Feasibility of the Meta Quest 3 for upper limb kinematic analysis: accuracy, consistency, agreement and reliability Authors: Gloria Campos, Gabriela Calvo, Adrián Borrego, Luis Millán, Álvaro Page, Roberto Llorens Published: 2026 ABSTRACT BackgroundUpper limb motor impairment is a frequent long-term consequence of a wide range of neurological and musculoskeletal conditions. Accurate assessment of motor function is essential to determine patient’s condition and plan adequate interventions. While clinical scales are easy to administer and not time-consuming, they could have limited accuracy and be biased. Kinematic analysis can provide detailed, quantitative information but usually depends on expensive, complex, and non-portable systems. Recent virtual reality head-mounted displays, with integrated hand-tracking, offer a promising, low-cost, and portable alternative to conduct upper limb kinematic assessment. However, the accuracy, consistency, agreement and reliability of kinematic variables derived from the hand tracking feature of these devices, as well as the impact of signal preprocessing on feature extraction, remain largely unexplored.MethodsThirty-nine healthy subjects performed five upper-limb tasks including tracking, point-reaching, and reach-and-grasp, while their movements were simultaneously recorded using a laboratory-grade motion capture system and the Meta Quest 3 headset. Multiple kinematic variables were extracted from the hand trajectories of both systems and analysed to evaluate accuracy, consistency, agreement and reliability. Additionally, the impact of signal smoothing on the extracted kinematic values was examined.ResultsMinimal absolute and relative errors were estimated between systems, showing high concordance and consistency across most kinematic variables and tasks. Importantly, the differences observed remained below the minimal detectable change of the motion capture system. Test-retest reliability of kinematic measures estimated by the Meta Quest 3 was moderate to good for nearly all variables with the only exception of the number of movement units, which consistently showed poor reliability across all tasks, indicating limited robustness of this metric. The signal smoothing was found to influence the magnitude of the estimated errors, highlighting the importance of optimizing the preprocessing to maximize the accuracy of the Meta Quest 3.ConclusionsOur findings support the feasibility of using the Meta Quest 3 as a cost-effective and portable alternative to laboratory-grade motion capture systems for kinematic evaluation of upper-limb motor function. However, caution is warranted regarding the selection of variables and tasks, particularly for metrics with limited robustness.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12984-026-02038-y. Background Upper limb motor impairment is a frequent long-term consequence of several neurological and musculoskeletal conditions, which are associated to disability and reduced functional independence [1]. A comprehensive and reliable assessment of motor function is essential to guide upper limb rehabilitation. In clinical practice, standardized scales and functional tests are commonly used, as they are easy to administer and not time-consuming. However, these instruments often rely on subjective scoring and may lack sensitivity to subtle improvements in sensorimotor performance [2, 3]. Spatiotemporal and kinematic measures obtained from different instrumented devices such as inertial sensors [4, 5] or camera-based motion capture systems (mocap) [6–8] can potentially overcome these limitations by providing objective, precise, and detailed information about the quality of movement, such as speed, smoothness and coordination [9]. The observed correlations between kinematic/spatiotemporal variables and clinical scales suggest that these measures capture clinically meaningful aspects of motor performance, supporting their potential value as objective markers of functional impairment [10]. Nevertheless, the clinical adoption of this kinematic assessment is limited by the high equipment costs, technical complexity, long setup times, and the need for controlled environment [11]. In addition, inertial and optical systems require attaching sensors or markers, respectively, to anatomical landmarks, which could lead to inaccurate localization and introduction of soft tissue or skin motion artifacts [12]. In the last years, less expensive and complex technological alternatives have been explored, including low-cost multitouch devices [13, 14] and depth sensors [15, 16], to overcome current limitations of kinematic analysis with promising results. In line with this, the latest generation of virtual reality head-mounted displays (HMDs), such as the Meta Quest 3 (Meta Platforms Inc., USA) feature markerless hand-tracking based on computer vision and artificial intelligence. These devices combine portability, affordability, ease of use, and eliminate the need for markers, potentially enabling broader application beyond specialized laboratories [17]. A few studies have compared HMD-based hand tracking with laboratory-grade optical mocaps powered by Vicon [18] (Vicon Motion Systems Ltd., UK) or OptiTrack [19] (NaturalPoint Inc., USA). Preliminary studies have reported small positional errors and strong correlations during simple reaching and manipulation tasks [17, 20–22], showing promising potential of the HMD-based hand trajectories. However, existing studies share some limitations that restrict the interpretation of the findings. First, most studies have focused on basic kinematic parameters (e.g., peak velocity and acceleration) without assessing other clinically meaningful aspects of movement such as smoothness and efficiency [23]. Second, existing works present methodological shortcomings that question their results. For instance, Borzelli et al. reported reliability metrics primarily reflecting association rather than absolute agreement, which may limit the detection of systematic bias between measurement systems [17]. Suwabe et al. investigated the validity of markerless hand tracking using the Lin’s Cross-correlation Coefficient [24], which combines Pearson correlation with a bias term to quantify agreement. However, this coefficient does not decompose sources of variability and does not provide complementary error indices, limiting its clinimetric utility. More robust indices, such as the intraclass correlation coefficient (ICC) to quantify reliability and the standard error of measurement (SEM) to estimate absolute error [25, 26], have rarely been applied in previous HMD validation studies. To the best of our knowledge, prior HMD validation studies have not reported ICC and SEM values within a unified framework that simultaneously addresses consistency, agreement, and intra-subject variability. Finally, previous studies with HMDs had commonly used low-pass filters with different cut-off frequencies to reduce noise [17, 20, 27]. However, no rationale has been provided for selecting those frequencies and alternative noise-reduction or preprocessing strategies have not yet been explored or implemented. Importantly, filter type and cut off frequency can significantly alter the estimation of movement parameters [28, 29]. This evidences the need to establish an optimal balance between noise reduction and preservation of the signal’s temporal fidelity [30], in order to optimize the clinical interpretation of the data provided by the HMD. Based on the existing evidence, we hypothesized that the HMD Meta Quest 3 could constitute a promising tool for estimating various kinematic characteristics of upper limb movement, provided that appropriate and accurate signal preprocessing is applied. To test the hypothesis, the present study pursued three objectives: first, to estimate the accuracy, consistency, and agreement of the Meta Quest 3 across several kinematic features relevant to upper limb function in comparison with a laboratory-grade mocap; second, to assess the reliability (defined as the degree to which repeated measurements obtained under the same conditions yield similar results) of HMD-based measurements; and third, to analyse the influence of signal smoothing on these features and define optimal pre-processing strategies. Methods Participants Healthy volunteers evenly distributed across the age range of 18 to 65 years were recruited from the staff and acquaintances of the research institute conducting the study. Recruitment criteria included (1) no medical history of neurological disorders, (2) no severe vision impairments and (3) no susceptibility to dizziness or discomfort associated with wearing head-mounted devices. A minimum of 30 participants was required based on an expected ICC of 0.75, null hypothesis of 0.5, three repeated measures per