Investigating cognitive workload in virtual reality motor interactions: a neuroergonomic approach using EEG

Virtual reality (VR) environments rely heavily on motor interaction techniques that can shape users’ cognitive workload and overall experience. This study examined three interaction methods Near, Ray, and GoGo select to compare their impacts on cognitive workload using electroencephalographic (EEG) measures and NASA-TLX ratings. Twenty healthy participants completed VR motor tasks while their EEG was recorded, and each participant performed all three interaction methods. Spectral EEG analysis showed mixed results: beta power differed significantly across techniques, theta power did not, and alpha power showed an omnibus effect without significant corrected pairwise differences. In contrast, the EEG workload index based on the frontal theta/parietal alpha ratio (Index 1) showed a significant task effect, with Ray select exhibiting the highest workload. Ray select also received higher NASA-TLX ratings for mental, physical, and temporal demands, as well as effort and frustration. Repeated-measures correlations indicated that Index 3 was significantly related to several NASA-TLX subscales, whereas Index 1 and Index 2 were not, suggesting that the EEG indices capture different aspects of workload rather than a single unified construct. Near select consistently showed the lowest workload, while GoGo select produced intermediate scores. These findings suggest that ray-based object selection requires greater cognitive workload, likely because it combines visuomotor precision with multitasking demands. Overall, the results indicate that EEG composite indices may be more sensitive than raw band-power measures for distinguishing workload differences across VR interaction techniques, and they support the use of neuroergonomic measures for guiding VR interface design.

Virtual reality (VR) is a computer-generated platform that creates and manages virtual environments. In the past decade, immersive VR systems have moved from research labs to everyday use, driven by advancements in graphics hardware and improved motion tracking. VR integrates multiple stimuli, including auditory, visual, and tactile inputs, to provide a captivating experience. The stereoscopic displays employed in VR headsets facilitate an immersive perception of presence from a first-person perspective, thereby causing the illusion of a three-dimensional environment (Stark, 1995; Lawton, 2006). The accurate tracking of both hand and head movements significantly enhances the realism of the experience, effectively simulating physical interactions and actions (Wu et al., 2019). The lower cost of VR systems has made them more accessible, expanding their use across healthcare, education, design, and entertainment (Hamad and Jia, 2022).

Progress in this technology has accelerated the development of various motion interface devices, making motor interaction a key approach for immersive VR experiences. Researchers like Slater (Slater, 2009) and Jacob (Jacob et al., 2008) have reinforced this concept through their investigations into motor interaction in VR. Head tracking plays a crucial role in nearly all immersive VR systems; as users turn their heads, the trackers adjust the virtual viewpoint, thereby enhancing the sense of immersion. An important aspect of interaction involves the selection and manipulation of objects within a virtual environment. For example, Bowman and Hodges studied remote object manipulation using two methods: augmented arm extension, known as GoGo interaction, and ray casting. Their findings indicate that grasping and manipulation should be treated as separate interactions in order to enhance usability. They discovered that ray casting is more intuitive, while arm-extension techniques provide better control, highlighting the necessity to develop methods that excel in both dimensions (Bowman and Hodges, 1997).

These insights underscore the significance of prioritizing user comfort and engagement for effective object manipulation in VR by closely emulating real-world interactions. This necessitates that designers develop intuitive interfaces that enable seamless handling of virtual objects, ultimately enhancing the overall user experience (Liu et al., 2024a). To accomplish this, developers should emphasize the creation of natural interfaces that encourage fluid interaction with virtual elements, thereby further elevating the user experience. As VR technology continues to advance, the difficulty of physical and cognitive engagement from users is increasing, and research is increasingly focused on ergonomics and neuroergonomics to deepen the understanding of user interactions. Ergonomics and human factors research examines human capabilities to enhance safety and efficiency (“Progress and Direction in Neuroergonomics,” 2019), while neuroergonomics merges neuroscience with ergonomics to investigate cognitive behaviors within work environments (Parasuraman, 2003). This approach improves the design of human-machine systems through cognitive modeling. Neuroergonomics investigates how different jobs impact brain performance, aiming to create work environments that enhance productivity and support mental wellbeing. A notable aspect is using VR to explore cognitive functions and manage work demands. VR provides a safe, cost-effective way to study these factors in controlled settings, making it a valuable tool for advancing workplace design, despite challenges in simulation accuracy and standardization (Kearney et al., 2006; Moosavi et al., 2022).

In neuroergonomics, a key focus is evaluating how different tasks influence cognitive load, or the mental effort required for task performance. This can be measured through psychophysiological indicators, task performance data, and subjective tools like the National Aeronautics and Space Administration Task Load Index (NASA-TLX) (Hart and Staveland, 1988) and the subjective workload assessment technique (SWAT) (Reid and Nygren, 1988). In VR tasks, a range of physiological signals can be monitored, including body temperature, heart rate, respiratory rate, galvanic skin response, electroencephalographic (EEG) activity, electromyography, eyelid closure, eye movements, and brain metabolic activity. These signals can be tracked using sensors attached to participants immersed in the virtual environment (Ahmadi et al., 2023). Variations in tasks and individual capabilities can increase cognitive load, potentially leading to overload that hinders performance (Liu et al., 2024b). This underscores the critical importance of effectively managing workloads in VR, particularly in vital areas such as training and industrial operations (Pribadi et al., 2024; Carvalho et al., 2025).

