8.40 prior uses of AR technology (SD = 14.52), reflecting substantial variability in prior AR experience.

All analyses were conducted in Minitab (2026 web version). Paired-samples t -tests were selected a priori as the primary inferential test because they directly correspond to the within-subjects, two-condition comparison structure of the study design and were the basis of the power analysis. Before hypothesis testing, normality of the paired difference scores was evaluated using the Ryan–Joiner normality test, Minitab’s implementation of a correlation-based normality test that is functionally equivalent to, and highly correlated with, the Shapiro-Wilk test (Ryan and Joiner, 1976), together with visual inspection of probability plots. Difference scores for all dependent variables were approximately normally distributed, except for the SUS, which exhibited a departure from normality attributable to a single extreme observation. Given the within-subjects design, sample size (N = 30), and the established robustness of paired-samples t -tests to isolated departures from normality, parametric analyses were retained for the primary hypothesis tests; a nonparametric Friedman test was additionally conducted as a sensitivity analysis for SUS given its normality violation (see Section 3.2).

To evaluate the effectiveness of the counterbalancing procedure, difference scores (Natural − WIMP) were compared between participants who completed the natural interaction condition first (NW) and those who completed the WIMP condition first (WN). No significant order effects were observed for SUS, presence, errors, corrective actions, missed steps, or subjective interaction smoothness (all p > 0.05). Significant order effects were observed for NASA-TLX total workload (F (1, 28) = 6.88, p = 0.014), task completion time (F (1, 28) = 5.21, p = 0.030), clarification requests (F (1, 28) = 36.84, p < 0.001), and subjective efficiency ratings (F (1, 28) = 5.65, p = 0.025), suggesting modest learning or familiarity effects for task performance and workload-related measures despite the counterbalanced design. Descriptive statistics and inferential results for all dependent measures are presented in Table 2.

Hypothesis one: usability and satisfaction

To evaluate H1, perceived usability was compared between the Natural interaction and WIMP conditions using the SUS (see Table 2). Participants reported significantly higher usability for the natural interaction condition (M = 79.83, SD = 14.32) than for the WIMP condition (M = 72.67, SD = 17.69). A paired-samples t-test indicated that this difference was statistically significant, t (29) = 2.79, p = 0.009, d = 0.51. These outcomes support H1, indicating that the natural interaction paradigm provided a more usable UX than the WIMP-based interface.

Because the SUS difference scores exhibited a minor departure from normality attributable to a single extreme observation, a nonparametric Friedman test was conducted as a sensitivity analysis, and this likewise showed a significant effect of the interaction condition, χ 2 (1) = 9.97, p = 0.002, confirming that the observed usability advantage for the natural interaction condition was robust to violations of the normality assumption.

Hypothesis two: cognitive load

Although no significant difference in cognitive workload was observed between the natural interaction and WIMP conditions, a significant order effect was identified. To further examine this effect, an exploratory paired-samples t -test was conducted to compare NASA-TLX scores between participants’ first and second experimental sessions, regardless of interaction condition. Cognitive workload was significantly lower during the second session (M = 33.98, SD = 15.68) than during the first session (M = 39.68, SD = 14.30), t (29) = 2.66, p = 0.013, d = 0.49, with a mean reduction of 5.70 points (95% CI [1.31, 10.09]). This finding suggests that participants experienced a reduction in perceived workload after becoming familiar with the experimental task and AR environment.

Hypothesis three: presence and immersion

Consistent with H3, participants reported a significantly greater sense of presence when interacting with the natural interaction paradigm than when using the WIMP-based interface, t (29) = 3.76, p = 0.001, d = 0.69 (see Table 2). The observed effect was moderate to large, indicating that embodied, multimodal interactions enhanced participants’ sense of presence within the AR environment.

To examine whether individual differences in immersive tendency influenced these findings, Pearson correlations were computed between the ITQ and presence scores for each interaction condition. Immersive tendency was not significantly correlated with presence in either the natural condition (r = 0.09, p = 0.629) or the WIMP condition (r = - 0.21, p = 0.264), indicating that the observed differences in presence were not attributable to participants’ baseline immersion propensity. These findings support H3.

