The Effects of a VR Training Program for Walker Avoidance Skill Improvement: A Randomized Controlled Trial

This study aimed to evaluate the effectiveness of our newly developed virtual reality head-mounted display (VR-HMD) “walker avoidance” game in reducing step-aside reaction time (SART) and enhancing agility in collision avoidance. Fifteen young adults in experimental group (EG) engaged in the “walker avoidance” game, while another 15 young adults in the control group (CG) played the “first touch” tutorial. The results showed the EG had significant decreases (p < 0.01) in both SART-standing and SART-walking when compared with pre-intervention measurements. Compared with the CG, the EG SART-standing exhibited significant decreases in both the first (p = 0.001) and second (p < 0.001) measurements post-intervention; the EG SART-walking demonstrated significant decreases in all (p < 0.05) measurements, except for pre-intervention measurement. One-dimensional statistical parametric mapping (spm1d) also demonstrated significant differences in most of the electromyography and forefoot/hindfoot ground reaction force results because the step-aside movement became quicker in the EG following training. After pushing the leg-heel contact, the EG participants made a toe-off sooner than the CG participants. Following two sessions of our newly developed “walker avoidance” game, conducted 1 week apart, the EG exhibited less collisions with virtual pedestrians and reduced reaction times to unpredictable directional change measurements compared with the CG. This study demonstrated the effectiveness of this targeted VR training program in improving motor function, which introduced a novel approach to rehabilitation as a digital therapy. It offers innovative perspectives and an approach for clinical rehabilitation, while also providing new ideas for the VR content development industry.

T HE step-aside movement is a common maneuver that involves shifting the body's center of mass by taking steps to change the direction of walking to evade obstacles or pedestrians.Our previous study found that individuals with decreased electromyography (EMG) ankle muscle activity are at a higher risk of stumbling during step-aside movement, as ankle muscle contractions in both the pushing and loading legs of the step-aside movement exceed those during normal walking [1].Robinovitch et al. reported that incorrect weight-shifting and stumbling were the two most frequent causes of falling [2].Because step-aside movements require substantial weight shifting, and their execution in an untrained and sudden manner to avoid danger tends to be unfamiliar and hasty, potentially causing incidents of stumbling, collisions, and falls.
Delgado and Der Ananian highlighted the potential of virtual exergaming (exercise gaming) in promoting physical activity and improving balance, posture, gait, and overall health in older adults [3].Huygelier et al. have reported that the beneficial effects of virtual reality applications, including virtual reality head-mounted displays (VR-HMD), on the health of older adults are not constrained by negative attitudes or cybersickness [4].Using a VR-HMD allows participants to immerse themselves in challenging or educational scenarios and enables them to practice new skills repeatedly in a corrective and non-threatening environment.Moreover, the enhanced immersion provided by new VR technologies is well suited for active learning or simulation-based learning [5].However, current evidence on the use of VR-HMD devices for rehabilitation training to enhance obstacle avoidance and reduce fall risk remains insufficient.Most related studies employed devices other than VR-HMD [6], [7], which limited the participants' immersive experience and potentially diminished training effectiveness.In studies that used VR-HMD devices, the VR software or games employed were not specifically designed for targeted behavioral training or were only used in combination with other interventions [8], [9].Nevertheless, these studies have consistently demonstrated significant functional improvements in various populations following rehabilitation training using VR devices.
Building on this concept and the aforementioned considerations, we developed a VR-HMD "walker avoidance" game designed to be played in the frontal plane from a standing position.This game, however, aims not only to reduce the step-aside reaction time (SART) in the standing position but also during walking, and additionally, to enhance agility in collision avoidance.In 1987, Schmidt and Young proposed that the transfer of learning is contingent upon the degree of "similarity" between two tasks, with tasks sharing commonalities in relative timing and sequencing tending to exhibit positive transfer effects [10], [11].We believe that virtual environments can replicate perceptual demands, encompassing the visual, auditory, and haptic aspects associated with a given task.Therefore, we developed both the "walker avoidance" game and the SART measurements in alignment with the simulation training method, and motor transfer principles outlined by Schmidt et al. in the book "Motor Control and Learning" [10].
The primary objective of this study was to assess the impact of the "walker avoidance" game on reducing unpredictable SART; furthermore, we aimed to determine whether the training effects of the game conducted solely in the frontal plane could effectively transfer to step-aside movements during walking.This study also sought to examine the associated changes in EMG, and ground reaction force (GRF).