Ensuring this balance is essential for user safety and overall performance. Research has extensively employed techniques such as Functional Near-Infrared Spectroscopy (fNIRS), EEG, Electrocorticography (ECoG), Magnetic Resonance Imaging (MRI), and Magnetoencephalography (MEG) to study how various cognitive factors influence neural control in both animal and human models (Latash et al., 2010; Koike et al., 2015). VR neuroimaging techniques offer valuable opportunities to enhance interaction, rehabilitation, and performance by providing neurobiological insights (Bohil et al., 2011; Ocklenburg and Peterburs, 2023; Sharma et al., 2024). Unlike MRI, MEG, fNIRS, and ECoG, which limit movement, EEG’s portability enables greater flexibility in movement (Lin et al., 2013; Song and Nordin, 2021; S. Krishna and Pulikkal, 2024). Improvements in EEG signal processing have increased spatial and temporal resolution (Janapati et al., 2023), making it well-suited for real-time cognitive workload assessment and enhancing VR applications across diverse fields.

EEG has become an essential tool for assessing cognitive workload in VR settings. Most studies analyze standard power spectral bands—theta, alpha, beta, and gamma—across frontal, central, and parietal regions. The fronto-parietal network shows decreased alpha power in the parietal area. The increased theta power in the frontal area as task difficulty increased, indicating higher cognitive load and greater engagement of working memory (Sauseng et al., 2005; Brouwer et al., 2012). Beta activity shows trends similar to those of alpha and may be associated with motor tasks (Pesonen et al., 2007; Scharinger et al., 2017). Frontal Theta/Parietal Alpha ratios and P300 amplitude are considered valuable performance markers (Holm et al., 2009; Puma et al., 2018). The indices Frontal Theta/Parietal Alpha ratio (Holm et al., 2009) and Frontal Midline Theta rhythm (Yamada, 1998) are effective indicators of cognitive load across different tasks. Most EEG research mainly examines alpha, theta, and beta band power (Borghini et al., 2014). Beta activity relates to working memory and thought regulation (Lundqvist et al., 2016), whereas theta activity reflects attentional and cognitive effort. During working memory tasks, theta power usually rises, and alpha power decreases compared to relaxed states (Klimesch, 1999). Distinct brain regions exhibit varying levels of effectiveness in different EEG frequency bands, with increased frontal midline theta power correlating positively with heightened memory load (Aricò et al., 2016). This relationship indicates that increasing cognitive demands adjust brain activity to reflect working memory and attention, especially in VR. These findings highlight the dynamic neural responses to cognitive tasks and their implications for attention and memory in immersive settings (Zakrzewska and Brzezicka, 2014).

The NASA Research Center has effectively employed the EEG engagement index to assess sensitivity to cognitive workload. This index, based on energies in the alpha, beta, and theta bands, shows that increased engagement correlates with higher cognitive load (Pope et al., 1995; Cheng et al., 2022). In literature, workload generally increased with difficulty. Still, engagement varied, sometimes decreasing after initially high levels on multi-level tasks, and was not consistently affected by difficulty in other complex tasks (Berka et al., 2007).

In VR, “Interaction Patterns and Techniques” are key for user communication and designing intuitive experiences. Patterns are reusable concepts across VR apps, helping users achieve common goals by describing interactions among users, the environment, and objects. Techniques are technology-specific methods for implementing patterns that illustrate user interaction with VR via input/output devices. For instance, the Hand selection pattern uses the GoGo technique to extend reach beyond arm length by mapping the initial reach and non-linearly enlarging the virtual hand, enabling precise selection of nearby and distant objects. The Pointing Pattern employs a ray extension technique for object selection, often favored over hand selection for distant or small objects when realistic interaction is not necessary (Jerald, 2015). Targeting small or distant objects are hard to target, which increases cognitive workload and difficulty (Argelaguet and Andujar, 2013; Mendes et al., 2017). Despite their popularity, the cognitive workload of motor tasks such as Near, Ray, and GoGo selection has not been discussed. These interaction techniques are among the most common ways users engage with VR. Therefore, this study examines the impact of VR motor interaction tasks on cognitive workload through EEG metrics—focusing on theta, alpha, and beta power and indices—and NASA-TLX scores. It further detects task-specific cortical patterns to clarify the neural demands associated with each technique. Key research questions include: R1: How do different VR motor interaction methods—Near select, Ray select, and GoGo select—impact cognitive workload as measured by EEG?

R2: How do subjective workload scores (NASA-TLX) relate to neurophysiological indicators of cognitive effort during these VR activities?

R3: Can specific EEG indices reliably differentiate cognitive demands across distinct VR motor interaction methods?

