One size fits none: a multi-method evaluation of VR interaction techniques across autistic and neurotypical users

Selecting suitable interaction techniques is critical for the effectiveness of autism-related virtual reality (VR), yet most evaluations still rely heavily on subjective feedback. This study applied a multi-method framework to compare two locomotion techniques (Teleport vs. Step) and two object interaction techniques (Hand-tracking vs. Controller) across autistic (N = 16) and neurotypical (N = 18) participants. Each technique was evaluated through three complementary lenses: trial duration as a behavioral efficiency indicator, EEG alpha and theta power spectral density as physiological proxies for attentional engagement and working-memory demand, and a direct perceived-ease prompt as a subjective indicator. Adaptation was quantified using trial-wise slopes across repeated exposure. Across the present sample, participants generally became faster over time in all four techniques. However, these behavioral gains did not consistently converge with EEG proxies or perceived-ease. The strongest group-sensitive finding emerged in locomotion: autistic participants showed sustained alpha suppression during Step, while neurotypical participants showed an alpha rebound. In object interaction, a subgroup of autistic participants who perceived Hand-tracking as easier showed alignment across all three measures, pointing to the potential value of low-abstraction and direct-action interfaces. These findings suggest that VR interaction techniques in autism-related contexts are best understood as multidimensional profiles rather than as universally optimal solutions. However, given the small and heterogeneous autistic field sample, substantial group differences in participant characteristics, and the use of a low-density consumer EEG headset, the findings should be interpreted as exploratory interaction profiles and physiological proxies rather than definitive population neural differences.

In recent years, researchers have increasingly explored Virtual Reality (VR) as an effective tool for autism interventions, given its ability to offer immersive and personalized experiences. VR environments offer a controlled, safe space for practicing social interactions, communication, and daily activities (Calderone et al., 2024; Halabi et al., 2017; Yazdanian et al., 2023), with numerous studies reporting positive outcomes (Farashi et al., 2024; Karami et al., 2021; Li et al., 2023). However, the effectiveness of VR-based interventions significantly depends on the usability and comfort of the interaction techniques provided to the users.

In VR design, a variety of interaction techniques can be employed, which Bowman broadly categorizes under three universal tasks: navigation, selection, and manipulation (Bowman, 2002). For instance, locomotion techniques determine how users move through a virtual space, while selection and manipulation techniques determine how they interact with virtual objects. Since techniques vary in motor demands, abstraction, and continuity, selecting the right one for a given context is critical.

Crucially, autistic users have diverse sensory preferences and cognitive profiles (Leekam et al., 2007). This variability implies that a single “best” interaction technique is unlikely to serve all users equally well. Most evaluations in autism-focused VR studies still rely on subjective measures, such as self-reports and third-party observations (Darmasti et al., 2023). While valuable, self-reports can be prone to bias, given that some autistic users may have difficulty articulating internal states or emotional responses (Pearl et al., 2017), and observer judgements might overlook subtle cognitive or emotional indicators.

In light of these factors, researchers have increasingly advocated complementary objective measures to enrich interpretation and reduce bias (Cooper et al., 2018; Kim et al., 2024). One such measure is electroencephalography (EEG), which can provide a physiological perspective on how users respond during VR interaction. In a few autism-related VR works, EEG has been used to examine cognitive load or sub-bands activity during tasks (Chrisilla et al., 2020), and studies with neurotypical users suggest that brainwave band frequency changes can signal mental workload or attentional states (Chang et al., 2022; Gupta et al., 2019). However, this evidence remains fragmented as most previous works involving EEG are primarily focused on designing an intervention tool or examining general task responses rather than comparing how specific interaction techniques differ across evaluative measures.

This study aims to address this gap by examining two pairs of widely used VR interaction techniques: Teleport vs. Step for locomotion, and Hand-tracking vs. Controller for object interaction. These techniques are evaluated through three complementary lenses: trial duration (behavioral efficiency), EEG alpha and theta power spectral density (physiological proxies of task-related demand), and a direct perceived-ease prompt. Rather than identifying a single “optimal” technique, this study seeks to profile how these interaction techniques differ across measures and groups, and what these patterns may imply for interaction technique selection in autism-related VR contexts.

Background theory and related works

This section begins by presenting the state of the art of evaluation practices in autism-focused VR studies. It then introduces the four VR interaction techniques examined in this study (Teleport, Step, Hand-tracking, Controller) and explains the rationale for selecting the four techniques under comparison. Finally, it presents the reasoning for focusing on alpha and theta EEG activity and concludes with the research gap along with the guiding question of the study.

Evaluation framework in autism-related VR studies

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by challenges in social communication, social interaction and the presence of restricted and repetitive behaviors (Black and Grant, 2014). Individuals with autism also perceive and process sensory information differently (Dunn, 1997) (Miller et al., 2007). This divergence can manifest as either an increased sensitivity (hypersensitivity) to sensory stimuli, where everyday sounds, sights, or touches are experienced with overwhelming intensity, or as a decreased sensitivity (hyposensitivity), where significant sensory input is required to engage the sensory systems effectively (Roth, 2010). These sensory processing differences can influence behavior, preference, daily interaction (Kyriacou et al., 2023), and may also shape how autistic individuals experience VR interaction techniques. Given these sensory and cognitive differences, it becomes important to take a closer look at how researchers have historically assessed interaction in autism-related VR studies.

Research on VR interactions for autism has typically relied on subjective feedback, such as user questionnaires and interviews (Darmasti et al., 2023). For example, Bozgeyikli et al. assessed locomotion and object selection/manipulation techniques primarily through subjective user feedback (Bozgeyikli et al., 2016a; Bozgeyikli et al., 2018; Bozgeyikli et al., 2016b), noting preferences for joystick-based locomotion and point-and-teleport over more complex gestures. Mei et al. examined the impact of a VR customizable virtual human (VH) on autistic users’ experience by measuring gameplay success rates and collected self-reported questionnaire data (Mei et al., 2015). These studies provided valuable insights into users’ perceptions and preferences, while also showing that evaluations in this area have primarily emphasized subjective and user-reported perspectives.