On that ground, we hypothesized that: 1.After the training, SART in both standing and walking would be shorter than before the training, and the experimental group (EG) would also exhibit shorter SART compared to the control group (CG) in both conditions.
2. The reduction in SART resulting from the "walker avoidance" game would persist for up to one week.
3. The training effect will have cumulative effects.

A. Participants
Thirty young, healthy participants (15 males and 15 females) were recruited for this study (Table I).Male participants had foot sizes ranging from 260 to 270 mm, whereas female participants had foot sizes ranging from 235 to 250 mm as dictated by the dimensions of the foot insole sensors.Individuals with balance issues, pes planus, chronic ankle instability, or prior major lower-extremity problems were excluded.Before each test session, all participants provided written informed consent.The randomized controlled trial study was approved by Yonsei University Mirae Institutional Review Board (approval no.1041849-202305-BM-081-02) and registered at the International Clinical Trials Registry Platform (KCT0009305).The study was conducted in accordance with the principles of the Declaration of Helsinki.

B. Experimental Setup
This study is designed as a randomized controlled trial employing simple randomization with a blocking strategy.After selecting the participants, they were assigned to either the EG or the CG using the random function in Microsoft Excel.Throughout the study, participants were kept blinded to their group allocation and the existence of the other group.
Before the first experiment, all participants were asked to walk at a normal speed to obtain EMG reference data.Subsequently, SART measurements (unpredictable directional change) during standing and walking were conducted to collect SART-standing and SART-walking data; EMG data of the tibialis anterior (TA), peroneus longus (PL), and soleus (SOL) muscles; and GRF data.These datasets were designated as pre-A data.Thereafter, EG participants engaged in the "walker avoidance" game, while those in the CG played the "first touch" game (Meta Oculus Quest 2 user tutorial).Subsequently, both groups underwent another round of SART measurements in standing and walking positions, and the same outcome data were collected and identified as post-A data.This marked the end of the first trial.One week later, the participants repeated the same procedure, but the data collected during this session were labeled pre-B and post-B data (Figure 1).
1) Walker Avoidance Game: "Walker avoidance" is a first-person obstacle avoidance game developed using Unity (version: 2020.3.28f1).In this study, the Meta Oculus Quest 2 was employed as the platform for the game; however, the game is also compatible with all other VR-HMDs apart from the Oculus Quest 2. The character and building models in the game were sourced from the Cube People -Demo asset (author: Fatty War) and Simple City pack plain asset (author: 255-pixel studios), respectively.Available on assetstore.unity.com.
Upon entering the game, players are positioned in the middle of a city on a pedestrian pathway that is divided into left and right sides by a white line.Apart from this pathway, players could observe various commercial buildings, streetlights, and pedestrians standing around them.Above the pathway, there are the words "Pass: 0" and "Hit: 0," which keep track of the number of successfully avoided and collided pedestrians, respectively, during gameplay.
Once the game began, the pedestrians started appearing intermittently in front of the player at a distance of 10 m, walking towards the player at a speed of 4 m/s, and then disappearing 6 m behind the player.As pedestrians may alter their paths midway, players must be prepared to avoid collisions by using step-aside movement.
The pedestrian dimensions were set as follows: width (x) = 1.1 m, height (y) = 1.5 m, depth (z) with backpack thickness = 1.5 m.The collision dimensions for the pedestrians were set as follows: width (x) = 1 m; height (y) = 1.5 m, and depth (z) = 1 m.Whenever a player successfully avoids a pedestrian, a crisp bell sound accompanies the action, and the number after the "Pass" label increments by one.Conversely, if a player collides with a pedestrian, the screen momentarily turns red, accompanied by a collision sound and controller vibration, and the number after the "Hit" label increments by one.
At any given time, only one pedestrian will appear, and the next pedestrian is typically set to appear after a random duration of 1 to 2 seconds.Pedestrians initially appear on either the left or right side of the appear point and may change direction at change point A, located 8 m in front of the player, or change point B, located 6 m in front of the player, or they may continue moving in the same direction (Figure 2A).This reduced the player's ability to predict the pedestrian's trajectory.Therefore, there are six possible walking trajectories for pedestrians: 1. from right to right; 2. from the right to the left side, with a leftward movement at point A; 3. from the right to the left side, with a leftward movement at point B; 4. from left to left; 5. from the left to the right side, with rightward movement at points A, and 6.From the left to the right side, with a rightward movement at point B. After the game started, the initial six occurrences were followed in order, after that, each trajectory appeared 20 times until the end of the game.The timing and paths of pedestrian appearances was determined using a random function in Microsoft Excel and remained unchanged after the start of the experiment to ensure a consistent game scenario for all participants.Hence, the participants were required to avoid 126 pedestrians in one game.The number of successfully avoided pedestrians was recorded for statistical analysis.See Video 1 for more detail.