The analysis is based on the following hypotheses: H1: The Ray select method causes a higher cognitive workload than other selection techniques, evidenced by increased frontal theta and decreased parietal alpha in the EEG.

H2: The Near select method will result in lower NASA-TLX workload scores compared to other selection methods.

H3: EEG index that combines theta and alpha power (such as the frontal theta/parietal alpha ratio) will effectively differentiate workload levels across VR tasks.

H4: EEG workload indices are expected to show associations with NASA-TLX subscales, indicating possible convergence between physiological and subjective workload measures.

The following parts of this manuscript are set up like this: Section 2 provides comprehensive details regarding the experimental design and methodology employed for EEG data and subjective measures analysis. Section 3 presents the findings from the EEG signal investigation, along with the statistical analyses conducted. Section 4 presents the discussion, inherent research limitations, and prospective avenues for further investigation. Section 5 encapsulates the conclusions drawn from the investigational findings.

The methodology will describe how participants were selected, outline the specific VR tasks used, and explain the techniques for EEG data collection and subjective measures analysis applied during the study.

Participants

This study investigates neural markers of cognitive workload during VR motor interaction tasks in twenty healthy participants aged 20 to 40 (average age = 27.1 ± 4.47 years), including fifteen males and five females. A priori power analysis was performed utilizing G*Power 3.1 software for a one-factor repeated-measures ANOVA (three interaction techniques; within-subjects). The results of the study suggested that a total sample size of 28 participants would be required. Data collection was limited to 20 participants due to resource constraints, which may affect sensitivity to small effects; results are interpreted with this limitation acknowledged. All participants had no history of neurological issues. Participants provided informed consent before participating. The study received approval from the Ethics Committee of the University Pendidikan Sultan Idris (No. UPSI/PPPI/PYK/ETIKA(M)/Jld. 19 (90)). Participants were compensated with 50 Malaysian ringgit for their participation.

Experimental setup

The research design is carefully structured to assess cognitive states during VR motor interactions. It includes several phases, from recruiting participants to analyzing data. (i) Participant Briefing: Participants receive an initial orientation and an informational brochure explaining the research aims, benefits of participation, and eligibility criteria. They must also complete a consent form. (ii) Equipment Familiarization: Participants are given time to become familiar with the VR headset, set up the EEG cap for comfort and usability, and complete baseline measurements of eye closure (EC) for 2 min and eye opening (EO) for 2 min (iii) EEG Recording Session: Participants engage in a 9-min session during which EEG data are recorded while performing specific tasks in a VR environment (see Figure 1). (iv) Task Feedback: After completing the tasks, participants provide feedback on their experience, focusing on cognitive workload and the difficulty level, assessed via the NASA-TLX digital self-assessment form.

Participants engage in sessions focused on motor skills and cognitive functions (Bowman and Hodges, 1997). They perform tasks in a sequence, with timed breaks to prevent fatigue. Figure 2 illustrates different task types: (a) Near select, where participants use a pinch gesture to choose a virtual cube (volume = 10 3 cm 3) positioned nearby at 30 cm distance, with visual cues—the cube appears red when out of range, blue when within range, and green when selected; (d) Ray select, where they project a ray from their hand to pick a virtual cube located at 150 cm, also accompanied by visual indicators; (e) GoGo select, where participants reach for a distant virtual cube, with non-linear hand movement once a threshold is crossed, and select using a pinch gesture with visual feedback. The session includes 12 trials with randomized tasks to reduce learning effects, lasting approximately 9 min and featuring rest periods. The entire session lasts about 70 min, including the experimental procedure briefing, EEG cap setup, baseline assessments, a demonstration of the virtual environment, virtual task execution, and NASA-TLX task feedback.

Data collection

The procedures were carried out at the CISIR Laboratory, part of the Electrical and Electronics Engineering Department at Universiti Teknologi PETRONAS, Malaysia. EEG signals from 64 channels were continuously recorded at a 512 Hz sampling rate using the high-resolution AntNeuro eegosportsTM EE-201 amplifier (eemagine Medical Imaging Solutions GmbH, Gubener Str.). Electrodes were attached to participants’ scalps according to the 10–20 international system, with a ground (GND) placed on the forehead and a reference at CP Z (Mohd Noor et al., 2023). Electrode impedances were maintained below 10 KΩ. The Meta Quest Pro VR headset features a resolution of 1800 x 1920 pixels per eye, a 106-degree field of view, and a 90 Hz refresh rate. It also provides Six Degrees of Freedom hand-tracking, enabling participants to perform tasks within a virtual environment.

EEG data preprocessing

The EEG data were first pre-processed using the open-source EEGLAB (v2024.2.1) (Delorme and Makeig, 2004). The sampling rate was then reduced to 256 Hz. Next, a bandpass filter between 1 and 40 Hz was applied with a zero-phase finite impulse response filter. The data was re-referenced using the common average method. Additionally, independent component analysis was performed to detect and remove artifacts caused by eye blinks and movements. The continuous EEG data were segmented into trials, with those exceeding ±300 μV in amplitude being systematically discarded. All processing steps were completed using MATLAB R2024b (Shardt, 2023).