Alongside these subjective approaches, some studies have begun to incorporate physiological measures as an additional layer of evaluation. Zhang et al. (Zhang et al., 2017) used EEG in a VR-based driving intervention context and showed that EEG-derived indices could capture cognitive load variation during VR task performance. Chrisilla et al. (2020) reported differences in EEG sub-band frequency powers between autistic and neurotypical participants during VR exposure, suggesting that VR experiences may be accompanied by measurable changes in neural activity. Studies involving neurotypical participants have similarly shown that VR interactions can influence brain activity and cognitive load (Chang et al., 2022; Gupta et al., 2019). Together, these studies suggest that EEG can provide a valuable complementary layer for evaluating VR interaction, motivating its inclusion alongside behavioral and subjective measures in the present study.

VR interaction techniques

Interaction techniques define how users move and interact within immersive environments and are commonly grouped into navigation, selection, and manipulation (Bowman, 2002). While “navigation” is the broader term used in foundational VR taxonomies, this study adopts the more specific term “locomotion” when referring to movement-based navigation techniques. To keep the investigation focused while covering core forms of VR interaction, this study examines two fundamental modes: locomotion and object interaction. These domains were selected because they are widely used across VR applications (Bowman et al., 2004; Jerald, 2015) and involve distinct forms of control and motor involvement, making them suitable as comparison targets in a multi-method evaluation.

Locomotion techniques

As one of the most frequently performed tasks in VR environments (Bowman et al., 2004; Di Luca, 2021), locomotion is central in VR. It enables users to move through and explore virtual spaces. In this study, two locomotion techniques are compared: Teleportation enables users to move instantly to a target location without visually traversing the space (Mine, 1995). It is widely adopted for its effectiveness in minimizing motion sickness (Bowman et al., 2004), a common challenge in VR, thus considered especially suitable for individuals with heightened sensory sensitivities.

Step moves the user in short, discrete increments using the joystick, preserving a sense of continuity relative to instant teleport. This technique offers a smoother motion than teleportation, with visual effects that mimic actual steps. However, it may increase task demand and the risk of motion sickness, as it requires users to process more complex motion cues.

These two locomotion techniques were selected because they represent a critical trade-off in VR navigation demands. Teleport minimizes continuous optic flow and self-motion cues by allowing discrete relocation to a target location, while Step preserves more continuous movement and spatial-updating demands through joystick-mediated incremental locomotion. Step was selected instead of walking-in-place, point walking, or other more embodied locomotion techniques because the study combined VR with EEG in a field-based setting. More embodied locomotion techniques would have introduced larger physical space requirements, safety demands, and motion artifacts, which were less suitable for the combined EEG-HMD setup and the heterogeneous participant sample. Comparing these two locomotion techniques which differ in continuity, abstraction, and potential sensory demands therefore provided a practical way to examine how locomotion choices relate to differences in user performance, physiological responses, and perceived-ease.

Object interaction techniques

Object interaction is a core part of many VR experiences as it reflects real-world actions such as picking up, moving, and interacting with objects. The present study compares two object interaction methods: ray-casting/controller and hand-tracking: Ray-casting/Controller functions as a point-and-click interface, offering users a laser-pointer-like mechanism to select and manipulate objects (Pietroszek, 2018). As the most commonly employed selection method in VR (Liu et al., 2023), it is valued for its precision, ease of learning and alignment with conventional interaction metaphors such as the computer mouse–features that benefit new VR users. This technique is referred to as “Controller” throughout the remainder of this article.

Hand-tracking, on the other hand, enables controller-free interaction using natural hand movements (Buckingham, 2021). This technique allows users to engage directly with virtual objects through gestures, offering a more intuitive experience. Its potential to enhance immersion and presence has made it increasingly relevant in modern VR applications.

By comparing these two methods, this study examines how two contrasting interaction styles–familiar and structured (controller) versus immersive and intuitive (hand-tracking)–are experienced and performed in a neurodiverse setting.

Alpha and theta EEG sub-bands

EEG signals can be decomposed into several frequency bands, with the most commonly recognized being Delta (0.5–4 Hz), Theta (4–8 Hz), Alpha (8–13 Hz), Beta (13–30 Hz), and Gamma (30–100 Hz) (Ernst Niedermeyer and Fernando Lopes da Silva, 2005). While all of these bands can provide useful information, this study focuses on alpha and theta because prior work has frequently associated them with task-related attention and workload-related processes during interaction. In the present study, these bands are treated as practical physiological indicators to complement behavioral and self-report measures in comparing VR interaction techniques.

Alpha (8–13 Hz) – Attentional engagement proxy

Alpha activity is often discussed in relation to attention regulation and task engagement. Importantly, alpha band activity is highly context-dependent (e.g., task engagement versus relaxed or idling states). Therefore, alpha is treated here as a comparative task-context proxy and interpreted conservatively in terms of relative differences across conditions and changes over repeated exposure, rather than as an absolute indicator of internal state. In line with prior literature (Jensen et al., 2002; Palva and Palva, 2011; Sauseng et al., 2005), decreases in alpha power (alpha suppression) are interpreted in this study as reflecting stronger task-related attentional engagement and active information processing. Conversely, increasing alpha power is interpreted as reduced suppression or rebound over time, which may be consistent with lower task-related engagement demand. In the context of this study, alpha is used as a proxy for how strongly participants are engaged with the VR task at a given point in time.

Theta (4–8 Hz) – working-memory effort proxy

Theta activity has often been associated with working-memory demand and task control (Jensen and Tesche, 2002; Missonnier et al., 2006; Zakrzewska and Brzezicka, 2014). Prior studies report that theta power tends to increase in tasks requiring higher cognitive effort or increased task difficulty (Mizuhara et al., 2004; Schapkin et al., 2020). Conversely, decreasing theta power may indicate diminishing cognitive demand as a task becomes more familiar or more automated over repeated trials. In this study, theta is therefore used as a proxy for working-memory effort during VR interaction.

Moreover, previous research has reported differences in alpha and theta oscillations among autistic individuals (Larrain-Valenzuela et al., 2017; Port et al., 2019; Shen et al., 2024), highlighting the relevance of investigating these sub-bands in neurodiverse user research. In addition to motivating the selection of alpha and theta, recent studies have also underscored the value of EEG in linking physiological data with subjective experiences. Raufi et al. found that EEG band ratios (alpha-to-theta and theta-to-alpha) could effectively classify self-reported mental workload (Raufi and Longo, 2022). At the same time, discrepancies between EEG findings and subjective evaluations have also been reported, particularly when sensory or emotional factors shape users’ self-reports (Yoto et al., 2007).

