This 2-stage cross-sectional development and usability study used single-session repeated-measures procedures in both stages. Stage 1 was a fixed-order, within-subjects exploratory data acquisition study with 40 participants (20 older and 20 younger adults), characterizing the time-series signal output of Balance Wood and the descriptive utility of the proposed signal-based measures across trials of varying difficulty. Stage 2 was a counterbalanced supplementary study in an independent sample of 18 younger adults. Reporting of the usability evaluation follows the American Psychological Association Journal Article Reporting Standards for quantitative research, as reflected in the Methods subsection headings below.

Both stages of this study were approved by the Research Ethics Committee of the Hino Campus, Tokyo Metropolitan University (R8-019). All participants in stages 1 and 2 provided written informed consent before participation; the stage 2 procedure (counterbalanced trial-order presentation and questionnaire administration) was covered under the same approved protocol. Before analysis, all recorded data were anonymized by the research team by replacing participant identifiers with numerical codes, and neither the analysis dataset nor this report contains any personally identifiable information. This study was conducted in accordance with the Declaration of Helsinki. Older-adult participants in stage 1 each received JP ¥6000 (approximately US $39, based on an exchange rate of approximately JP ¥153 per US dollar during the October-November 2025 data-collection period) as compensation; younger-adult participants in stage 1 and all stage 2 participants received no financial compensation. All images and videos in this paper and supplementary materials were reviewed to ensure that no study participant can be identified. The individual shown in Figure 1 and in Multimedia Appendices 1 and 2 is an author of this study, who provided written informed consent for the publication of their identifiable image and videos; no other identifiable individual appears in this paper or supplementary materials.

Balance Wood is a bilateral system for the tilt-angle signal analysis and continuous signal recording of ankle control during gameplay, converting coordinated foot-tilt input into control signals for an interactive serious game. The system consists of 2 Balance Wood units placed under the feet of the participant, a host computer running the game, and a visual feedback implementation (Figure 1). In the standard configuration, the feedback is presented via a screen-based game. This design follows previous balance-oriented exergame work [11,12,16].

A low-cost alternative based on an LED matrix, operating without a dedicated host computer or display screen, was also developed [29-33]; it was not used in the usability evaluation reported here. Video demonstrations of the screen-based and low-cost LED versions in use are provided in Multimedia Appendices 1 and 2, respectively.

Each Balance Wood unit comprises a tilting wooden device with an embedded M5StickC PLUS (M5Stack Technology Co, Ltd) module mounted at the center. This embedded module simultaneously serves 2 functions: angle sensing and wireless communication. Specifically, each unit measures the respective tilt state and transmits the signal via Wi-Fi. The host system receives the signals from the 2 Balance Wood units through User Datagram Protocol communication. The transmitted value is the roll component of the attitude estimate produced by the module’s on-board sensor-fusion routine, which corresponds to the sagittal-plane tilt of the platform. Denoting the left- and right-unit tilt angles at time t as θL(t) and θR(t), the host computes the combined control angle as their signed sum:

No other component angle enters the calculation, and no filtering, weighting, or dead band is applied. θc(t) is the variable that drives the game and, after the display scaling is reversed, is the variable used for all analyses reported here.

The attitude estimate is derived from the accelerometer and gyroscope signals of the module’s embedded inertial measurement unit. As low-cost micro-electro-mechanical-systems inertial sensors carry a fixed mounting and 0-point offset, an initial alignment (zeroing) was performed with the device flat before each session, so that all recorded signals were referenced to 0° [34]. Signals were sampled at approximately 9.5 Hz. Full hardware and signal-acquisition specifications are provided in Multimedia Appendix 3.

As the control variable is the signed sum of the 2 units, same-direction tilts from the 2 feet add, whereas opposite-direction tilts partially cancel; the interface therefore encourages coordinated bilateral movement. This simplified bilateral design lowers the difficulty of understanding how to operate the game [35-37].