Hypothesis four: task performance and navigation efficiency

Results for the objective performance and navigation measures are summarized in Table 2. No significant differences were observed between interaction conditions for task completion time, requests for clarification, or observed efficiency (all p > 0.05). However, participants committed significantly fewer errors, t (29) = - 2.36, p = 0.025, missed fewer procedural steps, t (29) = - 3.73, p = 0.001, and made fewer corrective actions, t (29) = - 2.48, p = 0.019, when using the natural interaction paradigm. Participants also demonstrated significantly smoother interactions in the natural condition, t (29) = 2.11, p = 0.043. Collectively, these findings supply partial support for H4, indicating that while natural interactions did not improve overall task completion speed, they were associated with more accurate and smoother task execution than the WIMP-based interface.

General results

Across the full set of measures as shown in Figure 3, natural interaction produced small-to-medium effects (Cohen, 2013) on subjective usability (d = 0.51) and behavioral accuracy (errors, missed steps, and corrective actions; d s = 0.43–0.68), and a medium-to-large effect on presence (d = 0.69). By contrast, effects on cognitive workload, task completion time, clarification requests, and observed efficiency were small and non-significant (d s ≤ 0.32). This pattern indicates that the advantages of natural interaction were concentrated in perceived usability, presence, and task accuracy rather than speed or efficiency, a distinction examined further in the next section.

This study compared natural interaction and traditional WIMP-based interaction in an AR environment across multiple UX and performance dimensions. Overall, the results show that natural interaction provided measurable benefits for usability, presence, and interaction quality, while offering more limited advantages for cognitive workload and objective task efficiency. Together, these results suggest that the main benefits of natural interaction may lie in improving the subjective quality and naturalness of the interaction experience rather than universally improving task performance.

The first hypothesis, that natural interaction would produce higher usability than the WIMP interface, was supported. Participants reported significantly higher usability ratings for the natural interaction condition, with no evidence of order effects. This finding is consistent with prior AR and VR research demonstrating that interaction techniques aligned with real-world behaviors are considered more intuitive, easier to learn, and more satisfying than indirect graphical interfaces (Derboven et al., 2012; Jankowski and Hachet, 2015). By lessening the need to translate intended actions into menu selections or indirect pointer operations, natural interaction likely improved perceived ease of use and overall interaction quality (Rodrigues et al., 2011; Tastan et al., 2022).

The second hypothesis, that natural interaction would decrease cognitive workload, was not supported. Although mean NASA-TLX scores were lower for the natural interaction condition, the difference was not statistically significant. Instead, participants’ workload was significantly lower during their second experimental session, regardless of the interaction condition, denoting a practice or familiarization effect (Albuquerque et al., 2020; Nenna and Gamberini, 2023; Sun and Nakajima, 2023). Participants likely devoted greater mental effort during their initial exposure to understanding the task, AR environment, and interaction paradigm, resulting in a higher perceived workload that diminished with experience. Because no significant interaction between condition and order was observed, this effect appears to reflect acclimation rather than a basic advantage of either interface. Similar results have been reported in immersive technology research, where initial novelty and learning demands often contribute substantially to perceived workload during initial exposure (Anders et al., 2012; Roques et al., 2025). The absence of significant differences across individual NASA-TLX subscales further suggests that any workload advantages associated with natural interaction were relatively small compared with the influence of task familiarity.

The third hypothesis, that natural interaction would increase presence, was supported. Participants reported significantly greater presence when interacting through natural multimodal techniques, and this effect was independent of presentation order. This finding aligns with presence theory and prior literature on immersive technology, suggesting that interaction methods that align with users’ real-world expectations strengthen the perception of “being there” in virtual and augmented environments (Bekele and Champion, 2019; Grubert, 2021). Interestingly, immersive tendency (ITQ) was not significantly associated with presence in either interaction condition. Previous work has often reported modest positive relationships between immersive tendency and experienced presence; however, these relationships are frequently inconsistent and appear to depend on the characteristics of the system, task, and participant population (Azofeifa et al., 2022; Bekele and Champion, 2019). In the present study, the interaction technique itself may have exerted a stronger influence on presence than stable individual differences in immersive tendency.