Importantly, alpha and theta dynamics here are not treated as direct readouts of internal cognitive states, but as proxies that complement behavioral and self-report measures in comparing VR interaction techniques. To maintain a focused analysis, other bands such as beta and gamma were excluded, as they were less directly aligned with the present study’s emphasis on task-related demand during VR interaction.

Research gap and question

Despite the growing body of work on VR for autism, specific interaction techniques in autism-related VR contexts are still rarely compared through an integrated evaluation framework (Darmasti et al., 2023). Existing studies have often emphasized either subjective experience, performance, or general task-related brain activity, but less often examined how these measures jointly characterize the use of particular interaction techniques. As a result, it remains unclear how different evaluative measures converge or diverge when comparing commonly used VR techniques in neurodiverse users.

To address this gap, the present study applies a multi-method evaluation framework to two pairs of widely used VR interaction techniques: Teleport versus Step for locomotion, and Hand-tracking versus Controller for object interaction. The study examines these techniques using three complementary measures: trial duration as a behavioral indicator, alpha and theta power spectral density as physiological proxies related to task demand, and a direct perceived-ease prompt as a subjective indicator. The primary research questions guiding this study are: RQ1: How do EEG alpha and theta proxies, behavioral performance, and perceived-ease vary across VR interaction techniques in autistic and neurotypical users?

RQ2: What do these combined measures imply for interaction technique selection in autism-related VR contexts?

To operationalize these questions, the following hypotheses are formulated per outcome domain: H1 (behavioral adaptation): across repeated trials, participants will show behavioral adaptation in both technique pairs, and differences in adaptation rates between techniques are expected.

H2 (alpha as attentional engagement proxy): different interaction techniques will show different alpha PSD patterns across repeated trials and may differ between participant groups.

H3 (theta as working-memory effort proxy): different interaction techniques will show different theta PSD patterns across repeated trials and may differ between participant groups.

H4 (perceived-ease): self-reported perceived-ease will not always align with the technique showing the strongest behavioral efficiency or the lowest physiological demand proxy, suggesting that subjective experience captures dimensions that are not fully reflected in objective measures.

The following section describes the study design, data collection procedure, and analysis strategy used to answer these questions.

This section details the methodological approach used in this study. It begins by describing the participant demographics and recruitment process, followed by a comprehensive outline of the experimental design, including the hardware, virtual tasks, and session procedures. Subsequently, it details the data collection and processing pipeline for all data streams and concludes by specifying the statistical approach used for the analyses.

Participants

Two participant groups were recruited using purposive sampling strategies appropriate to each context. The autistic group was recruited through local institutions, schools, and autism-focused organizations in Bandung, Indonesia, following ethical approval granted by KEP UNPAD 1 and official permission obtained from Bakesbangpol Kota Bandung 2. The neurotypical group was recruited from the Indonesian community in Berlin, Germany. This recruitment strategy was chosen to provide a neurotypical comparison sample while preserving linguistic and cultural comparability across groups. All participants, or their parents/caregivers where applicable, provided informed consent prior to participation.

For the autistic group, recruitment was conducted in a field setting where standardized clinical diagnosis documentation within participating institutions was often limited. Autism status was therefore identified through collaboration with local institutions and caregivers based on available records and participant information, while the intellectual functioning was assessed using the Wechsler-Bellevue Intelligence Scale (WBIS) (Wechsler, 1939), individually administered by certified Indonesian psychologists. The WBIS was utilized due to practical constraints specific to this context, including licensing costs, the absence of nationally representative normative data for newer Wechsler editions, and limited availability of clinicians trained to administer them, making it the most feasible instrument available for this field study.

Eighteen individuals were initially enrolled in the autism group; two enrolled participants were unable to complete the experiment. In both cases, the discontinuation was related to behavioral regulation challenges rather than technical discomfort; one remained persistently engaged with a preferred scenario and became distressed when required to transition, while the other was highly distracted and disengaged, which made the session incompletable. The final autism sample comprised 16 participants. The broad age and IQ range within this group reflects the practical realities of the recruitment landscape in this context. In Indonesia, limited awareness of neurodevelopmental disabilities and the absence of centralized early intervention services often contribute to delayed clinical diagnosis and heterogeneous institutional populations. This heterogeneity was treated as an ecologically grounded property of the sample, reflecting the diversity of autistic users encountered in real-world community settings. This broader demographic variability is examined in more depth in a companion study specifically focused on individual differences (Darmasti et al., 2025), whereas the present study concentrates on technique-level patterns within and across groups.

The neurotypical group consisted of 18 volunteers with no self-reported autism diagnosis. Recruiting participants from the same cultural and linguistic background as the autistic group was a deliberate methodological choice to reduce cultural and linguistic variability as a confounding factor. The purpose of this group was not to create a tightly matched control sample, but to provide a comparative reference point for interpreting technique-level patterns across measures. For the neurotypical group, IQ information was obtained through participant-reported prior IQ and was available for 16 of 18 participants.

As shown in Table 1, the autistic and neurotypical groups differed in age range, gender distribution and available IQ/intellectual-functioning information. The autistic field sample included both adolescents and adults, whereas the neurotypical comparison group consisted of adults only; the autistic group also had a higher proportion of male participants. These demographic characteristics reflect the practical realities of field-based recruitment in this context and should be considered when interpreting between-group comparisons. Furthermore, the available IQ/intellectual-functioning information presented in Table 1 is included only as contextual participant-characterization information and should not be treated as psychometrically equivalent across groups.

Experimental design and apparatus

This section describes the experimental setup, including the devices used, hardware selection and configuration, the experimental task, and the overall session flow.

Device selection

VR headset

This study utilized Oculus Quest 2, a standalone head-mounted display (HMD) with six degrees of freedom (6-DoF) tracking. It was selected because its lightweight design and untethered setup helped reduce physical burden during the session, which was particularly important given the sensory sensitivities present in part of the sample. Its display resolution (1832 × 1920 pixels per eye) and onboard processing were sufficient to support a stable and responsive VR experience throughout the experiment. Its compatibility with the concurrent EEG setup was also verified through internal testing prior to deployment.