The game has an invention patent pending with the Korean Patent and Trademark Office (2023.11.12).
2) SART Measuring Equipment Design: First, two sets of pressure sensor groups, each consisting of two pressure sensors connected in series, were connected to a 3-5V power supply.They were then connected in series to an arrow indicator light (using a combination of LED light strips) and to the Noraxon analog input, enabling a wireless connection to the Noraxon Ultium host.The SART measurement device was 3) SART Measurement Method: The SART measurement method was divided into standing and walking measurements.
To make the positions of the pressure sensors unpredictable and ensure that the participants performed the step-aside movement on the same material, we placed six identical yoga mats in a 2 × 3 grid pattern on the floor, with the mat containing the pressure sensors placed in the middle (second column).To decrease the learning effect and prevent participants from predicting the timing of LED light illumination, after each SART-walking measurement, we relocated the yoga mat with a pressure sensor (either at the back of the first row and second column or at the front of the second row and second column).After all measurements were completed, the participants expressed that they did not know the principles and timing of LED light illumination.
4) SART-Standing Measurement Procedure: Participants were instructed to stand in a stationary position on a yoga mat positioned between the pressure sensor and arrow indicator light.The feet were kept shoulder-width apart, and two stickers were affixed to mark the starting position in front of both toes.Another sticker was placed to the left of the front sticker at a distance equal to one shoulder width, representing the left endpoint.The same approach is applied on the right side to mark the endpoint.The participants were instructed to execute a step-aside movement on both sides while ensuring that the foot landing point was in proximity to the respective endpoint marker.After three practice trials on each side, participants returned to the starting position and focused on the arrow indicator light in front of them.At this point, one researcher positioned themselves behind the participant and unpredictably activated either pressure sensor, leading to the illumination of the arrow indicator light in either direction.Upon observing the illumination of the arrow indicator light in any direction, the participants were required to promptly execute a step-aside movement, ensuring that the foot landed near the corresponding endpoint (Figure 3 D-G).The time interval between light illumination and foot landing was recorded as the SART.Only data from the first successful attempt were considered for subsequent statistical analyses.
5) SART-Walking Measurement Procedure: Masking tape to create 1-meter-long reference lines parallel to the edges of the mat at both the left and right endpoints.The participants began walking 3 m away from the yoga mats and were directed to walk naturally toward the central yoga mat.For instance, when passing over the mat, the left foot triggers the pressure sensor beneath the left-side mat, illuminating the arrow indicator light on the right side.In this scenario, participants were required to execute a step-aside movement in the direction indicated by the arrow (right) while ensuring that the right foot landed near the right reference line and subsequently advanced in that direction, and vice versa (Figure 3 A-C).The time interval between the illumination of the arrow indicator light (3V voltage input time) and the moment the foot on the illuminated side touched the ground was measured as the SART.To prevent slippage, yoga mats were securely affixed to the ground using double-sided tape.
Before the measurement (prior to equipping the EMG sensors and insoles), the participants were provided with verbal instructions and on-site demonstrations to learn the SART-walking measurement method.After sufficient understanding, participants underwent up to five practice trials under guidance, and only upon successful completion of one trial did they proceed to the formal experimental measurement.If there were more than 10 consecutive failures, the participants were informed that they could not continue with the experiment because of the significant learning effects.After completing the practice session, the participants were instructed to perform two level walking trials to obtain the average data for EMG normalization.Once these steps were completed, the participants were considered to have completed all familiarization procedures and could proceed to the formal experimental measurement.

C. Data Collection
1) SART: SART-standing and SART-walking data were collected and extracted using the MR3.18®software.
2) Electromyography: Noraxon Ultium wireless surface EMG sensors were used to record the bilateral activities of the TA, PL, and SOL muscles.As recommended by SENIAM [12], the recording electrodes were placed 20 mm apart from the muscle fibers to be measured, and the ground electrode was located underneath the wireless sensors.The skin was prepared by shaving and cleaning with an alcohol swab before electrode attachment.The raw signals were recorded and processed using MR3.18®software.EMG data were recorded at a sampling frequency of 2000 Hz (2000 data points per second) and processed using the root mean square with a 200 ms epoch size.For amplitude normalization, the processed EMG data were divided by their average values during normal walking, and the normalized EMG data were used for statistical analysis.