The fourth hypothesis received partial support. Natural interaction led to fewer navigation errors, fewer missed procedural steps, fewer corrective actions, and smoother observed interactions, implying improved interaction quality and execution accuracy. However, no significant differences were observed for task completion time, clarification requests, or overall efficiency. This evidence suggests that natural interaction may improve users’ task performance without necessarily enabling them to complete those tasks more quickly. This distinction is important because natural interaction often encourages deliberate, embodied movements that highlight accuracy and confidence rather than speed (Büschel et al., 2018; Lee et al., 2012). Furthermore, both interfaces were designed to support successful task completion, which may limit opportunities to observe large differences in efficiency. Similar tradeoffs between interaction quality and objective speed have been reported in prior comparisons of natural and traditional interfaces, particularly when users are performing structured procedural tasks rather than exploratory activities (Dong et al., 2015; Shatilov et al., 2021).

Taken together, these results indicate that the primary value of natural interaction in AR goes beyond simple efficiency gains. Rather than dramatically reducing task time or cognitive demand, natural interaction appears to improve users’ subjective experience, increase immersion, and reduce execution errors while maintaining equivalent efficiency. These characteristics are particularly important in domains where accuracy, user confidence, and sustained engagement are valued alongside productivity, such as training, maintenance, healthcare, and decision support. Future research should examine whether these benefits become more pronounced with extended practice, more complex tasks, or collaborative AR environments, where the affordances of natural interaction may yield greater operational advantages.

Limitations

This study has several limitations. First, the findings were obtained using a single individual, high-stakes procedural AR training application, hardware platform, and task scenario, which may limit their generalizability to other AR systems and application domains. Second, although the sample size was sufficient to detect moderate within-subject effects, participants were recruited primarily from university and community populations rather than representative operational users. Finally, while the counterbalanced design reduced systematic sequencing bias, significant order effects were observed for several performance-related measures, suggesting that some learning occurred across repeated task execution. Additionally, this study relied on domain-general UX instruments rather than emerging AR-specific evaluation frameworks, which may capture spatial, embodied, and context-aware dimensions of AR interaction not represented in traditional usability, workload, and presence measures. As AR-specific instruments mature and gain broader validation, future work should incorporate them alongside established measures to more fully characterize AR-specific UX.

Future work

Future research should replicate these findings across diverse AR applications, user populations, and task contexts to further establish the generalizability of the observed effects. Additionally, the research should investigate why natural interaction improves usability, presence, and interaction quality without reducing perceived cognitive workload. Moreover, the work should examine how these interaction paradigms influence UX over the long term, during collaborative tasks, and in more complex operational scenarios. Finally, future studies should explore how individual interaction modalities (e.g., gesture, gaze, and voice) contribute independently to overall UX, thereby better informing the design of next-generation AR interfaces.

This study provides a controlled empirical comparison of natural and WIMP-based interaction paradigms in individual, high-stakes procedural AR training. Overall, the findings show that natural interaction produced higher perceived usability, greater presence, fewer interaction errors, fewer missed procedural steps, fewer corrective actions, and smoother observed interaction than the WIMP-based interface. These results support the value of embodied, multimodal interaction techniques for improving the quality and accuracy of AR UXs.

At the same time, natural interaction did not significantly decrease cognitive workload or improve task completion time, clarification requests, or overall efficiency. Instead, workload and several performance-related measures appeared to be influenced by order and familiarization, with participants reporting lower workload during their second session regardless of interaction condition. This suggests that the benefits of natural interaction may be most evident in perceived usability, presence, and interaction quality rather than immediate speed or workload reduction.

These findings have important implications for the design of future AR systems. Overall, this study provides empirical evidence that natural interaction offers meaningful advantages in UX over traditional WIMP-based interfaces in individual, high-stakes procedural AR training. While these benefits were reflected primarily in usability, presence, and interaction quality rather than cognitive workload or efficiency, they nevertheless highlight the value of designing interaction techniques that align with users’ natural behaviors. Rather than prioritizing speed alone, designers should consider how interaction techniques influence users’ perceptions of usability, immersion, and interaction quality. As AR continues to mature across training and operational domains, understanding these tradeoffs will be essential for developing interfaces that are not only effective but also intuitive, engaging, and supportive of human performance.