participant, a significance level of 0.05 and 80% power in the reliability analysis described below. Ten additional individuals were included to account for possible data loss and technical errors. After the experiment, the data of one participant was excluded for not completing the entire task. Consequently, the total sample included 39 healthy adults, 19 women (48.7%) and 20 men (51.3%), with a mean age of 41.4 ± 13.8 years, and 89.7% of them right-handed. The study was approved by the Ethics Committee of Universitat Politècnica de València (Approval No: P16280225). Written informed consent to participate in the study was obtained from all the participants before their enrolment. Instrumentation The movements of the participant’s dominant hand were simultaneously recorded using both a laboratory-grade mocap and the Meta Quest 3 headset while performing the different tasks. The mocap was equipped with 24 OptiTrack Prime14 and 4 OptiTrack Prime17W cameras, which recorded data within a 6 × 10 m calibrated volume at a sampling rate of 120 Hz. Passive infrared reflective markers were used to identify and track body segments. The Meta Quest 3, which features a Qualcomm Snapdragon XR2 Gen 2 processor and a screen resolution of 2064 × 2208 pixels per eye, provided additional hand tracking. The device is equipped with front-facing and side-facing cameras that enable full-colour pass-through and improved 6-DoF tracking. A custom ad hoc application was developed using Unity software (version 2022.3.10f1, Meta XR All-in-One SDK package [31]) to record the positions of all hand joints provided by the Meta SDK at a maximum frequency of 60 Hz. However, for the purposes of the present analysis, only the wrist position of the dominant hand was considered. Experimental procedure The experiment was conducted in the motion capture laboratory located at the Institute for Human-Centered Technology Research of the Universitat Politècnica de València (València, Spain). Participants were briefly introduced to the objective of the study. They were asked to sit in a chair positioned in front of a table, with their chest aligned to the central point of the target. The chair position was then adjusted to ensure they could comfortably execute the tasks. Each individual was then equipped with the virtual reality headset Meta Quest 3. The device was configured to operate in passthrough mode throughout the entire procedure, which allowed them to see the physical environment through the device’s cameras. A single reflective marker was positioned on the wrist, given that the HMD system provides tracking exclusively for the hand, and the analysis focused specifically on hand movement. Participants were required to perform 5 tasks: tracking in both a horizontal and vertical plane, point-reaching in both a horizontal and vertical plane and reach-and-grasp (Fig. 1). These tasks were selected based on previous literature, as they are commonly employed in kinematic assessments of upper-limb motor function and provide a controlled framework for the precise quantification of spatial and temporal movement features [9, 14, 32] For the horizontal and vertical tracking tasks, 5 points forming a semicircle were displayed on a plane, each positioned 30 cm from a central point and separated by 45° from one another around the center. Participants were instructed to reach the 5 targets in clockwise direction, always returning to the central point between targets, while continuously dragging their finger along the path (Fig. 1a). The point-reaching task in both planes consisted in 6 points arranged in 2 rows and 3 columns, with a separation of 15 cm between them, and a central resting point. Individuals were asked to touch the 6 points in random order, always returning to the central resting position between them and lifting the finger on the way to each target (Fig. 1b). Finally, for the reach-and-grasp task, an empty cup was positioned in front of the participants at a distance of 30 cm from the edge of the table. Participants were asked to reach forward to grasp the cup initiating the movement from the table edge, extend the arm to place the cup as far away as possible, release it, and return to the initial starting position (Fig. 1c). Fig. 1Experimental procedure of the study. A reflective marker was placed on the participants’ wrist for the motion capture system data acquisition. The participants were also equipped with the Meta Quest 3 headset, operating in passthrough mode. Participants completed 3 repetitions of five motor tasks: horizontal and vertical tracking (a), horizontal and vertical point-reaching (b), and a reach-and-grasp task (c). In the tracking tasks, five targets arranged as a semicircle (radius 30 cm; 45° spacing) were contacted sequentially in a clockwise order, with participants continuously dragging their finger and returning to the center between targets. In the point-reaching tasks, six targets were arranged in a 2 × 3 grid (15 cm spacing) around a central resting point. Participants touched targets in a random order, returning to the center between touches and lifting the finger during movements. In the reach-and-grasp task, an empty cup placed 30 cm from the table edge was grasped from the edge position, moved to the maximum reachable distance, released, and the hand returned to the start position. The gray points of the figure indicate the initial point or resting position. a Tracking task. b Point-reaching task; c Reach-and-grasp task Experimental procedure of the study. A reflective marker was placed on the participants’ wrist for the motion capture system data acquisition. The participants were also equipped with the Meta Quest 3 headset, operating in passthrough mode. Participants completed 3 repetitions of five motor tasks: horizontal and vertical tracking (a), horizontal and vertical point-reaching (b), and a reach-and-grasp task (c). In the tracking tasks, five targets arranged as a semicircle (radius 30 cm; 45° spacing) were contacted sequentially in a clockwise order, with participants continuously dragging their finger and returning to the center between targets. In the point-reaching tasks, six targets were arranged in a 2 × 3 grid (15 cm spacing) around a central resting point. Participants touched targets in a random order, returning to the center between touches and lifting the finger during movements. In the reach-and-grasp task, an empty cup placed 30 cm from the table edge was grasped from the edge position, moved to the maximum reachable distance, released, and the hand returned to the start position. The gray points of the figure indicate the initial point or resting position. a Tracking task. b Point-reaching task; c Reach-and-grasp task Participants engaged the 5 tasks sequentially and completed 3 repetitions for each of the 5 tasks, resulting in a total of 15 trials per individual. Short breaks of approximately 3–5 s were provided between repetitions to minimize fatigue, as well as 1-minute rests between tasks. Data analysis Signal preprocessing Recordings of the mocap underwent manual correction including artifact removal and gap-filling before being exported using Motive software (NaturalPoint Inc., USA) [33]. To ensure consistency across systems, mocap data was downsampled and exported at 60 Hz. HMD data was also resampled at 60 Hz as acquisition frequency included small fluctuations, which did not remain constant during recording. Both datasets were then aligned and preprocessed to enable direct comparison. Temporal synchronization was achieved using a functional approach based on cosine similarity between homologous wrist trajectories [34]. The procedure consisted of estimating a single global lag per trial, aligning the HMD and the mocap signals on a common 60 Hz timeline without applying non-linear time warping, and being the residual misalignment at most one frame (≈ 16.7 ms at 60 Hz). Spatial registration was then performed by estimating a rigid transformation to map HMD coordinates into the motion capture coordinate system [35]. After synchronization and spatial registration, signals from both systems were smoothed using a local regression approach with cubic fitting, treating the movement trajectories as smooth functions affected by random error. For each signal, the optimal bandwidth was determined independently by minimizing the autocorrelation of the residuals [36]. The HMD recordings were additionally smoothed by scaling the optimal bandwidth by factors of 1.5, 2.0, and 2.5. This strategy was not merely intended to reduce noise, but, following the objectives of the study, to assess how the degree of smoothing affects the kinematic variables estimated from the HMD. mocap Finally, kinematic features were computed from the smoothed trajectories of both systems after temporal synchronization and spatial