EEG cognitive load indices

The beta, alpha, and theta frequency powers (Borghini et al., 2014), as well as the (frontal theta/parietal alpha) (Holm et al., 2009), frontal midline theta (Yamada, 1998), and (beta/(alpha + theta)) (Pope et al., 1995). Indices were employed to assess the subjects’ cognitive status using EEG. The numerical indices corresponding to each segment of the VR engagement tasks were derived by utilizing the beta, theta, and alpha band power measurements individually (Equations 1–3). In this situation, Nc denotes the grouping of all EEG electrodes used. At the same time, Fm, Fr, and Par pertain to the electrodes allocated in the frontal midline, frontal, and parietal territories, correspondingly (Table 1).

Index 1 is calculated as the ratio of the mean theta frequency power recorded in the frontal zone to the mean alpha frequency power from the parietal zone (Equation 1) (Holm et al., 2009). Index 2 represents the average theta frequency power obtained from the frontal midline zone (Equation 2) (Yamada, 1998). Index 3 is defined by the ratio of average beta frequency power to the sum of average alpha and average theta frequency powers across all electrodes (Nc) (Equation 3) (Pope et al., 1995).

NASA-TLX data collection

NASA-TLX assessments were conducted post-task as participants completed VR motor interaction tasks. Participants utilized a digital interface to complete the NASA-TLX questionnaire after removing the VR headset.

The NASA-TLX assesses workload using six subscales: Mental Demand, Physical Demand, Temporal Demand, Performance, Effort, and Frustration. It employs pairwise comparisons and visual analog scale ratings (ranging from 1 to 21, indicating very low to very high load). Participants rated their perceived workload for each interaction technique, producing raw scores for each subscale and overall workload analysis.

This method effectively captured participants’ retrospective perceptions of workload, reducing memory decay and supporting real-time EEG data. Responses were collected in a quiet laboratory setting, with researchers available to clarify questions and avoid biasing ratings. All 20 participants completed assessments across three conditions, yielding 60 NASA-TLX profiles for analysis. Quantitative analysis, including measures such as frequency powers, index metrics, and NASA-TLX data, was conducted using a custom MATLAB algorithm. Statistical tests and analysis were performed using the Numiqo Team (2025) statistical tool (Numiqo Team (2025), no date).

The results provide detailed insights into how different VR motor interaction tasks Near select, Ray select, and GoGo select affect cognitive workload. This is measured using EEG spectral metrics (theta, alpha, beta powers, and cognitive indices) and subjective NASA-TLX reports. EEG analyses monitor physiological responses in real time during task performance, while NASA-TLX captures users’ perceptions afterward. Together, these methods verify workload effects and help in developing VR systems that balance immersion with cognitive load.

EEG analysis

Descriptive inspection of theta, alpha, and beta power, along with the cognitive workload indices, showed modest visual differences across the VR motor tasks. These results will help understand how motor interactions relate to mental effort and guide the development of VR systems that better meet user needs.

Figure 3 displays a topographic map of average frequency-band power across various tasks. During the selection tasks, theta power is predominantly increased over frontal electrodes, indicating greater activation of frontal cortex areas. This increase correlates with higher cognitive workload and executive function demands across the three interaction techniques. Conversely, alpha power decreases in the parietal and occipital regions, especially under Near and GoGo conditions, reflecting alpha desynchronization associated with attentional effort. The reduction in parietal alpha suggests the recruitment of attentional resources, implying that increased task difficulty lowers alpha power. Beta power remained relatively low and consistent across all interaction techniques, reflecting minimal variation in this frequency band across tasks.

Figure 4 shows the graphical representations of power distribution across different frequency bands. The analysis revealed task-specific changes in neural activity within these bands. The Ray select condition showed modest variations in median values, with somewhat higher theta power and slightly lower alpha and beta power.

Both Index 1 (median = 0.43 ± 0.06) and Index 2 (median = 0.33 ± 0.09) reached their highest values during the “Ray select” task, whereas Index 3 (median = 0.35 ± 0.05) was at its lowest in that condition, as shown in Figure 5.