EEG headset

EEG data were recorded using the Muse S Gen 2 (InteraXon Inc.), a dry-electrode consumer EEG headset. The device was chosen because it provided a practical balance between signal acquisition, comfort, and compatibility with the VR hardware. Compared with higher-density wet-electrode systems, the Muse S reduced setup time and physical burden for the stackable hardware configuration (EEG + VR), which was important in a field-based study involving autistic participants and repeated VR interaction. The headset records EEG data at 256 Hz from four electrodes (AF7, AF8, TP9, TP10), and the electrode placement is illustrated in Figure 1.

Hardware setup

Participants were first fitted with the Muse S Gen 2 headband, after which the Oculus Quest 2 was placed on top. The straps for both devices were carefully adjusted to ensure a secure fit while minimizing the pressure on the EEG electrodes and reducing the likelihood of motion artifacts. A final comfort and fit check was performed before the session began. The hardware configuration is presented in Figure 2.

Virtual tasks and environment

The virtual environment and tasks were designed to be minimalist in order to reduce unnecessary cognitive demand and to isolate the effects of the interaction technique. The experiment consisted of two main task blocks: a locomotion task and an object interaction task.

Locomotion task

Participants navigated along a simple virtual road by following arrow cues through a sequence of points to a finish line. This task was completed using two locomotion techniques: Teleport and Step. During development, the locomotion parameters were carefully calibrated so that both techniques allowed participants to reach the target destination in approximately the same minimum completion time, when used in their most efficient manner. Scene snapshots of the locomotion task are presented in Figure 3.

Object interaction task

In the virtual workspace, participants were required to pick up five objects from a desk and place them in a basket. This task was completed using two object interaction techniques: Hand-tracking and Controller. Scene snapshots of the interaction task are shown in Figure 4.

The step technique was implemented using the OVR Step Locomotion component (speed = 2, step distance = 3, cooldown time = 0.4 s), and Teleport was implemented using an XR projectile-curve ray (with configuration of velocity = 7, acceleration = 9.8, additional ground height = 0.1, additional flight time = 0.5, sample frequency = 20). Since Teleport relies on instantaneous Raycast-based relocation while Step uses continuous joystick-mediated movement, these parameters are not directly equivalent. During application development, the locomotion parameters were iteratively calibrated through developer testing so that both techniques allowed participants to reach the target destination in approximately the same minimum completion time when used in their most efficient manner, and thereby reducing built-in timing advantages between techniques.

Procedure and session flow

All participants experienced all four techniques. To minimize potential order effects, both the order of the two task blocks (locomotion and object interaction), and the order of techniques within each block were counterbalanced across participants using randomized assignment. The resulting distribution of the order after counterbalancing was comparable between both groups (see Table 2).

Each technique followed the same structure: one guided trial with the experimenter, followed by a series of independent trials. The study aimed for five independent trials per technique in addition to the guided trial. Each experiment session followed the procedure below: Orientation and baseline: participants began in a neutral VR lobby. A two-minute baseline EEG recording was obtained while participants engaged in a guided relaxation session in the MALOKA VR application (version 1.4.3).

Practice and trials (Task block #1 – Technique #1): the experimenter introduced the technique and assisted the participant through one guided trial. Participants then completed the independent trials.

Inter-technique break: a 1-min rest period was provided in the virtual lobby.

Practice and trials (Task block #1 – Technique #2): the same procedure was repeated for the second technique within the same task category.

Perceived-ease query: after completing both techniques within the task block, participants were asked for the technique they perceived as easier.

Inter-task break: after a short break, Steps 2–5 were repeated for the second task block.

Debriefing: the session concluded with final feedback and a short debriefing period.

Figure 5 provides an overview of the approximate timing of the procedure. The full session typically lasted approximately 15–25 min. When needed, research assistants provided support if participants became visibly confused or physically stuck, while preserving the participant’s autonomy in completing the independent trials.

To support dependable EEG acquisition in the field setting, data quality safeguards were applied both before and after each session. At the start of each session, signal quality was checked by confirming stable electrode contact and ensuring that the headset fit did not introduce excessive pressure. A brief VR calibration was also performed to verify tracking accuracy and reduce motion-related artifacts. During the data collection period, raw EEG streams and trial recordings from early sessions were periodically reviewed to identify any recurring issues such as signal dropouts. Fitting adjustments (Muse placement and/or VR strap tension) were also applied before subsequent sessions to maintain signal stability.

Data collection and processing

This section describes how behavioral, subjective, and EEG data were collected and prepared for analysis. It includes the operationalization of performance and perceived-ease measures, followed by the EEG preprocessing and feature extraction pipeline used to derive physiological indicators.

Performance metrics

In this study, performance is operationalized as trial duration (in seconds), measured from task initiation to task completion for each trial. Trial duration was selected as a task indicator of efficiency during repeated interaction, since the tasks were designed to be completed successfully and did not include standardized error logging (e.g., failed grabs or path deviation). It therefore captures how quickly participants achieved the task goal under each interaction technique, regardless of whether inefficiency arose from cautious movement, slower coordination, or other factors.

Furthermore, although the VR system was designed to log trial completion automatically using UNIX timestamps, preliminary inspection revealed occasional inaccuracies in the automated logs. A manual verification procedure was therefore applied. Two video sources were used for cross-checking: a direct screen recording of the participant’s in-headset view and an external camera recording of the participant and the casted VR screen. These sources were used to verify and, where necessary, correct the start and end times of each trial. Manual timestamp correction was performed by one trained-rater and a second-rater reliability estimate was not available. After validation, the resulting performance data were organized into a participant-by-technique trial dataset.

Subjective measures

As part of a broader semi-structured interview, participants were asked a direct perceived-ease question after completing each task block: “Which technique felt easier?”. This phrasing was chosen to minimize ambiguity and reduce unnecessary interpretive burden, particularly given evidence that autistic individuals may experience difficulties with figurative language and implicit meaning (Happe and Frith, 1996). These responses were treated as a perceived-ease indicator rather than as comprehensive usability assessments.

EEG processing

The EEG processing pipeline consisted of preprocessing, baseline normalization, trial-level feature extraction, and final metric selection. All signal processing and feature extraction were implemented in MATLAB using documented built-in functions from the Signal Processing Toolbox. Custom scripts were used to automate the workflow (batch processing, trial segmentation and data structuring), while core filtering and PSD estimation relied on standard MATLAB routines.

Given the low-density channel configuration of the headset and the study’s focus on band-limited power features, the processing pipeline prioritized conservative filtering, careful trial segmentation, and exclusion of unstable segments rather than component-based decomposition. The conceptual and technical details for each processing stage are described below.