registration, with all data expressed in the motion capture coordinate system. For the mocap, features were derived from a single optimally smoothed trajectory, whereas for the HMD, features were computed at each smoothing level (optimal bandwith and scaled bandwidths of 1.5, 2.0 and 2.5). This enabled both direct comparison between systems and evaluation of the preprocessing effects on measurement accuracy. Spatiotemporal and kinematic variables Existing studies on spatiotemporal and kinematic features of upper limb movement have focused on a broad set of variables involving joint angles and range of movement of the upper limb as well as trunk displacement, which have been showed acceptable reliability and reproducibility [37–39]. However, the HMD under study only provides information about the wrist, hand and fingers, and, consequently, the analysis was restricted to the wrist trajectories. Importantely, this information provides a meaningful basis for assessing upper limb function [40, 41]. Spatiotemporal and kinematic variables used in this study are described in Table 1. Table 1Description of the variables of the studyVariableDefinitionSpatiotemporal variablesTotal movement time (s)Elapsed time required to complete the entire task, measured from the onset of movement to the termination of the movement.Phase movement time (s)Elapsed time required to complete a movement phase.Kinematic variablesMean speed (cm/s)Average speed of the movement during the forward movements of the tracking and point-reaching tasks phases, and for the two phases of the reach-and-grasp task.Maximal speed (cm/s)Highest instantaneous speed reached during the forward movements of the tracking and point-reaching tasks phases, and for the two phases of the reach-and-grasp task.Curvature indexRatio between the actual length of the endpoint path and a straight-line distance connecting the initial and final positions [63], calculated for the forward movements of the tracking and point-reaching tasks phases, and for the two phases of the reach-and-grasp task.HarmonyCorrelation between the instantaneous distance from the initial point and the acceleration [64, 65] throughout total movement time.Number of movement unitsLocal velocity peaks where the amplitude between the preceding minimum and the next maximum exceeded 20 mm/s, and the time between successive peaks was at least 150 ms [38], calculated for total movement time.Root mean square of acceleration (cm/s2)Root mean square of the acceleration, measuring the overall magnitude of acceleration over total movement time. Table 1Description of the variables of the studyVariableDefinitionSpatiotemporal variablesTotal movement time (s)Elapsed time required to complete the entire task, measured from the onset of movement to the termination of the movement.Phase movement time (s)Elapsed time required to complete a movement phase.Kinematic variablesMean speed (cm/s)Average speed of the movement during the forward movements of the tracking and point-reaching tasks phases, and for the two phases of the reach-and-grasp task.Maximal speed (cm/s)Highest instantaneous speed reached during the forward movements of the tracking and point-reaching tasks phases, and for the two phases of the reach-and-grasp task.Curvature indexRatio between the actual length of the endpoint path and a straight-line distance connecting the initial and final positions [63], calculated for the forward movements of the tracking and point-reaching tasks phases, and for the two phases of the reach-and-grasp task.HarmonyCorrelation between the instantaneous distance from the initial point and the acceleration [64, 65] throughout total movement time.Number of movement unitsLocal velocity peaks where the amplitude between the preceding minimum and the next maximum exceeded 20 mm/s, and the time between successive peaks was at least 150 ms [38], calculated for total movement time.Root mean square of acceleration (cm/s2)Root mean square of the acceleration, measuring the overall magnitude of acceleration over total movement time. Description of the variables of the study Variables were estimated either from the entire trajectories or from specific phases of the trajectories. For the tracking and point-reaching tasks, each movement sequence from the central point to a target and back constituted a phase. For the reach-and-grasp task, two phases were defined: reaching the cup, and grasping and moving it forward. Variables included total and phase-specific movement times (MT); maximal speed, mean speed and endpoint trajectory deviations, determined by the curvature index, during the forward movements of each phase; and number of movement units (NMU) throughout total MT as an indicator of movement smoothness. Additionally, given the absence of a universally accepted standard smoothness metric [42], we proposed a movement harmony index, together with the root mean square (RMS) of acceleration as an indicator of movement abnormality [43]. Movement onset (beginning of wrist movement) was defined as the first time point where velocity exceeded 5 cm/s prior the first velocity peak above the 25th percentile of the velocity profile, and movement offset (end of wrist movement) as the first point where velocity dropped below 5 cm/s after the last velocity peak, also above the 25th percentile. Phase times for the tracking and point-reaching tasks were calculated by identifying the maximum peaks of the instantaneous distance from the initial movement position and determining the preceding and following time point at which velocity crossed the 5 cm/s threshold; for the reach-and-grasp task, movement phases were defined based on the velocity profile, using the same threshold criteria. The 5 cm/s velocity threshold was established to isolate intended goal-directed movement from minor postural adjustments of the arm unrelated to the task. Furthermore, this threshold ensures robustness against instrumental noise and inherent measurement fluctuations of the tracking system, which may present as residual motion even when the limb is static. This value proved to balance sensitivity to real movement with the exclusion of non-functional kinematic variations. Figure 2 shows a graphical example of a trial from the horizontal tracking task, illustrating how the trial time was segmented into different phases. Fig. 2Example of a horizontal tracking trial recorded with the motion capture system. The figure shows how the trial was divided into phases, with movement onset and movement offset marking the beginning and end of the movement. Both events were identified using a velocity threshold of 5 cm/s, which helped reduce the influence of small arm posture adjustments unrelated to the task, as well as residual fluctuations from the tracking system. These small signal variations, labeled as “noise” in the figure, may appear even when the limb is stationary. This threshold provided a practical balance between detecting true movement and excluding non-functional kinematic fluctuations Example of a horizontal tracking trial recorded with the motion capture system. The figure shows how the trial was divided into phases, with movement onset and movement offset marking the beginning and end of the movement. Both events were identified using a velocity threshold of 5 cm/s, which helped reduce the influence of small arm posture adjustments unrelated to the task, as well as residual fluctuations from the tracking system. These small signal variations, labeled as “noise” in the figure, may appear even when the limb is stationary. This threshold provided a practical balance between detecting true movement and excluding non-functional kinematic fluctuations Feature extraction based on these criteria, along with all preceding signal processing steps, was implemented in Python 3.10.11 using standard scientific libraries (NumPy [44] and SciPy [45]). Statistical analysis Statistical analysis covered measures of accuracy, consistency, agreement, and reliability. Before analysis, trials containing missing data due to tracking loss were removed, ensuring that at least two valid repetitions per task were retained for all participants. In total, only 7 trials were discarded [46]. ICC values obtained for the consistency, agreement and reliability analyses were interpreted as poor (ICC < 0.5), moderate (0.5–0.75), good (0.75–0.9) or excellent (ICC > 0.9) [25]. All data processing and statistical computations were performed in R (v4.4.2) [47]. Measurement accuracy A descriptive statistics including the median and IQR, and mean and standard deviation of absolute and relative errors (%) was reported for each variable and task to assess the accuracy of the HMD. Data normality was evaluated using the Shapiro-Wilk test. The non-parametric Wilcoxon signed-rank test was applied to establish statistical