EEG statistical analysis

Before conducting statistical analysis, the Anderson-Darling Normality Test was used to determine appropriate statistical methods. Most datasets were normally distributed (p > 0.05), except for Alpha power, which was not. Therefore, repeated-measures ANOVA was applied to parametric data, and the Friedman test was used for Alpha power, both at a 95% confidence level. A repeated-measures ANOVA was conducted to compare Index 1 across the three task conditions (Near select, Ray select, and GoGo select). Mauchly’s test indicated that the assumption of sphericity was not violated, χ 2 (2) = 1.59, p = 0.45. The analysis revealed a significant effect of task on Index 1, F (2, 38) = 7.09, p = 0.002, η 2 = 0.27. The mean Index 1 was highest for Ray select (M = 0.42, SD = 0.06), followed by Near select (M = 0.38, SD = 0.06) and GoGo select (M = 0.36, SD = 0.07). Bonferroni-adjusted post hoc tests showed that Ray select was significantly higher than GoGo select, with a mean difference of 0.06 and a 95% CI [0.01, 0.11], p = 0.004. The remaining pairwise comparisons were not significant. Index 2 did not differ significantly across the three task conditions, F (2,38) = 0.29, p = 0.75, η 2 = 0.02. The mean values were very similar across Near select (M = 0.39, SD = 0.09), Ray select (M = 0.40, SD = 0.08), and GoGo select (M = 0.40, SD = 0.09). Bonferroni-adjusted post hoc comparisons revealed no significant pairwise differences. Similarly, Index 3 showed no significant task effect, F (2,38) = 1.71, p = 0.195, η 2 = 0.08. The mean Index 3 was 0.38 (SD = 0.04) for Near select, 0.36 (SD = 0.05) for Ray select, and 0.37 (SD = 0.05) for GoGo select. Bonferroni-adjusted post hoc tests indicated no significant differences between any pair of conditions.

For Theta power, Mauchly’s test indicated that the assumption of sphericity was not violated, χ 2 (2) = 4.65, p = 0.098. There was no significant effect of condition, F (2, 38) = 0.75, p = 0.477, η 2 = 0.038. The mean scores were M = 0.38, SD = 0.08 for Near select; M = 0.38, SD = 0.07 for Ray select; and M = 0.39, SD = 0.08 for GoGo select. Bonferroni-adjusted post hoc tests revealed no significant pairwise differences between any conditions. For Beta power, Mauchly’s test indicated that the assumption of sphericity was not violated, χ 2 (2) = 0.60, p = 0.74. There was a significant main effect of condition, F (2, 38) = 10.80, p < 0.001, η 2 = 0.36. Descriptive statistics for the conditions were as follows: Near select (M = 0.38, SD = 0.07), Ray select (M = 0.34, SD = 0.07), and GoGo select (M = 0.38, SD = 0.06). Bonferroni-adjusted post hoc tests revealed significant differences between Near select and Ray select (mean difference = 0.04, p = 0.007) and between Ray select and GoGo select (mean difference = −0.04, p = 0.001). The difference between Near select and GoGo select was not significant. Alpha power was analyzed using a Friedman test. The test indicated a statistically significant difference among the conditions, χ 2 (2) = 6.3, p = 0.043, with a small effect size (Kendall’s W = 0.158). The median values for Near select, Ray select, and GoGo select were 0.168, 0.169, and 0.202, respectively. The corresponding mean ranks were 2.15, 1.55, and 2.30. Post hoc pairwise comparisons using the Dunn-Bonferroni method revealed no significant differences after adjustment for multiple comparisons. Specifically, the comparison between Near select and Ray select yielded an adjusted p = 0.173, Near select and GoGo select yielded an adjusted p = 1.000, and Ray select and GoGo select yielded an adjusted p = 0.053.

Correlation analysis of EEG indices

Repeated-measures correlation analysis was used to examine associations among the EEG workload indices while accounting for the within-participant repeated-measures design. As shown in Figure 6, there was no significant association between Index 1 and Index 2, r = 0.10, df = 39, p = 0.532, 95% CI [−0.21, 0.40]. The correlation between Index 1 and Index 3 was also not significant, r = 0.14, df = 39, p = 0.386, 95% CI [−0.18, 0.43]. However, Index 2 and Index 3 were significantly and moderately negatively correlated, r = −0.49, df = 39, p = 0.001, 95% CI [−0.70, −0.22].

NASA-TLX subjective workload assessment

The NASA-TLX assessment offered subjective workload measurements across six areas: Mental Demand, Physical Demand, Temporal Demand, Performance, Effort, and Frustration. Ratings were obtained for the three interaction methods to supplement the EEG-based workload indicators.

The Ray select technique consistently resulted in the highest workload ratings across most dimensions (see Figure 7). Scores for Mental (median = 13.5 ± 5.46), Physical (median = 11.5 ± 6.43), Temporal (median = 13.5 ± 6.2), Effort (median = 12 ± 4.12), and Frustration (median = 13.5 ± 5.82) were all higher in this condition, indicating that participants found Ray select to be cognitively demanding, time-pressured, and emotionally exhausting compared to other methods. Conversely, the perceived Performance (median = 13 ± 6.09) for the Ray select was generally lower, implying participants felt less confident about successfully completing tasks with this interaction.

The NASA-TLX results align with the EEG findings, particularly Index 1, suggesting that Ray select requires greater cognitive effort than the other interaction methods. The higher subjective workload reported for Ray select aligns with the significant beta effect and the frontoparietal workload pattern observed in Index 1, providing both perceptual and physiological evidence of the greater difficulty associated with this technique.