Preprocessing

The raw EEG data were recorded at 256 Hz. Each channel was first filtered using a 4 th -order Butterworth notch filter (48.5–51.5 Hz) to remove power-line noise, followed by a 4 th -order Butterworth bandpass filter (0.5–80 Hz) to remove very low-frequency drift and high-frequency noise. The filtered signal for each channel was then z-score normalized; consequently, all PSD and power values are dimensionless and reported in arbitrary units (a.u.).

Baseline extraction

To derive a reliable baseline, the initial relaxation recording was segmented into 8-s epochs. Following the FASTER statistical thresholding recommendations for artifact rejection (Nolan et al., 2010), epochs with signal variance in the lowest 25th percentile were identified as the most stable and averaged to form the baseline reference for each participant.

Trial feature extraction

Trial data were derived from the filtered EEG signals. For each trial, the EEG was filtered into the alpha and theta ranges using 4 th order Butterworth filters. The resulting band-filtered signals from all four channels were then spatially averaged to obtain a single low-dimensional time series for each band. This averaging strategy was chosen to provide a robust summary of band activity in a low-density headset, while reducing the influence of channel-specific noise. Two metrics were extracted from each trial and from the selected baseline segments, as formally defined in Equations 1, 2: Time-domain Band Power (referred to as ‘Power’): calculated as the mean squared amplitude of the spatially averaged EEG signal within each epoch. where x ¯ t is the spatially averaged EEG amplitude at time t.

Power Spectral Density (referred to as ‘PSD’): computed using Welch’s method with a 2-s sliding window, 50% overlap, and a 1024-point FFT. Welch’s method estimates PSD as: where w k t is the window function applied to segment k, and F · denotes the Fourier transform. Band-specific PSD values were obtained by numerical integration of the PSD across the alpha and theta frequency ranges using MATLAB’s bandpower function.

Metric selection

The full processing pipeline yielded eight candidate EEG metrics, including Power, PSD, Normalized Power, and Normalized PSD for both alpha and theta. Comparative inspection of these candidate metrics, including normality and distribution checks across conditions, indicated that the non-normalized PSD measures (for alpha and theta) showed the most consistent distributions compared to time-domain power and normalized metrics. This was also aligned with theoretical expectations, where PSD is widely accepted as a robust and commonly used feature in EEG analysis, particularly for cognitive and frequency-based studies. Unlike time-domain measures, PSD estimates frequency content by averaging across multiple tapered windows, thereby reducing the influence of transient fluctuations (Gasser et al., 1985).

Furthermore, the baseline-normalized metrics, though valuable in theory, were found to be less stable than the non-normalized. This is likely due to the variability of the baseline session in the “relaxation” VR application, which inadvertently induced heterogeneous responses across participants. While some participants remained still and relaxed, others asked questions or explored the scene. This behavioral inconsistency likely propagated into the normalization procedure and therefore reduced the stability of the resulting normalized measures. For this reason, the non-normalized alpha and theta PSD metrics were selected as the primary physiological measures for subsequent analyses. At the same time, the behavioral variability observed during baseline recording highlights a methodological note for future work; when feasible, baseline acquisition should be conducted under more tightly controlled resting conditions to reduce between-participant variability in arousal and task engagement.

Quantifying adaptation

For each participant and interaction technique, EEG features and trial duration were extracted on a trial-by-trial basis, yielding a sequence of alpha PSD, theta PSD and trial duration values across repeated trials. Figure 6 shows an example of the extracted trial series for one technique from a single participant.

To quantify adaptation across repeated exposure, a linear slope was calculated for each metric (trial duration, alpha PSD, theta PSD) for every participant and technique (see the illustration in Figure 7). This was implemented using a simple linear regression across the repeated independent trials, expressed as y = bx + c, where y is the trial-wise metric value and x is the trial index. The slope coefficient b was used as the primary analytic feature because it captures both the direction and magnitude of change over repeated exposure. In this way, each participant’s trial sequence was summarized as a single adaptation value per metric and technique.

In a small number of sessions, participants completed fewer than five independent trials, while in other sessions, some participants completed more than five trials when they remained comfortable and wished to continue. Among participants retained in the final dataset, all completed at least four independent trials per technique. All available independent trials were included in the slope estimation, and no additional participants were excluded at the analysis stage.

Statistical approach

This section describes the statistical procedures used to analyze the processed dataset. Prior to inferential testing, Shapiro–Wilk tests (Shapiro and Francia, 1972) were used to assess the distribution of each participant group and interaction technique condition. Across the tests, 62.5% of tested distributions were non-normal (see Supplementary Table S1 for the full test results). Given the relatively small sample size in each group (N < 30) and the substantial proportion of non-normal data, nonparametric approaches were used for all inferential analyses. The analyses were organized into three parts: Analysis 1: Comparisons of Adaptation Slopes across Techniques and Groups

Objective: To examine how adaptation patterns differed across interaction techniques and participant groups, and over repeated trials.

Method: Included both descriptive statistics and inferential components.

Descriptive statistics were used to summarize the direction and magnitude of adaptation slopes for both performance and EEG measures across techniques and groups.

Wilcoxon signed-rank tests were used for within-subject comparisons of techniques within each task pair (Teleport vs. Step, Hand-tracking vs. Controller).

Mann-Whitney U tests were used for between-group comparisons (autism vs. neurotypical) of slope values for each technique.

Analysis 2: Perceived-ease in Relation to Behavioral and EEG Adaptation Measures

Objective: To examine whether participants’ self-reported perceived-ease were aligned with the corresponding behavioral and EEG slope measures. This analysis was treated as descriptive and exploratory.

Method: Participants were first grouped according to the technique they perceived as easier within each task pair (Teleport vs. Step; Hand-tracking vs. Controller). Within each subgroup, the technique perceived as easier and harder were then compared using subgroup averages of the slopes. Alignment was coded separately for trial duration, alpha PSD, and theta PSD according to the following criteria:

Performance alignment (behavioral adaptation): perceived-ease is coded as aligned if the technique perceived as easier showed a more negative trial-duration slope (faster improvement) than the alternative technique.

Alpha alignment (attentional engagement proxy): perceived-ease is coded as aligned if the technique perceived as easier showed a less negative or more positive alpha PSD slope than the alternative technique, indicating relatively reduced alpha suppression or greater rebound over trials.