comparisons of the variables’ distributions for each system as normality was not always met. Consistency and agreement analysis Consistency metrics assess whether different measurement systems rank subjects in the same relative order, being insensitive to systematic bias. For instance, a high value of consistency between systems in our study would indicate the HMD is capable of evaluate subjects in the same way as the mocap, although measures could not be directly comparable between systems. Intraclass correlation coefficient ICC(3,1) and SEM [48] were computed to assess consistency based on a 2-way mixed-effects model following Eq. (1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{ijk}=\mu\:+{T}_{i}+{\gamma\:}_{j}+{\left(T\gamma\:\right)}_{ij}+{\epsilon\:}_{ijk}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{ijk}$$\end{document} represents the measure for the subject i, system j and repetition k; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mu\:$$\end{document} the overall mean; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{T}_{i}$$\end{document} the random effect of subject i; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\gamma\:}_{j}$$\end{document} the fixed effect of system j; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left(T\gamma\:\right)}_{ij}$$\end{document} the random effect associated with the interaction between the subject i and system j; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\epsilon\:}_{ijk}$$\end{document} the residual error for each measurement. The interaction term accounts for subject-specific differences in how systems perform, capturing variability not explained by subjects or systems alone. Absolute agreement metrics, in turn, evaluate if measurement systems produce identical or numerically equivalent values, accounting for any systematic biases between them. To quantify absolute agreement, ICC(2,1) was estimated using the proposed model of Eq. (1), with the measurement system treated as a fixed effect. This approach of estimating the measurements of agreement by the variance components makes the ICC(2,1) equivalent to the Concordance Correlation Coefficient under this model [49, 50]. The fixed effect of the measurement system was incorporated into the ICC calculation by estimating its empirical variance as the deviation of the means from the overall mean [51]. Comparison between ICC(2,1) and ICC(3,1) may help identify if discrepancies between the mocap and the HMD are primarily due to systematic bias or inherent unreliability of the HMD measurements. To further assess absolute agreement at the individual measurement level, Bland Altman analysis was performed with a linear mixed-effects model to account for the presence of repeated measures per subject [52, 53]. For subject i and repetition k, the difference between OptiTrack and HMD values (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{D}_{ik}$$\end{document}) was modelled by Eq. (2):2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{D}_{ik}={\mu\:}^{\mathrm{*}}+{\alpha\:}_{i}+{\epsilon\:}_{ik}$$\end{document} where µ is the global bias, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}_{i}\sim N(0,{{\sigma\:}_{u}}^{2})$$\end{document} the subject-specific random effect, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\epsilon\:}_{ik}\sim N(0,{\sigma\:}^{2})$$\end{document} the residual error. The bias corresponds to the mean systematic difference between systems, while the 95% limits of agreement (LoAs) were defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mu\:\pm\:1.96\sqrt{{{\sigma\:}_{u}}^{2}+{\sigma\:}^{2}}$$\end{document}. Confidence intervals for bias and LoAs were estimated by parametric bootstrap. Reliability analysis Reliability, serving in this study as an indicator of measurement repeatability, was assessed with the ICC and SEM independently calculated for each technology. Given the duration of the experimental procedure, participants could have experienced fatigue or learning effects, which might influence measurement reliability. Therefore, an additional mixed model shown in Eq. (3) was defined:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{ij}=\mu\:+{\alpha\:}_{i}+{{\delta\:}_{i}+\epsilon\:}_{ij}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{ij}$$\end{document} is the measure for subject i and system j; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mu\:$$\end{document} is the overall mean of the measurements; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}_{i}$$\end{document} is the random effect of subject i; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\delta\:}_{i}$$\end{document} is the fixed effect of the repetitions for subject i (capturing test-retest variability), and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\epsilon\:}_{ik}$$\end{document} the residual error. This modelling approach allows decomposition of variance components to obtain ICC and SEM estimates that properly reflect both intra-subject variability and repetition effects, with ICC representing the proportion of subject variance relative to total variance (excluding repetition), and SEM corresponding to the square root of the residual variance, reflecting measurement error. Finally, the minimal detectable change (MDC), which represents the magnitude of change necessary to exceed the measurement error of repeated measures, was calculated for the 95% CI [54]. In this context, the MDC helps determine whether changes in kinematic metrics reflect actual motor change or remain within the bounds of measurement error. Results The comparative analysis between the HMD measures estimated for the four different degrees of smoothing (no smoothing, x1.5, x2 and x2.5) revealed that the data smoothed with a factor of 1.5 provided the best balance between noise reduction and signal preservation in terms of absolute error. This smoothing factor was therefore retained for all subsequent analyses (see supplementary material). An example of this in the horizontal tracking task can be observed on Fig. 3. Fig. 3Wrist trajectory during a representative trial of the horizontal tracking task. The figure shows a 2D top-down view of the 3D trajectories of the wrist captured by both the Meta Quest 3 and the mocap during one trial of the horizontal tracking task Wrist trajectory during a representative trial of the horizontal tracking task. The figure shows a 2D top-down view of the 3D trajectories of the wrist captured by both the Meta Quest 3 and the mocap during one trial of the horizontal tracking task Measurement accuracy The accuracy assessment of the HMD compared to the mocap for each of the tasks is presented in Tables 2, 3 and 4. Graphical representations of the distributions of each variable are also available on Figs. 4, 5 and 6. Table 2Mesurements and discrepancies between the motion capture system and the Meta Quest 3 for the tracking tasksVariableMotion capture system[Mdn (Q1-Q3)]Meta Quest 3[Mdn (Q1-Q3)]Absolute error[Mean ± sd]Relative error[%]SignificanceHorizontal Total MT (s)10.30 (9.13–11.55)10.27 (9.16–11.50)0.08 ± 0.110.68%p < 0.01 Phase MT (s)1.91 (1.74–2.09)1.87 (1.72–2.06)0.04 ± 0.051.80%p < 0.01 Mean speed (cm/s)26.35 (23.71–29.27)25.69 (23.48–28.67)0.99 ± 0.703.77%p = 0.059 Maximal speed (cm/s)44.50 (39.44–49.83)45.44 (39.73–49.79)2.14 ± 1.814.76%p = 0.237 Curvature index1.01 (1.01–1.01)1.01 (1.01–1.02)0.01 ± 0.000.62%p < 0.01 Harmony0.36 (0.28–0.42)0.38 (0.32–0.45)0.04 ± 0.0312.70%p < 0.01 NMU5 (2–8)4 (2–8)0.83 ± 0.6518.02%p < 0.01 RMS acceleration (cm/s2)98.92 (83.73–119.26)92.90 (73.20–106.49)9.89 ± 6.419.57%p < 0.01Vertical Total MT (s)11.19 (10.31–12.83)11.16 (10.25–12.76)0.17 ± 0.171.39%p = 0.253 Phase MT (s)1.94 (1.81–2.21)1.96 (1.78–2.09)0.10 ± 0.094.62%p = 0.153 Mean speed (cm/s)20.65 (17.88–23.57)20.54 (17.47–23.85)0.91 ± 0.594.41%p = 0.900 Maximal speed (cm/s)32.25 (29.30–38.05)33.65 (28.74–39.38)1.79 ± 1.385.39%p = 0.028 Curvature index1.01 (1.01–1.02)1.02 (1.02–1.03)0.02 ± 0.011.06%p < 0.01 Harmony0.32 (0.24–0.37)0.36 (0.25–0.42)0.06 ± 0.0418.81%p < 0.01 NMU6 (3–8)5 (3–8)1.00 ± 1.1226.56%p < 0.01 RMS acceleration (cm/s2)72.47 (56.08–85.13)68.41 (51.22–80.89)8.73 ± 6.3111.02%p < 0.01MT: movement time; NMU: number of movement units; RMS: root mean squareStatistically significant differences between measures obtained by both systems are highlighted in bold Table 2Mesurements and discrepancies between the motion capture system and the Meta Quest 3 for the tracking tasksVariableMotion capture system[Mdn (Q1-Q3)]Meta Quest 3[Mdn (Q1-Q3)]Absolute error[Mean ± sd]Relative error[%]SignificanceHorizontal Total MT (s)10.30 (9.13–11.55)10.27 (9.16–11.50)0.08 ± 0.110.68%p < 0.01 Phase MT (s)1.91 (1.74–2.09)1.87 (1.72–2.06)0.04 ± 0.051.80%p < 0.01 Mean speed (cm/s)26.35 (23.71–29.27)25.69 (23.48–28.67)0.99 ± 0.703.77%p = 0.059 Maximal speed (cm/s)44.50 (39.44–49.83)45.44 (39.73–49.79)2.14 ± 1.814.76%p = 0.237 Curvature index1.01 (1.01–1.01)1.01 (1.01–1.02)0.01 ± 0.000.62%p < 0.01 Harmony0.36 (0.28–0.42)0.38 (0.32–0.45)0.04 ± 0.0312.70%p < 0.01 NMU5 (2–8)4 (2–8)0.83 ± 0.6518.02%p < 0.01 RMS acceleration (cm/s2)98.92 (83.73–119.26)92.90 (73.20–106.49)9.89 ± 6.419.57%p < 0.01Vertical Total MT (s)11.19 (10.31–12.83)11.16 (10.25–12.76)0.17 ± 0.171.39%p = 0.253 Phase MT (s)1.94 (1.81–2.21)1.96 (1.78–2.09)0.10 ± 0.094.62%p = 0.153 Mean speed (cm/s)20.65 (17.88–23.57)20.54 (17.47–23.85)0.91 ± 0.594.41%p = 0.900 Maximal speed (cm/s)32.25 (29.30–38.05)33.65 (28.74–39.38)1.79 ± 1.385.39%p = 0.028 Curvature index1.01 (1.01–1.02)1.02 (1.02–1.03)0.02 ± 0.011.06%p < 0.01 Harmony0.32 (0.24–0.37)0.36 (0.25–0.42)0.06 ± 0.0418.81%p < 0.01 NMU6 (3–8)5 (3–8)1.00 ± 1.1226.56%p < 0.01 RMS acceleration (cm/s2)72.47 (56.08–85.13)68.41 (51.22–80.89)8.73 ± 6.3111.02%p < 0.01MT: movement time; NMU: number of movement units; RMS: root mean squareStatistically significant differences between measures obtained by both systems are highlighted in bold Mesurements and discrepancies between the motion capture system and the Meta Quest 3 for the tracking tasks MT: movement time; NMU: number of movement units; RMS: root mean square Statistically significant differences between measures obtained by both systems are highlighted in bold Table 3Mesurements and discrepancies between the motion capture system and the Meta Quest 3 for the point-reaching tasksVariableMotion capture system[Mdn (Q1-Q3)]Meta Quest 3[Mdn (Q1-Q3)]Absolute error[Mean ± sd]Relative error[%]SignificanceHorizontal Phase MT (s)1.87 (1.71–2.05)1.87 (1.72–2.11)0.06 ± 0.053.00%p < 0.01 Mean speed (cm/s)32.84 (30.19–36.22)33.89 (30.37–36.56)1.59 ± 1.504.69%p = 0.240 Maximal speed (cm/s)67.60 (61.25–74.55)71.01 (63.54–78.92)3.97 ± 3.035.93%p < 0.01 Curvature index1.11 (1.07–1.16)1.12 (1.08–1.19)0.03 ± 0.032.41%p < 0.01 Harmony0.65 (0.60–0.69)0.65 (0.60–0.69)0.02 ± 0.033.27%p = 0.359 NMU21 (17–24)15 (13–17)5.68 ± 2.5327.06%p < 0.01 RMS acceleration (cm/s2)117.41 (104.51–132.51)95.58 (85.69–106.21)21.62 ± 8.3918.21%p < 0.01Vertical Phase MT (s)1.56 (1.41–1.67)1.66 (1.48–1.83)0.14 ± 0.179.69%p < 0.01 Mean speed (cm/s)25.24 (23.46–27.21)24.46 (2.22–26.38)1.84 ± 2.016.97%p = 0.048 Maximal speed (cm/s)46.07 (41.63–50.02)48.81 (44.15–53.54)3.84 ± 2.948.63%p < 0.01 Curvature index1.12 (1.09–1.15)1.11 (1.06–1.15)0.04 ± 0.032.76%p = 0.311 Harmony0.59 (0.54–0.64)0.61 (0.52–0.66)0.04 ± 0.036.02%p = 0.138 NMU17 (15–20)15 (13–18)2.90 ± 1.9216.39%p < 0.01 RMS acceleration (cm/s2)84.78 (74.80–94.81)64.72 (58.73–72.29)18.46 ± 8.1621.46%p < 0.01MT: movement time; NMU: number of movement units; RMS: root mean squareStatistically significant differences between measures obtained by both systems are highlighted in bold Table 3Mesurements and discrepancies between the motion capture system and the Meta Quest 3 for the point-reaching tasksVariableMotion capture system[Mdn (Q1-Q3)]Meta Quest 3[Mdn (Q1-Q3)]Absolute error[Mean ± sd]Relative error[%]SignificanceHorizontal Phase MT (s)1.87 (1.71–2.05)1.87 (1.72–2.11)0.06 ± 0.053.00%p < 0.01 Mean speed (cm/s)32.84 (30.19–36.22)33.89 (30.37–36.56)1.59 ± 1.504.69%p = 0.240 Maximal speed (cm/s)67.60 (61.25–74.55)71.01 (63.54–78.92)3.97 ± 3.035.93%p < 0.01 Curvature index1.11 (1.07–1.16)1.12 (1.08–1.19)0.03 ± 0.032.41%p < 0.01 Harmony0.65 (0.60–0.69)0.65 (0.60–0.69)0.02 ± 0.033.27%p = 0.359 NMU21 (17–24)15 (13–17)5.68 ± 2.5327.06%p < 0.01 RMS acceleration (cm/s2)117.41 (104.51–132.51)95.58 (85.69–106.21)21.62 ± 8.3918.21%p < 0.01Vertical Phase MT (s)1.56 (1.41–1.67)1.66 (1.48–1.83)0.14 ± 0.179.69%p < 0.01 Mean speed (cm/s)25.24 (23.46–27.21)24.46 (2.22–26.38)1.84 ± 2.016.97%p = 0.048 Maximal speed (cm/s)46.07 (41.63–50.02)48.81 (44.15–53.54)3.84 ± 2.948.63%p < 0.01 Curvature index1.12 (1.09–1.15)1.11 (1.06–1.15)0.04 ± 0.032.76%p = 0.311 Harmony0.59 (0.54–0.64)0.61 (0.52–0.66)0.04 ± 0.036.02%p = 0.138 NMU17 (15–20)15 (13–18)2.90 ± 1.9216.39%p < 0.01 RMS acceleration (cm/s2)84.78 (74.80–94.81)64.72 (58.73–72.29)18.46 ± 8.1621.46%p < 0.01MT: movement time; NMU: number of movement units; RMS: root mean squareStatistically significant differences between measures obtained by both systems are highlighted in bold Mesurements and discrepancies between the motion capture system and the Meta Quest 3 for the point-reaching tasks MT: movement time; NMU: number of movement units; RMS: root mean square Statistically significant differences between measures obtained by both systems are highlighted in bold Table 4Descriptive analysis and error between the motion capture system and the HMD Meta Quest 3 for the reach-and-grasp taskVariableMotion capture system[Mdn (Q1-Q3)]Meta Quest 3[Mdn (Q1-Q3)]Absolute error[Mean ± sd]Relative error[%]SignificanceEntire movement Total MT (s)3.45 (3.17–3.88)3.47 (3.10–3.89)0.11 ± 0.092.99%p = 0.046 Harmony0.24 (0.14–0.35)0.30 (0.16–0.39)0.05 ± 0.0411.11%p < 0.01 NMU1 (1–2)1 (0–1)0.45 ± 0.46p = 0.053 RMS Acceleration (cm/s2)160.50 (117.90–190.41)127.70 (91.51–163.82)28.75 ± 16.6218.59%p < 0.01Reaching phase Phase MT (s)0.70 (0.64–0.78)0.74 (0.67–0.80)0.04 ± 0.045.05%p < 0.01 Mean speed (cm/s)32.77 (27.49–37.06)30.74 (26.08–36.01)2.48 ± 1.987.58%p < 0.01 Maximal speed (cm/s)52.63 (45.87–61.10)52.33 (46.7–60.23)4.03 ± 3.357.58%p = 0.456 Curvature index1.10 (1.07–1.13)1.11 (1.07–1.13)0.02 ± 0.021.35%p < 0.01Grasp and forward phase Phase MT (s)1.5 (1.28–1.73)1.53 (1.33–1.72)0.08 ± 0.075.55%p < 0.01 Mean speed (cm/s)18.21 (14.11–23.65)18.19 (13.85–22.37)1.40 ± 1.107.42%p = 0.091 Maximal speed (cm/s)48.34 (40.06–59.90)46.34 (37.24–59.87)4.08 ± 3.098.75%p < 0.01 Curvature index1.16 (1.12–1.23)1.23 (1.16–1.32)0.08 ± 0.066.48%p < 0.01MT: movement time; NMU: number of movement units; RMS: root mean squareStatistically significant differences between measures obtained by both systems are highlighted in bold Table 4Descriptive analysis and error between the motion capture system and the HMD Meta Quest 3 for the reach-and-grasp taskVariableMotion capture system[Mdn (Q1-Q3)]Meta Quest 3[Mdn (Q1-Q3)]Absolute error[Mean ± sd]Relative error[%]SignificanceEntire movement Total MT (s)3.45 (3.17–3.88)3.47 (3.10–3.89)0.11 ± 0.092.99%p = 0.046 Harmony0.24 (0.14–0.35)0.30 (0.16–0.39)0.05 ± 0.0411.11%p < 0.01 NMU1 (1–2)1 (0–1)0.45 ± 0.46p = 0.053 RMS Acceleration (cm/s2)160.50 (117.90–190.41)127.70 (91.51–163.82)28.75 ± 16.6218.59%p < 0.01Reaching phase Phase MT (s)0.70 (0.64–0.78)0.74 (0.67–0.80)0.04 ± 0.045.05%p < 0.01 Mean speed (cm/s)32.77 (27.49–37.06)30.74 (26.08–36.01)2.48 ± 1.987.58%p < 0.01 Maximal speed (cm/s)52.63 (45.87–61.10)52.33 (46.7–60.23)4.03 ± 3.357.58%p = 0.456 Curvature index1.10 (1.07–1.13)1.11 (1.07–1.13)0.02 ± 0.021.35%p < 0.01Grasp and forward phase Phase MT (s)1.5 (1.28–1.73)1.53 (1.33–1.72)0.08 ± 0.075.55%p < 0.01 Mean speed (cm/s)18.21 (14.11–23.65)18.19 (13.85–22.37)1.40 ± 1.107.42%p = 0.091 Maximal speed (cm/s)48.34 (40.06–59.90)46.34 (37.24–59.87)4.08 ± 3.098.75%p < 0.01 Curvature index1.16 (1.12–1.23)1.23 (1.16–1.32)0.08 ± 0.066.48%p < 0.01MT: movement time; NMU: number of movement units; RMS: root mean squareStatistically significant differences between measures obtained by both systems are highlighted in bold Descriptive analysis and error between the motion capture system and the HMD Meta Quest 3 for the reach-and-grasp task MT: movement time; NMU: number of movement units; RMS: root mean square Statistically significant differences between measures obtained by both systems are highlighted in bold Fig. 4Absolute values of the spatiotemporal and kinematic variables for the horizontal and vertical tracking tasks. Boxplot comparison of the spatiotemporal and kinematic variables measured with the laboratory-grade motion-capture system (blue) and the Meta Quest 3 (orange) in the horizontal and vertical tasks Absolute values of the spatiotemporal and kinematic variables for the horizontal and vertical tracking tasks. Boxplot comparison of the spatiotemporal and kinematic variables measured with the laboratory-grade motion-capture system (blue) and the Meta Quest 3 (orange) in the horizontal and vertical tasks Fig. 5Absolute values of the spatiotemporal and kinematic variables for the horizontal and vertical point-reaching tasks. Boxplot comparison of the spatiotemporal and kinematic variables measured with the laboratory-grade motion-capture system (blue) and the Meta Quest 3 (orange) in the horizontal and vertical point-reaching tasks. RMS: root mean square Absolute values of the spatiotemporal and kinematic variables for the horizontal and vertical point-reaching tasks. Boxplot comparison of the spatiotemporal and kinematic variables measured with the laboratory-grade motion-capture system (blue) and the Meta Quest 3 (orange) in the horizontal and vertical point-reaching tasks. RMS: root mean square Fig. 6Absolute values of the spatiotemporal and kinematic variables for the reach-and-grasp task. Boxplot comparison of the spatiotemporal and kinematic variables measured with the laboratory-grade motion-capture system (blue) and the Meta Quest 3 (orange) in the reach-and-grasp task. RMS: root mean square Absolute values of the spatiotemporal and kinematic variables for the reach-and-grasp task. Boxplot comparison of the spatiotemporal and kinematic variables measured with the laboratory-grade motion-capture system (blue) and the Meta Quest 3 (orange) in the reach-and-grasp task. RMS: root mean square Errors from movement times, both total (total MT) and phase-specific (phase MT), were small across all tasks, being the total movement time the variable with the lowest relative error (below 3%), with no significant differences observed between measurement systems except in the horizontal tasks (Table 2). Speed related features also exhibited good performance, with relative errors ranging approximately between 3% and 8%. Finally, variables related with movement smoothness and quality were the most unstable across tasks, showing heterogeneous behaviours and reaching higher errors, with a maximum relative error of 21.5% for RMS acceleration in the point-reaching vertical task (Table 3). Consistency and agreement analysis Findings from the consistency and agreement analyses are presented in Tables 5, 6 and 7. Table 5Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the tracking taskVariableConsistencyAgreementBland-AltmanICC (3,1)SEMICC (2,1)Lower LoAsUpper LoAsBias% Out of LoAsHorizontal Total MT (s)0.780.800.78-0.340.430.059.41% Phase MT (s)0.630.130.62-0.150.210.036.96% Mean speed (cm/s)0.742.210.74-2.172.900.361.79% Maximal speed (cm/s)0.724.350.71-6.495.15-0.676.25% Curvature index0.280.010.19-0.020.01-0.012.18% Harmony0.490.070.47-0.140.08-0.034.96% NMU0.342.690.33-2.143.390.624.92% RMS acceleration (cm/s2)0.7812.410.74-10.0627.168.555.46%Vertical Total MT (s)0.850.780.85-0.740.840.055.00% Phase MT (s)0.690.140.68-0.300.370.035.74% Mean speed (cm/s)0.821.890.82-2.622.680.035.00% Maximal speed (cm/s)0.862.740.85-5.854.01-0.923.71% Curvature index0.300.010.17-0.030.01-0.016.18% Harmony0.520.070.50-0.170.12-0.036.90% NMU0.422.420.41-3.164.240.544.88% RMS acceleration (cm/s2)0.829.180.79-12.5125.626.567.90%MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Table 5Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the tracking taskVariableConsistencyAgreementBland-AltmanICC (3,1)SEMICC (2,1)Lower LoAsUpper LoAsBias% Out of LoAsHorizontal Total MT (s)0.780.800.78-0.340.430.059.41% Phase MT (s)0.630.130.62-0.150.210.036.96% Mean speed (cm/s)0.742.210.74-2.172.900.361.79% Maximal speed (cm/s)0.724.350.71-6.495.15-0.676.25% Curvature index0.280.010.19-0.020.01-0.012.18% Harmony0.490.070.47-0.140.08-0.034.96% NMU0.342.690.33-2.143.390.624.92% RMS acceleration (cm/s2)0.7812.410.74-10.0627.168.555.46%Vertical Total MT (s)0.850.780.85-0.740.840.055.00% Phase MT (s)0.690.140.68-0.300.370.035.74% Mean speed (cm/s)0.821.890.82-2.622.680.035.00% Maximal speed (cm/s)0.862.740.85-5.854.01-0.923.71% Curvature index0.300.010.17-0.030.01-0.016.18% Harmony0.520.070.50-0.170.12-0.036.90% NMU0.422.420.41-3.164.240.544.88% RMS acceleration (cm/s2)0.829.180.79-12.5125.626.567.90%MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the tracking task MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Table 6Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the point-reaching tasksVariableConsistencyAgreementBland-AltmanICC (3,1)SEMICC (2,1)Lower LoAsUpper LoAsBias% Out of LoAsHorizontal Phase MT (s)0.750.130.75-0.220.15-0.035.32% Mean speed (cm/s)0.672.630.67-5.124.40-0.356.60% Maximal speed (cm/s)0.814.300.77-11.745.82-2.966.19% Curvature index0.760.030.73-0.090.06-0.024.77% Harmony0.680.040.67-0.070.06-0.014.68% NMU0.293.610.16-3.1414.025.444.14% RMS acceleration (cm/s2)0.749.320.432.9440.3121.627.70%Vertical Phase MT (s)0.450.170.41-0.560.33-0.118.58% Mean Speed (cm/s)0.382.520.37-5.337.391.036.49% Maximal Speed (cm/s)0.643.940.57-12.626.45-3.085.72% Curvature Index0.520.040.52-0.100.110.016.67% Harmony0.690.050.69-0.120.10-0.017.02% NMU0.163.460.14-6.099.371.645.72% RMS Acceleration (cm/s2)0.5010.370.30-10.1343.9116.895.56%MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Table 6Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the point-reaching tasksVariableConsistencyAgreementBland-AltmanICC (3,1)SEMICC (2,1)Lower LoAsUpper LoAsBias% Out of LoAsHorizontal Phase MT (s)0.750.130.75-0.220.15-0.035.32% Mean speed (cm/s)0.672.630.67-5.124.40-0.356.60% Maximal speed (cm/s)0.814.300.77-11.745.82-2.966.19% Curvature index0.760.030.73-0.090.06-0.024.77% Harmony0.680.040.67-0.070.06-0.014.68% NMU0.293.610.16-3.1414.025.444.14% RMS acceleration (cm/s2)0.749.320.432.9440.3121.627.70%Vertical Phase MT (s)0.450.170.41-0.560.33-0.118.58% Mean Speed (cm/s)0.382.520.37-5.337.391.036.49% Maximal Speed (cm/s)0.643.940.57-12.626.45-3.085.72% Curvature Index0.520.040.52-0.100.110.016.67% Harmony0.690.050.69-0.120.10-0.017.02% NMU0.163.460.14-6.099.371.645.72% RMS Acceleration (cm/s2)0.5010.370.30-10.1343.9116.895.56%MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the point-reaching tasks MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Table 7Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the reach-and-grasp taskVariableConsistencyAgreementBland-AltmanICC (3,1)SEMICC (2,1)Lower LoAsUpper LoAsBias% Out of LoAsEntire movement Total MT (s)0.730.280.73-0.370.450.043.48% Harmony0.720.090.70-0.190.10-0.045.70% NMU0.100.820.10-1.511.950.228.95% RMS acceleration (cm/s2)0.7425.640.64-17.4172.1827.895.70%Reaching phase Phase MT (s)0.630.070.60-0.140.08-0.033.20% Mean speed (cm/s)0.694.350.68-7.2310.021.393.51% Maximal speed (cm/s)0.676.830.67-12.8711.58-0.645.27% Curvature index0.660.020.63-0.050.03-0.016.84%Grasp and forward phase Phase MT (s)0.740.180.74-0.340.28-0.033.54% Mean speed (cm/s)0.862.340.86-3.764.700.472.59% Maximal speed (cm/s)0.836.100.82-10.0314.332.155.00% Curvature index0.590.070.48-0.210.06-0.073.07%MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Table 7Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the reach-and-grasp taskVariableConsistencyAgreementBland-AltmanICC (3,1)SEMICC (2,1)Lower LoAsUpper LoAsBias% Out of LoAsEntire movement Total MT (s)0.730.280.73-0.370.450.043.48% Harmony0.720.090.70-0.190.10-0.045.70% NMU0.100.820.10-1.511.950.228.95% RMS acceleration (cm/s2)0.7425.640.64-17.4172.1827.895.70%Reaching phase Phase MT (s)0.630.070.60-0.140.08-0.033.20% Mean speed (cm/s)0.694.350.68-7.2310.021.393.51% Maximal speed (cm/s)0.676.830.67-12.8711.58-0.645.27% Curvature index0.660.020.63-0.050.03-0.016.84%Grasp and forward phase Phase MT (s)0.740.180.74-0.340.28-0.033.54% Mean speed (cm/s)0.862.340.86-3.764.700.472.59% Maximal speed (cm/s)0.836.100.82-10.0314.332.155.00% Curvature index0.590.070.48-0.210.06-0.073.07%MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement Consistency and agreement analysis between the motion capture system and the Meta Quest 3 for the reach-and-grasp task MT: movement time; NMU: number of movement units; RMS: root mean square; ICC: intraclass correlation coefficient; SEM: standard error of measurement; LoAs: limits of agreement For all tasks, both measures of consistency and agreement showed similar values, with agreement being slightly lower than consistency in some cases, reflecting only a minimal sensitivity to systematic bias. ICC values generally ranged from moderate to good, except for NMUs in all tasks and, more notably, for nearly all variables of the vertical point-reaching task (Table 6). SEM values calculated from the mixed-effects model (thus reflecting error between systems) were low and stable across tasks, indicating that the HMD provides measurements that are largely consistent with the mocap. Bias from Bland-Altman plots was close to zero in most cases, with the exception of RMS of acceleration, which consistently showed positive values, reflecting a more pronounced systematic bias for this variable. Overall, the percentage of samples falling outside the limits of agreement was generally around 4–8% (Tables 5, 6 and 7). Reliability analysis Results for the reliability analysis are presented in Tables 8, 9 and 10. Table 8Reliability analysis between the motion capture system and the Meta Quest 3 for the tracking tasksVariableMotion capture systemMeta Quest 3Test-retest reliabilitySEMMDCTest-retest reliabilitySEMMDCHorizontal Total MT (s)0.740.882.450.710.912.54 Phase MT (s)0.580.150.410.520.150.40 Mean speed (cm/s)0.702.236.190.692.537.02 Maximal speed (cm/s)0.684.2711.850.684.9213.64 Curvature index0.600.000.010.380.000.01 Harmony0.470.070.200.330.080.22 NMU0.203.108.580.202.877.95 RMS acceleration (cm/s2)0.8012.6635.080.7213.1336.41Vertical