NASA-TLX statistical analysis

Before conducting statistical analysis, the Anderson-Darling Normality Test was used to determine appropriate methods. NASA-TLX ratings showed clear task-related differences across all subscales. Mental Demand, with Mauchly’s test, indicated that the assumption of sphericity was not violated, χ 2 (2) = 0.26, p = 0.88. There was a significant main effect of condition, F (2, 38) = 29.12, p < 0.001, η 2 = 0.61. The mean scores were M = 3.00 (SD = 2.29) for Near select, M = 12.40 (SD = 5.46) for Ray select, and M = 8.85 (SD = 5.62) for GoGo select. Bonferroni-adjusted post hoc tests revealed that all pairwise differences were significant. Specifically, Near select was significantly lower than Ray select (mean difference = −9.40, p < 0.001) and GoGo select (mean difference = −5.85, p < 0.001). Ray select was significantly higher than GoGo select (mean difference = 3.55, p = 0.031). Physical Demand, Mauchly’s test indicated that the assumption of sphericity had been violated, χ 2 (2) = 6.97, p = 0.031. A Friedman test was conducted. The results indicated a statistically significant difference among the conditions, χ 2 (2) = 15.925, p = 0.0003, with a moderate effect size, Kendall’s W = 0.3981. The mean values for Near select, Ray select, and GoGo select were 1.275, 2.425, and 2.3. Post hoc pairwise comparisons using the Dunn-Bonferroni method revealed that Near select differed significantly from Ray select (mean-rank difference = −1.15, adjusted p = 0.001) and from GoGo select (mean-rank difference = −1.025, adjusted p = 0.004). No significant difference was found between Ray select and GoGo select (mean-rank difference = 0.125, adjusted p = 1). Temporal Demand, Mauchly’s test indicated that the assumption of sphericity was not violated, χ 2 (2) = 0.58, p = 0.75. There was a significant main effect of condition, F (2, 38) = 23.95, p < 0.001, η 2 = 0.56. Descriptive statistics for the conditions were as follows: Near select (M = 2.75, SD = 2.00), Ray select (M = 12.10, SD = 6.20), and GoGo select (M = 9.25, SD = 5.43). Bonferroni-adjusted post hoc tests revealed that the mean score for Near select was significantly lower than both Ray select (mean difference = −9.35, p < 0.001, 95% CI [-12.45, −6.25]) and GoGo select (mean difference = −6.50, p < 0.001, 95% CI [-9.15, −3.85]). The difference between Ray select and GoGo select was not significant (mean difference = 2.85, p = 0.17, 95% CI [-0.07, 5.77]). Performance: Mauchly’s test indicated that the assumption of sphericity was not violated, χ 2 (2) = 0.55, p = 0.76. There was a significant main effect of condition, F (2, 38) = 15.69, p < 0.001, η 2 = 0.45. Descriptive statistics for each condition are as follows: Near select (M = 19.95, SD = 1.28), Ray select (M = 13.00, SD = 6.09), and GoGo select (M = 15.35, SD = 5.38). Bonferroni-corrected post hoc tests revealed that Near select scores were significantly higher than Ray select scores (mean difference = 6.95, p < 0.001) and GoGo select scores (mean difference = 4.60, p = 0.003). The difference between Ray select and GoGo select was not significant (mean difference = −2.35, p = 0.24). Effort, Mauchly’s test indicated that the assumption of sphericity was not violated, χ 2 (2) = 1.18, p = 0.55. There was a significant main effect of condition, F (2, 38) = 25.12, p < 0.001, η 2 = 0.57. The mean scores were M = 3.80 (SD = 3.16) for Near select, M = 12.30 (SD = 4.12) for Ray select, and M = 10.10 (SD = 5.64) for GoGo select. Bonferroni-adjusted post hoc tests revealed that the Near select condition was significantly lower than both the Ray select condition (mean difference = −8.50, p < 0.001, 95% CI [-10.94, −6.06]) and the GoGo select condition (mean difference = −6.30, p < 0.001, 95% CI [-8.73, −3.87]). There was no significant difference between the Ray select and GoGo select conditions (mean difference = 2.20, p = 0.39, 95% CI [-0.72, 5.12]). Finally, Frustration Mauchly’s test indicated that the assumption of sphericity was not violated, χ 2 (2) = 0.47, p = 0.79. There was a significant main effect of condition, F (2, 38) = 14.81, p < 0.001, η 2 = 0.44. Descriptive statistics for the conditions were as follows: Near select (M = 4.25, SD = 6.21), Ray select (M = 11.60, SD = 5.82), and GoGo select (M = 9.25, SD = 6.37). Bonferroni-adjusted post hoc tests revealed that the mean score for Near select was significantly lower than Ray select (mean difference = −7.35, p < 0.001) and GoGo select (mean difference = −5.00, p = 0.004). The difference between Ray select and GoGo select was not significant (mean difference = 2.35, p = 0.261).