In the recorded tilt-angle signal, forward tilt is represented by negative values and backward tilt by positive values; all main outcomes are described in terms of the forward and backward directions. All recorded signals correspond to the physical inclination of the foot device. For display purposes only, the host system applies a fixed 3/5 scaling factor to the on-screen bucket, so displayed and physical angles are not interchangeable: a displayed angle of 11° corresponds to approximately 18.33° of physical device tilt (Multimedia Appendix 3). The recorded data file stores the display-scaled values; the factor was divided out before analysis, so unless otherwise stated, all angular values reported here are unscaled physical angles.

The recorded signal represents the ankle-controlled tilt angle of the Balance Wood device, generated as participants drove the device through ankle flexion-extension. This device-level angle reflects ankle-driven control behavior rather than a direct anatomical measurement of ankle joint range of motion; the relationship between the device tilt and the underlying ankle joint angle is mediated by the foot-device interface and was not separately calibrated in this study.

In the standard screen-based implementation, a bucket is displayed in the center of the game interface, with a water source on each side. Water flows randomly from either the left or the right source, and each flow lasts for a randomly selected integer duration between 7 and 12 seconds. Participants are required to tilt the bucket toward the active source to collect the incoming water. Forward tilt corresponds to tilting the bucket to collect water from the left source, whereas backward tilt corresponds to tilting the bucket to collect water from the right source. The interface was designed so that the movement goal could be understood from the visual task outcome rather than from verbal instruction alone. This follows previous exergame work emphasizing intuitive, movement-oriented feedback [13,15,18,38-42].

In both display implementations, successful directional control allows the bucket to collect water from the currently active source; the low-cost version retains the identical bilateral ankle-input structure and directional control principle, consistent with compact wearable and embedded sensing approaches in rehabilitation and functional evaluation [14,31-33,43]. If the bucket is not tilted sufficiently toward the active source, water cannot be collected, and movement that is excessively fast causes water to leak. These 2 visual outcomes let participants judge their own directional accuracy and speed regulation from the state of the game.

To elicit sustained, near-maximal excursions rather than passive low-effort movement, the system applies a direction-specific adaptive challenge-penalty rule, tracked independently for the forward and backward directions. This design logic follows broader principles of interactive systems, where a task must balance guidance and challenge to elicit a range of movements from the user [17,19,44-46]. The rule arms for a given direction only once the displayed bucket tilt reaches 11°, a playability threshold determined empirically during development as the lower bound at which the water-flow simulation both engages water collection and retains the collected water; it disarms when the movement returns close to the neutral position. Once armed, maintaining the movement at or above a fixed proportion of the current maximum peak avoids a penalty, whereas failing to do so returns the displayed bucket smoothly toward the midline; sustained performance at that criterion raises the peak threshold for the direction, subject to an observation window that reverts the increase if the criterion is not maintained. The penalty is applied to the displayed bucket angle only: the recorded signal always corresponds to the actual device tilt of the participant and never includes the penalty offset, and no score penalty or change to the water-flow rate is applied. The complete rule set is given in Multimedia Appendix 3, and the numeric value of every parameter is listed in Table S1 in Multimedia Appendix 3.

The performance of each participant was examined across 3 consecutive Balance Wood trials, and the resulting time-series signals were described using the proposed exploratory measures. The dataset comprised 40 participants: an older-adult group (n=20) and a younger-adult group (n=20). Including both age groups allowed the system to be exercised across a range of baseline movement capabilities.

The 3 formal trials differed inherently in difficulty, linked directly to the size of the falling water drops: lower difficulty meant larger drops and easier collection. Trial 1 was the easy condition; trial 2 was the medium condition, preset so that completing all water collection required close to 3 minutes, although faster participants could finish 10 to 20 seconds earlier; trial 3 was the high-difficulty condition, intentionally designed so that the task could not be fully completed even with maximal effort, to obtain a complete dataset of ankle behavior under maximal challenge. All 3 conditions ran under the same maximum duration of 180 seconds, and each trial ended at whichever came first: completion of the water-collection target for that condition or expiry of the 180-second limit. As completion requirements differed across trials, raw duration and holding time in seconds were not treated as direct evidence of performance; the analysis instead used the extracted signal-based measures.