Total MT (s)0.820.862.390.820.862.39 Phase MT (s)0.770.130.350.620.150.41 Mean speed (cm/s)0.831.774.900.762.256.24 Maximal speed (cm/s)0.882.396.630.833.148.71 Curvature index0.590.000.010.320.010.02 Harmony0.460.060.180.520.080.22 NMU0.392.617.240.272.567.09 RMS acceleration (cm/s2)0.878.4323.380.838.2122.77MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Table 8Reliability analysis between the motion capture system and the Meta Quest 3 for the tracking tasksVariableMotion capture systemMeta Quest 3Test-retest reliabilitySEMMDCTest-retest reliabilitySEMMDCHorizontal Total MT (s)0.740.882.450.710.912.54 Phase MT (s)0.580.150.410.520.150.40 Mean speed (cm/s)0.702.236.190.692.537.02 Maximal speed (cm/s)0.684.2711.850.684.9213.64 Curvature index0.600.000.010.380.000.01 Harmony0.470.070.200.330.080.22 NMU0.203.108.580.202.877.95 RMS acceleration (cm/s2)0.8012.6635.080.7213.1336.41Vertical Total MT (s)0.820.862.390.820.862.39 Phase MT (s)0.770.130.350.620.150.41 Mean speed (cm/s)0.831.774.900.762.256.24 Maximal speed (cm/s)0.882.396.630.833.148.71 Curvature index0.590.000.010.320.010.02 Harmony0.460.060.180.520.080.22 NMU0.392.617.240.272.567.09 RMS acceleration (cm/s2)0.878.4323.380.838.2122.77MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Reliability analysis between the motion capture system and the Meta Quest 3 for the tracking tasks MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Table 9Reliability analysis between the motion capture system and the Meta Quest 3 for the point-reaching tasksVariableMotion capture systemMeta Quest 3Test-retest reliabilitySEMMDCTest-retest reliabilitySEMMDCHorizontal Phase MT (s)0.750.130.350.720.150.40 Mean speed (cm/s)0.672.547.040.652.847.86 Maximal speed (cm/s)0.853.399.400.824.5112.50 Curvature index0.870.020.060.760.040.11 Harmony0.750.030.090.620.040.10 NMU0.213.9010.820.433.138.68 RMS acceleration (cm/s2)0.808.4123.310.758.8424.50Vertical Phase MT (s)0.820.080.240.530.180.49 Mean speed (cm/s)0.751.544.260.372.657.34 Maximal speed (cm/s)0.772.908.050.644.2111.66 Curvature index0.820.020.050.550.040.12 Harmony0.860.030.080.670.050.14 NMU0.153.279.050.313.319.18 RMS acceleration (cm/s2)0.727.7821.570.648.7524.25MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Table 9Reliability analysis between the motion capture system and the Meta Quest 3 for the point-reaching tasksVariableMotion capture systemMeta Quest 3Test-retest reliabilitySEMMDCTest-retest reliabilitySEMMDCHorizontal Phase MT (s)0.750.130.350.720.150.40 Mean speed (cm/s)0.672.547.040.652.847.86 Maximal speed (cm/s)0.853.399.400.824.5112.50 Curvature index0.870.020.060.760.040.11 Harmony0.750.030.090.620.040.10 NMU0.213.9010.820.433.138.68 RMS acceleration (cm/s2)0.808.4123.310.758.8424.50Vertical Phase MT (s)0.820.080.240.530.180.49 Mean speed (cm/s)0.751.544.260.372.657.34 Maximal speed (cm/s)0.772.908.050.644.2111.66 Curvature index0.820.020.050.550.040.12 Harmony0.860.030.080.670.050.14 NMU0.153.279.050.313.319.18 RMS acceleration (cm/s2)0.727.7821.570.648.7524.25MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Reliability analysis between the motion capture system and the Meta Quest 3 for the point-reaching tasks MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Table 10Reliability analysis between the motion capture system and the Meta Quest 3 for the reach-and-grasp taskVariableMotion capture systemMeta Quest 3Test-retest reliabilitySEMMDCTest-retest reliabilitySEMMDCEntire movement Total MT (s)0.662.300.820.690.310.87 Harmony0.700.090.240.670.100.27 NMU0.100.882.490.000.802.22 RMS acceleration (cm/s2)0.808.4123.310.758.8424.50Reaching phase Phase MT (s)0.530.080.200.620.070.20 Mean speed (cm/s)0.694.1411.480.654.8413.43 Maximal speed (cm/s)0.686.3317.540.647.4820.75 Curvature index0.670.020.060.560.030.08Grasp and forward phase Phase MT (s)0.700.190.520.700.190.52 Mean speed (cm/s)0.862.406.650.862.366.53 Maximal speed (cm/s)0.825.9916.600.836.3717.66 Curvature index0.690.050.140.480.090.25MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Table 10Reliability analysis between the motion capture system and the Meta Quest 3 for the reach-and-grasp taskVariableMotion capture systemMeta Quest 3Test-retest reliabilitySEMMDCTest-retest reliabilitySEMMDCEntire movement Total MT (s)0.662.300.820.690.310.87 Harmony0.700.090.240.670.100.27 NMU0.100.882.490.000.802.22 RMS acceleration (cm/s2)0.808.4123.310.758.8424.50Reaching phase Phase MT (s)0.530.080.200.620.070.20 Mean speed (cm/s)0.694.1411.480.654.8413.43 Maximal speed (cm/s)0.686.3317.540.647.4820.75 Curvature index0.670.020.060.560.030.08Grasp and forward phase Phase MT (s)0.700.190.520.700.190.52 Mean speed (cm/s)0.862.406.650.862.366.53 Maximal speed (cm/s)0.825.9916.600.836.3717.66 Curvature index0.690.050.140.480.090.25MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change Reliability analysis between the motion capture system and the Meta Quest 3 for the reach-and-grasp task MT: movement time; NMU: number of movement units; RMS: root mean square; SEM: standard error of measurement; MDC: minimal detectable change The results of reliability highlighted the similarity between systems, with nearly identical values for spatiotemporal variables across the three tasks and similar values for the remaining measures. Test-retest ICCs indicated moderate to good reliability in the investigated variables, except from NMU, which showed poor reliability across all tasks, and harmony, with poor reliability only in the tracking task (Table 8). The only task that yielded notably lower values for the HMD compared with the mocap was the vertical point-reaching task, as evidenced by the mostly poor ICC values (Table 9). Alternatively, when considering measurements, the tasks with the lowest overall repeatability were the horizontal tracking task and the reaching phase of the reach-and-grasp task (Tables 8 and 10). The comparison of the SEM from the consistency analysis (Tables 5, 6 and 7) with MDC of the mocap (Tables 8, 9 and 10), evidenced that all SEM values were found to be below the corresponding MDCs, indicating that the measurement error between systems is smaller than the minimal change that the reference system could detect. Discussion This study investigated the accuracy, consistency and agreement with a laboratory-grade mocap, as well as reliability, of upper limb spatiotemporal and kinematic variables obtained with the Meta Quest 3 in healthy individuals. The results demonstrated that the HMD is capable of producing similar measurements compared to the mocap, with relative errors below 10% with the exception of smoothness-related variables, generally moderate to good consistency and agreement, and high repeatability in terms of reliability. Additionally, the impact of signal smoothing was also examined, demonstrating the importance of applying appropriate preprocessing and filtering procedures before extracting features from the signals. Although some differences between measures of both systems reached statistical significance, these minor discrepancies are unlikely to exceed prior reported thresholds considered clinically meaningful. A minimal change of approximately 15% in performance based on kinematic metrics has been suggested to constitute a clinically important difference [55, 56]. In the present study, the relative errors between the measures obtained by the Meta Quest 3 and the laboratory-grade mocap were consistently below this threshold. The only exceptions to this were NMU and the RMS of acceleration, which exceeded the 15% threshold in most cases. The discrepancies in these measures could be explained by the lack of reliability and repeatability of the NMU. As for the RMS of acceleration, it is important to consider this measure could be particularly susceptible to larger errors because acceleration corresponds to the second derivative of position, which amplifies high-frequency noise and small tracking inaccuracies. The higher relative errors during the vertical tracking task compared to the horizontal task may be due to the relative position of the hand with respect to the headset cameras during vertical movements, which can increase tracking difficulty. In particular, these movements may place the hand closer to the limits of the field of view or involve orientations that are less favorable for tracking, as well as small head movements that can further affect tracking accuracy. The moderate to good ICC values from the consistency and agreement analysis in both measurement systems showed that the majority of variability reflected real inter-subject differences rather than measurement error or differences between systems. The remaining variability, captured by the SEM and representing the measurement error between systems, remained below the MDC calculated for the mocap for all variables, with the