Repeated-measures correlation analysis was also used to examine the relationships between the EEG workload indices and NASA-TLX subscales, accounting for the repeated-measures structure of the data. As shown in Table 2, Index 1 was not significantly associated with Mental Demand, Physical Demand, Temporal Demand, Performance, Effort, or Frustration, all p s > 0.05. Index 2 also showed no significant relationships with the NASA-TLX subscales. In contrast, Index 3 was significantly and negatively correlated with Mental Demand, r = −0.39, df = 39, p = 0.013, Physical Demand, r = −0.34, df = 39, p = 0.029, Temporal Demand, r = −0.38, df = 39, p = 0.014, Effort, r = −0.36, df = 39, p = 0.020, and Frustration, r = −0.37, df = 39, p = 0.018, and it was positively correlated with Performance, r = 0.46, df = 39, p = 0.003. Overall, these cross-measure correlations did not directly support H4, although the condition-level ordering of Ray > GoGo > Near was broadly consistent across physiological and subjective measures.

This study used a neuroergonomics approach to investigate cognitive states during VR motor tasks, providing detailed insights into the neural processes involved. It enabled a systematic analysis of EEG signals across different interaction types, facilitating assessment of cognitive workload and subjective responses. The results emphasize a complex connection between VR motor tasks and mental load, highlighting the significance of task design in improving user experience and performance in immersive environments.

Cognitive correlates for VR motor interactions

Ray select showed the highest subjective workload ratings, and descriptive EEG trends also suggested slightly greater cognitive demand, although these EEG differences were not statistically significant. Near select showed the lowest workload patterns, with GoGo select positioned between the two, offering a partial response to research question 1. Ray select displayed a descriptive pattern of higher theta and lower alpha and beta power, together with elevated Index 1 and Index 2 and lower Index 3 values. However, because only Index 1 showed a significant task effect, these patterns should be viewed as descriptive trends rather than strong indicators of interaction-driven differences in cognitive experience. Although the statistical results for band powers, Index 2, and Index 3 were not significant, the significant main effect for Index 1 and the post hoc Ray-GoGo difference demonstrate the sensitivity of Index 1. EEG topographies showed increased frontal theta power and reduced parietal alpha power during selection tasks, patterns commonly associated with increased cognitive control and attentional focus (Sauseng et al., 2005; Brouwer et al., 2012). Ray select showed slightly higher theta values and somewhat lower alpha and beta values in the descriptive plots; however, these trends were not statistically significant and should be interpreted as visual patterns rather than indicators of sensorimotor changes (Berka et al., 2007).

H1 was only partially supported: the predicted spectral pattern was not detectable in raw theta and alpha power alone, but it emerged most clearly in Index 1. This suggests that the workload effect was better captured by the composite frontoparietal ratio than by individual band powers. The raw theta and alpha results did not provide significant pairwise confirmation. The absence of significant raw band-power effects does not contradict the hypothesis; rather, it suggests that the underlying spectral shifts were modest and that the composite ratio captured a more stable frontoparietal workload signature than either band alone. Additionally, the results support H3, because Index 1 significantly differentiated Ray from GoGo select. Frontal theta and posterior alpha showed similar descriptive patterns across tasks, and since these effects were not significant, Ray select cannot be identified as uniquely producing higher cognitive load. (Yamada, 1998; Holm et al., 2009). Conversely, beta activity was generally low and spread across tasks, with particularly weak beta levels in Ray select, implying decreased sensorimotor involvement and a shift toward more effortful, cognitive control and multitasking rather than smooth motor function (Berka et al., 2007; Lundqvist et al., 2016).

Interestingly, conventional power metrics and composite indexes exhibited limited sensitivity to task-level variation. Most frequency-band indices did not show significant pairwise differences, except for Index 1, while Beta showed a significant condition effect and Alpha showed an omnibus difference without corrected pairwise differences. This distinguished Ray select from GoGo select. This suggests that relying solely on power-based global markers may not capture subtle differences in VR interaction requirements. More detailed spatial or temporal metrics, or personalized calibration, may be needed to identify workload changes specific to each condition. Furthermore, the significant negative correlations between Index 2 and Index 3, indicate that these indices reflect different aspects of neural workload; higher cognitive load appears to be linked to lower engagement or efficiency, as shown in Index 3 (Pope et al., 1995; Berka et al., 2007).

EEG results match previous research showing that cognitive workload is indicated by spectral changes in the theta, alpha, and beta bands, including increases in frontal theta and reduction of parietal alpha (Tremmel et al., 2019; Alessa et al., 2023). Composite measures such as the frontal theta/parietal alpha ratio (Mondellini et al., 2024) and the beta/(alpha + theta) engagement index (Pope et al., 1995; Berka et al., 2007) has also been used to evaluate workload and attention, with similar patterns observed in interactive VR and AR studies (Raufi and Longo, 2022; Karmakar et al., 2024). These results support the idea that the current effects are due to task-driven changes in cognitive demand influenced by interaction design.