Theta alignment (working-memory proxy): perceived-ease is coded as aligned if the technique perceived as easier showed a more negative theta PSD slope than the alternative technique, indicating decreasing working-memory demand across repeated exposure.

Analysis 3: Association between EEG and Performance Adaptation

Objective: To examine whether changes in physiological measures (EEG-derived indicators) were associated with changes in behavioral efficiency (performance) over repeated trials.

Method: For each participant and interaction technique, alpha and theta PSD slopes were correlated with trial duration slopes using Spearman’s rank correlation.

Effect sizes were reported as rank-biserial correlations for Wilcoxon signed-rank and Mann-Whitney U tests. Spearman’s ρ is reported as the effect-size estimate for correlation analyses. No formal correction for multiple comparisons was applied; p-values are therefore unadjusted and should be interpreted cautiously. Taken together, these analyses provided a triangulated basis for interpreting behavioral adaptation, EEG-derived indicators, and perceived-ease across VR interaction techniques. The results are presented in the following section.

This section reports the empirical findings in three parts. It first presents adaptation patterns across interaction techniques based on trial duration, alpha PSD and theta PSD slopes. It then reports the alignment between perceived-ease and objective measures. Finally, it presents the correlations between EEG and performance slopes.

Adaptation patterns across interaction techniques

Participant adaptation was examined using slopes of trial duration, alpha PSD, and theta PSD computed across available independent trials for each participant and each technique. To facilitate comparison across interaction techniques, Figure 8 summarizes the group-mean trial duration, alpha PSD, and theta PSD slopes for all four interaction techniques. Detailed analysis results for the locomotion (Teleport vs. Step) and object interaction (Hand-tracking vs. Controller) tasks are presented separately in the following subsections. Full descriptive statistics, inferential test results, and full distribution plots for this subsection are provided in Supplementary Material (Supplementary Tables S2–S8; Supplementary Figures S1–S2).

Locomotion (teleport vs. step)

Performance slopes

As shown in Table 3, both autistic and neurotypical participants showed improved performance across trials for the locomotion techniques, evidenced by the consistent negative means for trial duration slopes. In both groups, the mean slope was more negative for Teleport than for Step. The autism group showed greater variability in trial-duration slopes than the neurotypical group.

Alpha PSD slopes

A clear divergence was observed in the alpha band. Between-group comparisons using the Mann–Whitney U test showed a significant difference in alpha slopes for the Step technique (U = 40, p < 0.001, rank-biserial r = −0.722). Descriptively, the autistic group showed negative mean alpha PSD slopes, whereas the neurotypical group showed positive mean alpha PSD slopes. Figure 9 shows the participant data distribution of this Step alpha pattern. Furthermore, within-group comparisons using the Wilcoxon signed-rank test showed a significant difference for the Teleport vs. Step condition in the neurotypical group (W = 39, p = 0.043, rank-biserial r = −0.544).

Theta PSD slopes

Theta slopes showed no significant within-group or between-group differences for either locomotion technique. Descriptively, the autistic group showed more negative theta slopes for both Teleport and Step, whereas the neurotypical group showed a negative mean slope for Teleport and a near-zero positive mean slope for Step.

Object interaction (hand-tracking vs. controller)

Performance slopes

As shown in Table 4, both groups again showed performance improvement over time with both techniques, as indicated by the negative mean of trial duration slopes. In the neurotypical group, the mean slope was more negative for Controller than for Hand-tracking, while in the autism group, the mean slopes were negative for both techniques and relatively similar in magnitude.

Alpha PSD slopes

No significant within-group or between-group differences were observed for alpha slopes in either object interaction technique. Descriptively, the neurotypical group showed positive mean alpha slopes for both Hand-tracking and Controller, whereas the autism group showed negative mean alpha slopes for both techniques.

Theta PSD slopes

No significant between-group or within-group differences were observed for theta slopes in either object interaction technique. Descriptively, the autism group showed slightly positive theta slopes for both object interaction techniques.

Alignment of perceived-ease with adaptation patterns

This subsection reports the alignment between participant’s stated perceived-ease and the corresponding performance, alpha, and theta slope patterns, based on the predefined alignment criteria described in Section 3.4. In the neurotypical group, participants who perceived Teleport (N = 10) as easier showed alignment across all three measures. On the other hand, participants who perceived Step (N = 8) as easier showed no alignment across any of the measures. For object interaction task, participants who perceived Controller (N = 16) as easier showed alignment in performance and theta, but not in alpha. Participants who perceived Hand-tracking (N = 2) as easier showed no alignment across any of the measures. These patterns are summarized in Table 5.

In the autism group, participants who perceived Teleport (N = 12) as easier showed alignment in performance, but not in alpha or theta. Meanwhile, participants who perceived Step (N = 4) as easier showed alignment in performance and alpha slope, but not in theta. For object interaction, participants who perceived Hand-tracking (N = 6) as easier showed alignment across all three measures, whereas participants who perceived Controller (N = 10) as easier showed alignment in performance and theta, but not in alpha. These patterns are summarized in Table 6. Across both groups, the observed alignment patterns varied by interaction pair and by measure. Full subgroup values are provided in Supplementary Material (Supplementary Tables S9–S12).

EEG–performance correlations

Table 7 presents Spearman’s correlations between EEG and trial duration slopes for each interaction technique. Significant positive correlations between alpha and theta slopes were observed in several conditions, including Teleport in both groups, Controller in the neurotypical group, and Step in the autism group. However, correlations between EEG and trial duration slopes were largely non-significant. The p-values above 0.05 are reported for transparency but are not interpreted as evidence of association.

Taken together, the results show that behavioral, physiological, and stated perceived-ease patterns varied across technique pairs and participant groups, with no single measure telling a consistent story across all conditions.

This study compared two pairs of VR interaction techniques using three complementary evaluative lenses: behavioral performance, EEG-derived indicators, and perceived-ease. Across the results, these measures did not always converge. The clearest differentiation appeared in the locomotion pair, particularly between Teleport and Step, whereas the object interaction pair showed more mixed patterns. In addition, the two participant groups did not show identical profiles across techniques and measures. Taken together, these findings suggest that interaction technique evaluation in autism-related VR contexts benefits from considering multiple measures in parallel, rather than relying on behavioral efficiency, physiological indicators, or perceived-ease alone.