3) Ground Reaction Force: Forefoot ground reaction force (FGRF), heel ground reaction force (HGRF) data were collected using knitted shoes with a Noraxon Ultium Insole.The Noraxon Ultium Insole utilizes eight force-sensitive registers positioned in the major weight-loading foot areas of the sole, and the collected sensor data are processed in real-time using interpolation techniques to approximate (hysteresis < 5%, error < 3%) the pressure distribution across the entire foot.The FGRF and HGRF data were automatically normalized to the body weight of each participant using a Noraxon insole.

D. Statistical Analysis
The reliability of the SART measurements were assessed by calculating the intraclass correlation coefficient (ICC).For this analysis, the SART pre-A data from the pilot test, which included 12 participants (EG = 6), were extracted to determine the ICC.
Two-way repeated-measures analysis of variance (ANOVA) power analysis using G-Power software was performed to compute the minimum sample size requirement.The minimal sample size was determined to be 24 (experimental group: 12, control group: 12) according to the effect size (f = 0.40) and power (1 -β = 0.95) on the difference in SART-walking between the experimental and control groups using pilot data.
All reaction time data and collision number were analyzed separately using the Shapiro-Wilk test to test the assumption of a normal distribution.A one-way repeated-measures ANOVA was performed to compare the SART differences Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
between pre-A, post-A, pre-B, and post-B in all participants.Bonferroni's post-hoc test was applied to account for type I errors.An independent samples t-test was used to compare the SART differences between the EG and CG at all time points.A Wilcoxon signed-rank test was conducted to compare the differences in the number of collisions with virtual pedestrians in the "walker avoidance" game between the first and second trials.SPSS was used for the statistical analysis of the SART data and collision numbers, with statistical significance set at p < 0.05.
As in our previous publications, the differences in EMG muscle activity, and GRF during the step-aside movement in the walking condition were statistically examined across the entire step-aside movement using one-dimensional statistical parametric mapping (SPM1d) in Python, using the open-source software package SPM1d [1], [13], [14].SPM one-way and two-way repeated-measures ANOVA and SPM post hoc twosample t-tests were used for statistical analysis.

E. Region of Interest
We established a region of interest (ROI), that represented a specific time interval within a one-dimensional measurement domain.In all the experiments, the earliest point in time when the LED light was activated was designated as the starting point of the pushing phase ROI, corresponding to data point 30 (0.015s after pushing leg heel contact).The latest toe-off time (data point 2300, 1.15s after heel contact) among all participants was marked as the endpoint of the pushing leg ROI.Additionally, in all experiments, the earliest moment when the loading leg made heel contact (at data point 1100, 0.55s after heel contact) was identified as the starting point of the loading phase ROI.Because all the CG participants completed the step-aside movement later than the EG participants, the intersection of the CG loading leg HGRF and CG loading leg FGRF mean value trajectories (at data point 2366) was defined as the endpoint of the loading phase, occurring approximately 1.20 seconds after heel contact of the pushing leg.Consequently, data collected from the pushing leg heel contact up to 1.20 seconds later were extracted for statistical analysis, totaling 2400 data points.Therefore, the data of all participants from the pushing leg heel contact (0.00s) for up to 1.20 s were extracted.Subsequently, different ROIs were applied for the SPM statistical analysis of the push and loading phases.
All thirty participants completed the experiment.Oneway repeated-measures ANOVA Bonferroni's post hoc test (Figure 4C) indicated that within-group comparisons in the EG showed significant decreases (p < 0.01) in both SART-standing and SART-walking when compared to the pre-A measurement.However, the post hoc test also found significant increases (p < 0.01) compared post-A with pre-B in both SART-standing and SART-walking.All statistical methods also tested the within-group data for CG similarly but did not find any significant differences.Furthermore, when conducting between-group comparisons, the EG SART-standing exhibited significant decreases in both post-A (p < 0.001) and post-B (p < 0.001) measurements.Similarly, the EG SART-walking demonstrated significant decreases in post-A (p < 0.001), pre-B (p = 0.013), and post-B (p < 0.001) measurements compared to the CG.Furthermore, the number of collisions with virtual pedestrians in the "walker avoidance" game decreased by 37.08% during the second training session, and the one-tailed Wilcoxon signed-rank test also showed a statistically significant decrease (p < 0.05).