NASA-TLX corroboration of EEG findings

The NASA-TLX analysis shows that Ray select imposes a much higher subjective workload than the Near and GoGo selects. This directly addresses research question 2, which asks how NASA-TLX scores relate to neurophysiological markers, and question 3, which asks whether EEG indices can reliably distinguish different cognitive demands across methods. Participants reported higher mental, physical, temporal, effort, and frustration demand, along with lower perceived performance, for Ray select. Statistical tests confirmed significant differences across all subscales, with Near select having the lowest scores. This pattern corresponds partly with the EEG results, as only Index 1 showed a significant task-related effect, specifically distinguishing Ray from GoGo. Repeated-measures correlations between the EEG workload indices and NASA-TLX subscales further indicated that Index 1 and Index 2 were not significantly associated with the subjective measures, whereas Index 3 showed significant negative correlations with Mental Demand, Physical Demand, Temporal Demand, Effort, and Frustration, and a positive correlation with Performance. Together, these findings suggest that the different EEG indices capture distinct aspects of workload rather than a single unified construct. Further non-parametric tests showed Near select significantly differed from Ray and GoGo on all subscales, while Ray and GoGo did not differ. This supports a low-demand (Near) versus higher-demand (Ray/GoGo select) profile, with Ray generally showing the highest subjective workload. These results provide partial convergent evidence for H4, but the convergence is stronger at the condition level than in the direct EEG–NASA-TLX correlations. As shown by the broadly consistent Ray > GoGo > Near select ordering at the condition level, the pattern suggests agreement between subjective and physiological measures in overall workload ranking, even though the correlation coefficients indicate that only Index 3 is consistently related to the NASA-TLX subscales. This convergence highlights the value of combining physiological and subjective measures to assess VR cognitive load and informs the design of interaction techniques to reduce user burden.

These findings have significant implications for VR motor interaction design. Techniques like Near select appear to reduce cognitive and emotional strain by utilizing natural, embodied movements and lowering uncertainty in target selection. Conversely, ray-based methods, which are common in VR, are perceived as more taxing, as indicated by higher NASA-TLX scores for Ray select. Human mental resources are limited; when a task exceeds these limits, performance declines and errors increase. According to Multiple Resource Theory, humans have separate “buckets” for visual, auditory, spatial, and verbal processing. Dual tasks are handled well if they draw on different resources, but face challenges when resources overlap. Ray Select likely produced the highest workload because it required simultaneous visual aiming, spatial judgment, and motor confirmation within a single tightly coupled control loop, whereas GoGo distributed these operations across a more sequential reach–pinch sequence and Near Select minimized the spatial extent of the interaction. From a closed-loop control perspective, Ray Select likely increased the gain and temporal precision required for working correction, making the sensorimotor loop less stable and more costly to maintain, whereas GoGo reduced the need for continuous correction by spreading the action over two phases. Although this study does not directly measure precision or multitasking, subjective ratings suggest that some aspects of ray-based interaction increase perceived workload (Berka et al., 2007; Wickens, 2008; Midha et al., 2021).

Limitations and future directions

Several limitations should be considered when interpreting these findings. First, the sample size was modest and below the original a priori target, which limits statistical power, particularly for detecting small-to-medium effects and for estimating correlations with precision. As a result, some null findings may reflect limited sensitivity rather than the absence of an effect. Second, the analyses were conducted primarily at the condition level, so intra-individual trial-to-trial variability was not modeled explicitly. Future trial-level mixed-effects analyses would be valuable for determining whether the observed workload effects remain stable across repetitions. Third, the EEG analysis was limited to linear spectral power measures and a frontoparietal index; therefore, spatial statistics, connectivity measures, time-frequency decomposition, and nonlinear complexity metrics were not examined. Fourth, the study used a single VR headset and a single interaction implementation set, which limits generalizability to other hardware platforms and input mappings. Finally, the present analysis did not systematically examine participant-level moderators such as prior VR experience, spatial ability, or interaction preference, although these factors may influence both subjective and physiological workload. Future studies with larger and more diverse samples should address these sources of variability to better characterize how VR interaction design affects workload.

This neuroergonomics study reveals that VR motor interaction techniques impact cognitive workload differently. Ray select demonstrated the highest workload, with Index 1 providing the clearest EEG discrimination. These differences depend on the task, influenced by selection distance (30 cm Near, 150 cm Ray/GoGo) and visuomotor demands. Spectral power results were mixed: Theta was not significant, Beta showed a significant condition effect, and Alpha showed an omnibus effect without significant corrected pairwise differences. EEG Index 1 and NASA-TLX suggest Ray’s increased cognitive demands due to visuomotor multitasking and precision tasks. The hypotheses are partly confirmed: H2 and H3 are supported more clearly, H1 is only partially supported, and H4 is supported only partially, because the EEG–NASA-TLX correlations were not uniformly significant, although the condition-level ordering remained broadly consistent across physiological and subjective measures. This emphasizes key factors for developing cognitively sustainable VR interfaces for training, rehabilitation, and critical applications. However, EEG power and other indices (besides Index 1) showed limited sensitivity at the group level, highlighting the need for more localized measures, personalized calibration, and larger samples to better assess workload. Future research should include varied participants, analyze physiological sensor data and NASA-TLX at the trial level, and develop real-time neuroadaptive systems that adjust interaction modes based on frontal theta/alpha ratios. These improvements will help create immersive technologies that enhance user performance and prevent cognitive overload.