only exceptions being the curvature index and harmony in the vertical tracking task. This indicates that the magnitude of disagreement between devices did not exceed the intrinsic repeatability limits of the reference system itself. In other words, the additional uncertainty introduced by the Meta Quest 3 remains within the natural measurement variability of the laboratory-grade mocap and is therefore unlikely to meaningfully alter the interpretation of the recorded kinematic outcomes. Furthermore, similarity between ICC(3,1) and ICC(2,1) values indicated that there is no substantial systematic bias between devices. In other words, the two systems do not differ consistently in their absolute measurement levels. Instead, any discrepancies between them are mainly random rather than systematic. This means that one method does not systematically overestimate or underestimate the measurements compared to the other, supporting that both systems produce comparable values in absolute terms. Nevertheless, from the perspective of functional assessment, where the primary interest often lies in detecting changes over time or differences between conditions, the presence of a fixed bias could be considered as irrelevant. The results of the Bland-Altman analysis further supported the comparability of measurements. The close-to-zero bias values indicated minimal systematic differences between devices. The presence of both positive and negative bias values across variables evidences no consistent pattern, suggesting that no systematic tendency occurred for the Meta Quest 3 to over- or underestimate measurements relative to the mocap. The low percentages of samples falling outside the LoAs suggested that large discrepancies between systems were rare. The comparable reliability between measures of both systems, with variables demonstrating good or bad reliability in one system also exhibiting similar behaviour in the other, suggests that measurement consistency is largely determined by the nature of the task and the specific variable under investigation rather than the technology used. Importantly, the moderate to good reliability demonstrated by nearly all variables confirms that both technologies provide stable measurements across repetitions. However, the reliability was not uniform across tasks, and the vertical point-reaching condition consistently showed lower agreement and repeatability compared to the remaining tasks. This directional dependency suggests that tracking performance may be influenced by movement configuration relative to the headset cameras. Although the present study was not designed to isolate the underlying cause, this finding highlights the importance of considering task geometry when interpreting HMD-derived kinematic measures. The exception constituted by the NMU, with poor reliability across the three tasks, indicate that measurements for the same subject were highly variable and cannot be considered trustworthy for detecting longitudinal changes under the conditions and tasks presented in this study. The poor reliability observed for NMU under the present conditions highlights its limited instrumental stability within this experimental framework. In contrast, RMS acceleration and movement harmony demonstrated greater robustness in terms of agreement and repeatability. However, their behaviour varied across tasks, indicating that the stability of kinematic descriptors depends on task characteristics. Withing the scope of this study, RMS acceleration and movement harmony may therefore constitute more stable instrumental variables than NMU for describing movement quality. An important practical observation from our study is that hand tracking with the Meta Quest 3 was highly robust. Across all participants and tasks, tracking loss was extremely rare: only 7 trials out of 585 (3 repetitions × 5 tasks × 39 subjects) had to be discarded, typically due to brief occlusions or movements momentarily outside the optimal field of view. This indicates that, under the experimental conditions used, which may not fully reflect the demands of other interactive tasks, the device reliably captured hand positions without substantial interruptions, further supporting its suitability for kinematic assessment in controlled research and clinical settings. Several of the kinematic variables analyzed in this study have already been correlated to functional motor performance clinical scales of upper-limb impairment [10]. Mean and maximal speed showed strong positive correlations with the Fugl-Meyer Assessment [57] and Wolf Motor Function Test [58], indicating that faster movements generally reflect better motor capacity; as well as total and phase-specific MT, whose negative correlation with these scales may reinforce that longer durations are associated with poorer performance. Smoothness related metrics showed positive correlations with the Modified Ashworth Scale [59], suggesting that these variables may be sensitive to increased muscle tone and spasticity. Finally, trajectory deviations, captured here as the curvature index, were mainly related to performance on the Box and Blocks Test [60], indicating that straighter and more efficient movements are associated with greater functional motor ability. Several limitations should be considered when interpreting the results of this study. First, participants performed the tasks while wearing the HMD, which may have influenced their natural movements and could differ from real-world performance. Although this does not compromise the validity of the findings, it should be considered when evaluating how well these movements correspond to those performed without the headset. Additionally, although movement time is a widely used spatiotemporal metric, it should be interpreted with caution in self-paced tasks, as it may reflect not only motor performance but also individual movement strategies and task engagement [61]. Second, all participants were healthy individuals without upper-limb deficits, which could limit the generalizability of the findings to people with upper limb deficits [62]. Further research is needed to specifically evaluate the use of Meta Quest 3 in clinical populations to confirm our findings. Third, the sample size was relatively small, which may limit the generalizability of the findings. Fourth, the selected tasks may not capture the full range of upper limb movements, so caution should be exercised when extrapolating these results to other tasks or functional contexts. However, despite these constraints, the methodology used in our study and its findings support that the Meta Quest 3 device offers a highly accessible option to optical systems, which require expensive cameras, dedicated laboratory space and high setup times. Its portability and ease of setup make kinematic analysis performed with the Meta Quest 3 suitable for clinics and research settings with limited resources, avoiding the logistical constraints of traditional motion capture. Most metrics demonstrated sufficient consistency and agreement, supporting their potential applicability for quantitative upper-limb assessment in research and structured clinical contexts. Nevertheless, it is important to acknowledge the observed results, most notably the poor reliability of the NMU variable, as well as lower reliability observed in the horizontal tracking and the reaching phase of the reach-and-grasp task, highlighting the need to select appropriate tasks and outcome measures. Conclusion This study evaluated the accuracy, consistency, agreement and reliability of upper-limb kinematic measurements obtained with Meta Quest 3 compared to OptiTrack, a laboratory-grade mocap, using a robust statistical framework. The results demonstrated that the HMD could provide measurements with small errors and moderate-to-good reliability for most variables, with discrepancies remaining below previously reported thresholds considered clinically meaningful in nearly all cases. The poorest results were observed for NMU, which showed higher variability and limited reliability despite being a widely used metric to estimate movement smoothness. The signal smoothing level affected the results, revealing that the use of a smoothing factor of x1.5 provides the optimal balance between noise reduction and signal preservation. Given its low cost, portability, and ease of setup, the Meta Quest 3 represents a promising and accessible option for upper-limb motion analysis in structured research settings and controlled clinical contexts. However, caution is warranted in task and variable selection. Future studies are needed to confirm these findings in clinical populations with upper-limb impairments and to validate the use of HMDs for rehabilitation assessment. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Supplementary Material 1.