Participants were healthy adults who were able to perform repeated ankle flexion-extension safely and to provide written informed consent. Individuals with any acute or chronic condition that could interfere with safe participation—including lower-limb injury or pain, neurological or vestibular disorders affecting balance, or any contraindication to light physical activity—were excluded. For the older-adult group, participants were community-dwelling older adults; individuals unable to understand the task instructions or to provide informed consent were also excluded.

The younger-adult groups (stages 1 and 2) were recruited by convenience sampling from the Tokyo Metropolitan University community (Hino Campus) and the associated laboratory community, whereas the stage 1 older-adult group was recruited from the general community with support from the JST (Japan Science and Technology Agency) Moonshot Research and Development program (Moonshot Goal 3; JPMJMS2034). Data were collected at the Hino Campus of Tokyo Metropolitan University. In stage 1, the older-adult group was assessed between October and November 2025, and the younger-adult group between February and March 2026; stage 2 was conducted in June 2026.

As a development and usability study, the sample sizes were chosen to be appropriate for early-stage usability evaluation rather than for confirmatory hypothesis testing. No formal a priori power analysis was conducted because the study is descriptive and exploratory in nature and is not designed to test inferential hypotheses; accordingly, the reported inferential statistics are interpreted as exploratory and are reported with CIs to convey precision. Stage 1 included 40 participants (20 participants per age group), and stage 2 included 18 younger adults, consistent with sample sizes commonly used in the usability and feasibility studies of serious games.

Only age and sex were collected; no other anthropometric or experience variables were recorded. Characteristics for all 3 groups are summarized in Table 1.

All sessions used an identical seated setup, standardized by target posture rather than by fixed equipment dimensions. Participants sat upright with the thighs approximately horizontal and both legs directed forward, so that hip flexion was approximately 90°, with the knees flexed to approximately 90°, the shanks hanging vertically, and each foot placed at the center of its Balance Wood unit; in this position, the platform was horizontal and was defined as 0° at session zeroing. Chair height was adjustable from 45 to 60 cm and was set to obtain this posture. Each unit was mounted on a movable wooden base board that permitted rotation in the sagittal plane only, mechanically preventing mediolateral tilt; with the boards in contact, the center-to-center distance between units was 13 cm, widened to approximately 22 cm, or at most 30 cm, for participants requiring a wider stance. All distances were measured directly. Participants wore their own shoes, which were required to be flat-soled, and the feet were not strapped or otherwise fixed to the platforms. Movement outside the ankle was not mechanically constrained; participants were instructed to maintain this seated posture and to control the device with the ankles.

Before the 3 formal trials, participants completed a low-difficulty practice session in which the instructor explained the game content, operating procedure, and precautions while observing performance in real time; formal testing began once the participant was confirmed to have understood the operation. Participants then completed the 3 trials consecutively in a fixed order (trial 1 to trial 2 to trial 3), and the system automatically recorded the tilt-angle signals of the device throughout testing [42,47].

A representative signal from trial 3 of a single participant is displayed (Figure 2), covering the full 180-second trial; each peak corresponds to 1 period of maintained forward or backward ankle movement. Some peaks show an amplitude decrease followed by a rapid rise, reflecting adjustment after the challenge-penalty mechanism was triggered. In many peaks, the latter half reached a greater amplitude than the former half, and peaks later in the trial tended to be larger than the initial ones. Such waveforms illustrate the signal features on which the proposed exploratory measures are computed [47].

The parameters used in this study fall into 2 categories. System parameters of the adaptive challenge-penalty mechanism—including the 11° displayed-angle minimum valid peak threshold (equivalent to approximately 18.33° physical tilt), the 0.3° threshold increment, the 95% maintenance criterion, the 7-second observation window, and the 4-second penalty cancellation window—were fixed during system development and constitute part of the experimental environment shared by all participants. These parameters shaped the data-generating process and cannot be retrospectively altered without rerunning the experiment. Offline analysis parameters—including the middle 80% peak retention window, the κ=0.8 coefficient for the directional holding threshold, and the second-half-vs-first-half comparison rule for effective flexion-extension classification—were chosen during signal processing and could in principle be varied to evaluate the robustness of derived measures. This distinction is referenced in the Limitations subsection of the Discussion section.