Performance gains, but measures diverge

A first pattern observed in the present sample is that participants generally became faster across repeated trials in all four techniques. However, the EEG and subjective measurements showed that these behavioral changes did not necessarily imply a corresponding shift in physiological and subjective states. In practical terms, participants could become more efficient at completing the task while still showing ongoing task-related adaptation demands in the EEG proxies or did not necessarily perceive the technique as being easier.

This pattern is consistent with prior observations that behavioral and physiological indices can dissociate (Yoto et al., 2007). In addition, this also aligns with the foundational work of Nielsen and Levy (Nielsen and Levy, 1994), which highlighted that performance and perceived-ease do not always align; it underscores that objective gains in efficiency do not inherently signify a parallel reduction in users’ subjective or physiological burden.

Locomotion: continuity matters

The clearest divergence appeared in the locomotion task. In both groups, Teleport showed more negative mean trial-duration slopes than Step, suggesting faster short-term behavioral adaptation. This pattern is consistent with the broader VR locomotion literature, where teleportation is often treated as a practical default because it minimizes optical flow and is commonly associated with lower risk of VR sickness (Prithul et al., 2021). At the same time, other work has also shown that teleportation can reduce continuity of movement (Clifton and Palmisano, 2020) and weaken presence (Slater and Steed, 2000). In this study, this trade-off logic provides a useful interpretive lens: Teleport appears to be advantageous when the immediate design priority is efficient onboarding and lower short-term demand.

This interpretation becomes more compelling as locomotion was also the domain in which the strongest group-sensitive EEG pattern emerged. For Step, autistic participants showed negative alpha slopes whereas neurotypical participants showed positive mean alpha slopes, and this between-group difference was statistically significant. Since alpha is interpreted conservatively here as a task-related proxy, this pattern is best understood as evidence of different adaptation dynamics rather than a marker of higher workload. A key distinction between Teleport and Step is sensory continuity; unlike Teleport, Step preserves continuous self-motion cues and optic flow, which may remain adaptively demanding even when task completion becomes faster.

This sample-specific between-group pattern may also be interpreted through the lens of sensory continuity within the Step technique. Unlike Teleport, Step preserves incremental self-motion cues and optic-flow changes, which may remain adaptively demanding even as task completion becomes faster. One possible explanatory framework is predictive processing (Clark, 2013): continuous motion cues may require sustained sensory monitoring when incoming sensory input is not easily discounted through prior expectations. This suggests that locomotion techniques differing in continuity and abstraction may interact differently with users’ sensory-cognitive profiles, even when overall performance improved in both cases. However, since the autistic field sample is also not tightly matched with the neurotypical reference group, the present data cannot explicitly determine whether this pattern is attributable to the autism diagnosis, intellectual functioning, developmental factor, gender distribution or their combination.

Object interaction: directness shapes perceived-ease

The object interaction findings tell a somewhat different story. Both Hand-tracking and Controller showed negative mean trial-duration slopes in both groups, again indicating short-term behavioral adaptation. Unlike locomotion, however, the object interaction pair did not yield strong between-group differences in either behavioral slopes or EEG proxies. The following descriptive patterns are therefore interpreted cautiously and used only to contextualize the technique profiles observed in this sample. Neurotypical participants showed faster behavioral adaptation with Controller, whereas autistic participants showed relatively similar behavioral adaptation across Hand-tracking and Controller. This descriptive pattern is compatible with prior VR work showing that controllers can provide more reliable input, faster completion times, and lower perceived workload than hand tracking in reach-grab-place tasks, particularly when technical issues like tracking loss or self-occlusion interfere with gesture-based input (Hameed et al., 2023).

At the same time, the subgroup perceived-ease findings show that Controller efficiency is not the whole story. In the autism group, participants who perceived Hand-tracking as easier showed an alignment across performance, alpha, and theta adaptation patterns. This is particularly meaningful because Hand-tracking differs from Controller not only in hardware but also in the translation it requires between intention and action.

This can be understood through the foundational HCI concept of the Gulf of Execution–the gap between a user’s intention and the physical actions required by the interface to achieve it (Norman, 1988). Hand-tracking, by mirroring natural hand movements, likely presents a much smaller Gulf of Execution than an abstracted tool like a controller, reducing the translation cost associated with controller mappings and button sequences. This also resonates with the concept of embodied interaction, where action and meaning are grounded in bodily engagement rather than in abstract commands (Dourish, 2001) and the concept of direct manipulation (Shneiderman, 1983), as Hand-tracking is often theorized to reduce the distance between user intention and system action. This perceived-ease pattern observed in a small subgroup of autistic participants, suggests cautiously that for users who may find abstraction effortful, the intuitiveness–directness, and congruence of an interface may contribute to its overall usability.

However, this does not mean Hand-tracking is universally better. The present results do not support a simple claim of Hand-tracking superiority. Rather, the findings point to a trade-off: the Controller remains advantageous when accuracy and reliability are central, whereas Hand-tracking may provide a better experiential and adaptation fit for users who benefit from lower abstraction and more direct-action mapping.

The multi-method evaluation framework

Beyond the divergence between performance and EEG patterns, the perceived-ease findings further illustrate this broader interpretation. In several conditions, stated ease did not fully align with the technique showing the strongest behavioral improvement or the most favorable EEG adaptation pattern. This divergence should not be interpreted as a weakness of the subjective measure; rather, it reflects the fact that subjective and objective measures are not designed to capture the same construct. Trial duration in the present study appears to capture short-term procedural adaptation, alpha and theta PSD proxies capture a different dimension of how users adapt to interaction demands over repeated exposure, and the perceived-ease prompt captured a low-burden experiential judgement.

A useful way to frame this is in terms of interaction-technique profiling: rather than asking which technique is globally best, the triangulated approach asks what kind of profile each technique produces across behavioral efficiency, physiological adaptation, and stated ease. This logic is also well supported by the theory of Task–Technology Fit, which argues that performance benefits depend on the match between the technology and the task it supports (Goodhue and Thompson, 1995).

The illustration of this triangulation is presented in Figure 10. This multi-method approach is particularly beneficial for autistic participants, whose motor proficiency, engagement, and subjective comfort may each tell a different yet equally important part of the usability story–and where no single measure alone would be sufficient to reveal it.