In this study, valid signal acquisition involved 2 complementary levels of processing. First, direction-specific excursions (peaks) were identified for peak-based analyses such as typical directional amplitude and effective flexion-extension proportion. Second, clearly artifactual or nonrepresentative segments were excluded from the continuous trial signal to obtain an effective analyzed signal for time-based outcomes such as directional holding proportion and overspeed proportion.

For peak-based analyses, a peak was defined as a direction-specific excursion bounded by return toward the neutral level, and analysis focused on the stable portion of each excursion: only the middle 80% of each peak was treated as the valid signal region, the first and last 10% being treated as transition periods between forward and backward movement. A symmetric 10% trim was used as a simple operational rule, retaining most of each excursion while reducing the influence of direction-reversal transitions at both ends.

Clearly artifactual or nonrepresentative segments were also excluded, such as isolated initial transitional segments or extreme abnormal segments whose waveform patterns differed markedly from surrounding peaks. Segment exclusion was performed using predefined rules and verified by visual inspection. This operated within each participant’s recorded data and did not exclude any participant; no participant-level data were missing.

All 58 enrolled participants (stage 1: 20 older and 20 younger adults; stage 2: 18 younger adults) completed all study procedures and were included in the analysis, with no exclusions or withdrawals (the flow of participants through the 3 groups is shown in Figure 3). In stage 2, all 18 participants provided complete questionnaire responses, with no missing data for the usability outcomes.

The descriptive statistics of the main outcomes in the older-adult group are summarized (Table 5). When all 3 trials were combined by pooling valid excursions across the 3 trials, the effective flexion-extension proportion was 0.838 in the forward direction and 0.742 in the backward direction.

The results of the Mauchly test are reported for all 5 outcomes (Table S2 in Multimedia Appendix 3). The sphericity assumption was violated only for the typical backward amplitude (W=0.5432, P=.0041, ϵGG=0.686), therefore, Greenhouse-Geisser corrected results were reported for that outcome. The repeated-measures ANOVA revealed nominally significant trial effects for the typical forward amplitude (F₂,₃₈=4.169, P=.023, ηp²=.180), and the typical backward amplitude (F₁.₃₇₃,₂₆.₀₈₄=4.338, P=.036, ηp²=.186). The forward holding proportion (F2,38=3.122, P=.056, ηp2=.141) and the backward holding proportion (F2,38=2.563, P=.090, ηp2=.119) did not reach nominal significance. The overspeed proportion remained stable across the 3 trials (F2,38=0.416, P=.663, ηp2=.021). Under the Bonferroni-corrected threshold of P<.01, neither amplitude effect reached formal significance, and both should be read as preliminary despite their large effect sizes. Post hoc paired comparisons with Bonferroni correction were conducted for the 2 outcomes demonstrating trial effects. For the typical forward amplitude, no pairwise comparison remained significant after correction, although increases from trial 1 to trials 2 and 3 were observed descriptively. By contrast, for the typical backward amplitude, the increase from trial 1 to trial 2 remained significant after Bonferroni correction (t19=3.518, P=.0023, Padj=.0069, dz=0.787). The increase from trial 1 to trial 3 was additionally retained as a meaningful descriptive change, although it did not remain significant after correction (Table S3 in Multimedia Appendix 3).

Shapiro-Wilk testing indicated approximately normal trial-level distributions apart from the typical backward amplitude in trial 3. A supplementary Friedman test was therefore conducted for that outcome as a sensitivity analysis. It was additionally significant (χ22=9.30, P=.0096, Kendall W=0.233; Table S4 in Multimedia Appendix 3) and did not change the interpretation of the repeated-measures ANOVA result.

Overall, typical amplitudes were higher in the later trials than in trial 1, the clearest pairwise change being the increase in typical backward amplitude from trial 1 to trial 2. The forward holding proportion increased gradually across trials, the backward holding proportion peaked in trial 2 before declining in trial 3, and the overspeed proportion remained stable. The effective flexion-extension proportion suggested a direction-dependent movement pattern rather than a uniform change across both directions.

The descriptive statistics of the main outcomes in the younger-adult group are summarized (Table 6). When all 3 trials were combined by pooling valid excursions across the 3 trials, the effective flexion-extension proportion was 0.779 in the forward direction and 0.787 in the backward direction.

The results of the Mauchly test are reported for all 5 outcomes (Table S5 in Multimedia Appendix 3). The sphericity assumption was violated only for the typical forward amplitude; therefore, Greenhouse-Geisser corrected results were reported for that outcome. The repeated-measures ANOVA showed no significant trial effect for any of the 5 main inferential outcomes in the younger-adult group, with effect sizes that were small to medium at most (partial ηp²=0.083 or below). Accordingly, no formal post hoc paired comparisons were performed for the younger-adult group.

Shapiro-Wilk testing indicated that most trial-level distributions in the younger-adult group were approximately normal, although deviations from normality were observed for the typical forward amplitude in all 3 trials and for the overspeed proportion in trial 3. Therefore, supplementary Friedman tests were conducted for these outcomes as sensitivity analyses. These tests were not significant for the typical forward amplitude (χ22=1.90, P=.3867, Kendall W=0.048) or the overspeed proportion (χ22=1.30, P=.5220, Kendall W=0.033; Table S4 in Multimedia Appendix 3) and thus did not change the overall interpretation of the repeated-measures ANOVA results. Overall, the younger-adult group showed minimal within-session trial-wise change: typical amplitudes, holding proportions, and the overspeed proportion all remained relatively stable across the 3 trials, and the forward and backward effective flexion-extension proportions were similar to one another.

This study developed Balance Wood, a bilateral ankle-controlled serious game, and evaluated its usability and the kinds of gameplay-derived signals it records in healthy adults. Across both stages, the system produced continuous tilt-angle signals from which the proposed exploratory measures could be computed.

In stage 1, the proposed measures could be computed for every participant, and each group is described separately. Within the older-adult group, typical directional amplitudes increased across the 3 trials; within the younger-adult group, the measures remained comparatively stable. The overspeed proportion varied little across trials in both groups. Taken separately, they illustrate how continuous signal tracking can characterize movement patterns beyond simple task completion.

In stage 2, under a fully counterbalanced administration in younger adults, the signal-derived measures showed no significant differences across difficulty conditions or presentation positions, and the device tilt angle agreed closely with a digital goniometer at the device level. On the structured questionnaire, perceived usability was good, safety concerns and discomfort were rated low, and participants reported using predominantly the ankle and adjacent structures rather than proximal segments.

Taken together, the 2 stages provide complementary evidence for the signal-derived framework. Stage 1 showed that the measures could be extracted from repeated gameplay and describe within-session patterns. Stage 2 extended the framework to an independent, counterbalanced younger-adult sample, in which all 7 measures were again derived and no statistically significant effects of difficulty or presentation position were detected. The methodological contribution of stage 2 is therefore that the framework remains applicable when difficulty and presentation position are separated by design. Accordingly, stage 1 patterns are best interpreted as descriptions of repeated gameplay rather than effects attributable specifically to task difficulty or order. Stage 2 findings remain exploratory because of the small, younger-adult-only sample and heterogeneous prior exposure.

Previous exergame studies and reviews have evaluated usability and acceptability across younger and older adult populations [10,13,20,42]. As usability depends on interaction design and use context, findings from other exergame configurations cannot be assumed to transfer directly to Balance Wood’s seated, bilateral, ankle-focused interface. In stage 2, the mean SUS score was 81.5, above the conventional benchmark of 68 [49] and descriptively higher than the score of 70.5 reported for a home-based exergame evaluated in a similarly aged healthy sample [10]. Participants also reported that visual feedback helped them understand the task and adjust their speed, rated safety concern and pain or discomfort low, and indicated predominant reliance on the ankle and adjacent structures. Ratings for enjoyment and willingness to use regularly were 3.78 and 3.61, respectively, on a 5-point scale. These comparisons remain descriptive given the small, younger-adult stage 2 sample and heterogeneous prior exposure to the system.