Design implications

The practical implications of the study are therefore conditional: the most suitable technique depends on the combination of behavioral efficiency, task-related physiological adaptation, and perceived-ease required for a given user and task context. From that perspective, the following technique-specific recommendations are conditional and goal-dependent: Teleport appears to be a reasonable option when the immediate priority is behavioral efficiency or physical ease and rapid task mastery. As the processing demand is relatively lower than the continuous movement, it allows users to achieve movement goals with lower mental effort, making it suitable for tasks such as onboarding, tutorials, or productivity applications.

Step may remain valuable in scenarios where continuity of movement is important, or when users prefer more spatially continuous navigation.

Hand-tracking may be a preferable choice for tasks aimed at achieving an intuitive and natural user experience. It also may offer a coherent fit for users who benefit from lower abstraction and more direct-action mapping.

Controller remains a strong practical baseline for object interaction, especially when the tasks demand higher precision and predictability. Its deterministic nature can provide a stable interaction, which is useful in scenarios with a low tolerance for error.

Ultimately, these recommendations are conditional starting points. In practice, they are best treated as hypotheses to be checked via a short and iterative loop: profile the user, run brief trials, compare performance with engagement proxies (e.g., alpha/theta) and expressed perceived-ease, then adjust. Combined with profile information (e.g., prior VR exposure, motivation), this approach helps select techniques that match the immediate goal such as efficiency, engagement, precision, or intuitiveness. This fit-based implication is also supported by the companion manuscript on the same dataset, which showed that individual factors such as age, IQ, and technological familiarity shaped performance and EEG outcomes differently across groups (Darmasti et al., 2025). More broadly, it also aligns with neurodiversity-affirming XR work and recent adaptive-VR reviews that emphasize customization and personalization as core design directions (Maddalon et al., 2024; Schmidt and Glaser, 2026).

Limitations and future work

While this study yields valuable insights into VR interaction techniques, several limitations should be acknowledged. First, a central limitation concerns group comparability. The autistic and neurotypical groups differed substantially in IQ/intellectual-functioning information, age range, gender distribution, recruitment context, and prior technology/VR experience. These factors limit causal interpretation of between-group comparisons, particularly for the EEG-derived and performance outcomes. Furthermore, although intellectual functioning in autistic group was assessed using WBIS as the most feasible instrument in the local field context, WBIS is an older measure and may limit comparability with contemporary IQ assessments. Its use also raises concerns about contemporary validity and possible Flynn-effect distortion (Flynn, 1984; Trahan et al., 2014), which therefore compromises comparability with studies using newer instruments. The findings should therefore be understood as exploratory patterns from a small and heterogeneous autistic field sample, rather than as stable estimates of autistic users as a population. In addition, autism is heterogeneous in sensory profile, communication style, motor ability, support needs, and prior technology experience; these dimensions were not systematically characterized in the present study and may have contributed to the observed variability.

Second, EEG was recorded with a four-channel dry-electrode consumer headset integrated with an HMD. This improved feasibility and comfort in a field-based setting, but it also limited spatial specificity and made the EEG measures more appropriate as practical physiological proxies than as localized neural markers. Recent VR-EEG reviews note exactly these trade-offs, including hardware integration challenges, limited recording sites, and signal-quality constraints (Li et al., 2024). Third, wearable and dynamic EEG recordings remain vulnerable to motion artifacts, which persist as a primary interpretive challenge in movement-rich settings (Giangrande et al., 2024).

Fourth, the modest sample size limits the statistical power of the analysis and the generalizability of the findings. Expanding the study with larger cohorts is needed to further validate the results and explore more subtle effects. Fifth, the interactive baseline protocol introduced participant variability and future work should use a standardized resting-state baseline.

Finally, future work could strengthen this line of research by using more controlled baselines, combining EEG with other low-burden measures and modeling trial-level data more directly. Future studies could expand the multi-method framework by incorporating other physiological metrics, such as other EEG sub-bands (e.g., beta, gamma) or complementary measures (e.g., eye tracking, heart rate variability), to provide an even richer picture of the user’s states and finally moving toward adaptive VR systems.

This study evaluated VR interaction techniques using a multi-method evaluation framework, combining performance as an indicator of efficiency, EEG as a proxy for task-related engagement, and succinct self-reports for perceived ease. The findings suggest that VR interaction techniques in autism-related contexts are best understood as multidimensional profiles rather than as techniques to be judged by which performs best on average. In this sense, a profiling approach asks what pattern each technique produces across behavioral efficiency, physiological adaptation, and perceived ease, and whether that pattern fits the user and task at hand.

Across both groups, participants showed consistent behavioral adaptation in all four techniques, but the accompanying EEG and perceived-ease patterns varied by interaction type and subgroup. The clearest divergence appeared in locomotion, where Step showed a group-sensitive alpha adaptation pattern despite behavioral improvement. In the present sample, autistic participants showed sustained alpha suppression across repeated Step trials, while neurotypical participants showed alpha rebound. However, this between-group pattern should be treated as an exploratory observation, as the two groups differed substantially in IQ/intellectual functioning, age distribution, and other participant characteristics. The present data therefore cannot determine whether this pattern reflects autism-specific profiles or broader group differences. In the object interaction task, Hand-tracking showed full alignment across measures for a subgroup of autistic participants who perceived it as easier, despite stronger performance gains with Controller overall.

At a practical level, the findings suggest that no single interaction technique universally outperforms the others. Instead, the most suitable choice depends on the user profile and task goal. Teleport appears especially useful when rapid task mastery and lower processing demand are prioritized, while Step may remain valuable where spatial continuity matters. For object interaction, Hand-tracking offers a more direct and low-abstraction experience, while Controller provides reliable precision in error-sensitive tasks. Rather than fixed prescriptions, these should therefore be treated as conditional starting points to be refined through brief iterative profiling that combines behavioral performance, physiological engagement proxies, and perceived ease.

Taken together, the findings address both guiding questions of this study. For RQ1, all four techniques generally supported behavioral adaptation, but EEG and perceived-ease measures diverged in technique- and group-specific ways, most notably in the alpha dynamics of Step. For RQ2, this divergence is itself the key implication: interaction technique selection in autism-related VR contexts is best approached as profiling rather than universal optimization. By presenting a multidimensional view that balances efficiency, engagement, and ease, this work may help highlight pathways to support the development of more inclusive and impactful VR technologies, where neurodiversity is not an edge case to accommodate, but a design reality to embrace.