Controlled laboratory work has characterized bilateral ankle-dorsiflexion force control using customized force-measurement platforms and target-matching tasks [3]. Although informative, this approach was not designed as a serious game and relied on a laboratory-oriented setup. Exergame studies have also quantified movement using 3D motion capture [51] or force-platform center-of-pressure data [52]. Balance Wood instead integrates sensing, visual feedback, and continuous signal recording in a low-cost, seated serious-game interface, deriving direction-specific measures directly from its own bilateral tilt-control signal. Following further validation, this design may facilitate repeated recording in home or community settings.

Finally, prior work has used self-adaptive serious games to adjust task demands during motor learning [53], while related exergame research has examined therapeutic ankle outcomes [54]. In Balance Wood, however, adaptation served a different methodological purpose: rather than primarily targeting motor-learning or therapeutic outcomes, the challenge-penalty mechanism progressively updated the direction-specific peak criterion to encourage repeated, sustained excursions near each participant’s current peak, supporting the capture of high-amplitude movement for descriptive signal analysis.

Balance Wood provides a compact, relatively low-cost method for collecting continuous bilateral ankle-controlled gameplay signals in 1 session; stage 2 required approximately 30 minutes per participant and no laboratory-grade motion-capture equipment for device-level tilt recording. The platform can support standardized signal acquisition across repeated trials alongside end point outcomes. Favorable usability and feedback comprehension, with low discomfort and safety concerns, support further evaluation in applied settings. Following external validation, the system could be evaluated in community settings, small clinics, or home-based research, potentially lowering the logistical barriers to repeated recording of ankle-controlled behavior. At present, Balance Wood is a research and development platform rather than a clinical assessment or diagnostic instrument. Before independent or unsupervised deployment, future studies should examine biomechanical validity, between-session test-retest reliability, usability, and signal characteristics in older and clinical populations, footwear-related slippage, device fixation, and safety procedures.

First, the samples were small (20 per group in stage 1; 18 in stage 2), and several outcomes did not survive multiplicity correction; the results are therefore presented as exploratory and should be read through effect sizes and descriptive patterns rather than P values alone. Second, generalizability is limited by the single-site, healthy, and partly younger-adult samples (stage 2 younger adults only, predominantly male, recruited from the university), and between-session test-retest reliability was not assessed. Validation in older and clinical populations is planned as future work.

Third, the signal-derived measures are exploratory descriptors, not validated biomechanical constructs, and may be specific to the Balance Wood task. As the adaptive challenge-penalty mechanism was designed to shape gameplay, its influence on the recorded signals cannot be separated retrospectively; establishing the biomechanical meaning of the measures and systematically varying the mechanism’s parameters are left for future work. The near-limit holding threshold was anchored to each participant’s own trial 2 amplitude, so the resulting holding proportions should be read as relative to each participant’s medium-difficulty performance rather than against an external standard. Fourth, no concurrent validation against gold-standard anatomical measures (eg, goniometry) was performed, so the measures characterize device-level rather than anatomical ankle control; mapping device tilt to the anatomical ankle angle remains for future work.

This study introduced Balance Wood, a low-cost bilateral serious game that converts ankle-controlled device tilt into continuous gameplay signals, and reported its first development-and-usability evaluation in healthy adults. The system showed good perceived usability (above the established SUS benchmark), a positive and low-risk experience, close agreement between its device-level tilt readings and a digital goniometer, and predominantly ankle-based control. Its novelty lies not in gamification itself but in packaging a low-cost, bilateral, ankle-specific control interface with continuous signal capture into a single, easy-to-use device, which distinguishes it from existing systems that record continuous movement data and are typically laboratory-grade, costly, or oriented toward whole-body or upper-limb movement. Its principal contribution is a reproducible, usability-tested platform with characterized device-level accuracy and formally defined candidate measures. These measures require clinical or biomechanical validation before home- or community-based research; in the near term, the system remains a research tool.