In the pushing leg of the step-aside movement, SPM twoway repeated-measures ANOVA found major regions with statistically significant differences in the group effect of PL (0.78-1.00 s), SOL (0.76-0.94 s), and FGRF (0.58-0.70 s, 0.82-0.98s).The SPM post hoc two-sample t-test showed that after heel contact of the pushing leg, the SOL and FGRF results of the EG participants exhibited earlier peak values and toe-off sooner (Figures 5, 6).The SPM{t} results indicating significant decreases compared to that of the CG (Figures 5, 6), which signified earlier completion of the push phase in the EG.The starting point of the major region was 0.89 s in post-A (Figure 5A), and by post-B, it had advanced to 0.77 s (Figure 5C) after pushing leg heel contact.The SPM post hoc two-sample t-test results for the FGRF also indicated significant decreases compared with the CG (Figures 6).The starting point of these major regions was 0.88 s in post-A (Figure 6A), advancing to 0.83 s in post-B (Figure 6C).Additionally, we also found a significant increased major region between 0.61 s and 0.67 s in post-B (Figure 6C).
In the loading leg of the step-aside movement, two-way repeated-measures ANOVA revealed significant differences in TA (0.56-0.65 s, 0.90-1.00s), PL (0.57-0.70 s, 0.92-1.07s), HGRF (0.77-0.85 s), and FGRF (0.75-0.86 s) between the EG and CG.The SPM post hoc two-sample t-test analyzed post-B data and showed significant decreases during 0.89-1.05s in TA, 0.93-1.04s in PL, and 1.01-1.18s in HGRF,  significant increases during 0.58-0.67s in PL and 0.72-0.87s in HGRF compared to that of the CG (Figure 7).The average HGRF trajectories provided a more detailed representation of the differences in heel contact timing (Figure 7C).Additionally, the TA and PL muscle activation decreased (Figures 7A, B).

IV. DISCUSSION
The SART results demonstrated that, following training with the VR-HMD "walker avoidance" game, participants exhibited significantly quicker responses during unpredictable step-aside movements.Within-group comparisons in the EG revealed significant decreases (p < 0.01) in both SART-standing and SART-walking from pre-A measurements.Betweengroup comparisons showed that the EG exhibited significant decreases in SART-standing post-A and post-B (p < 0.001) measurements and in SART-walking in post-A (p < 0.001), pre-B (p = 0.013), and post-B (p < 0.001) measurements compared with the CG.Thus, our first hypothesis was verified (Figure 4C).
Consistent with our second hypothesis, the statistically significant decreases observed in pre-B compared with pre-A under both conditions, as well as in the between-group comparison during SART-walking at pre-B, indicate that the training effects were maintained and retained over the ensuing week.However, did not identify a significant difference in the between-group comparison for the pre-B SART-standing condition.Additionally, we discovered significant increases (p < 0.01) when comparing post-A with pre-B in both the standing and walking conditions.Thus, SART significantly rebounded after 1 week, although it remained lower than that at the pre-training level.When participants underwent another training session at this point, the SART reached a new low (post-B) [15].Thus, the effects of VR training diminished rapidly within a week, although these effects can accumulate to some extent.This finding is consistent with our third hypothesis, which can be attributed to the fact that learning and consolidation constitute continuous processes that are mediated by factors such as sleep.The intermittent repetition of learning over several days facilitated the solidification and deepening of skills, and the skills tended to be better preserved in this way [16].
Furthermore, after two training sessions, the average SART-standing and SART-walking values for the EG decreased by 0.28 s and 0.18 s, respectively, compared with those of the CG participants (Figure 4A, B).The SPM results confirmed these statistically significant differences (Figures 5 and 6). Figure 5 illustrates the progressively earlier SOL muscle activation in the EG during pushing leg heel contact to achieve push-off as the number of training sessions increased, whereas the CG maintained a consistent pattern.A similar trend was observed for the FGRF and other dependent variables (Figure 6).The SPM results provided evidence that all the statistically significant differences noted above resulted from the SART becoming shorter and the step-aside movement becoming quicker in the EG following VR training.One significant increase in the major region of the FGRF of the EG was observed between 0.61 s and 0.67 s in post-B compared with that of the CG (Figure 6C), which was not statistically different between post-A and pre-B.This finding demonstrates that after two sessions of VR training, when the FGRF of the CG participants increased, the majority of the EG participants had already reached their peak FGRF, enabling them to push off the leg more effectively.A similar trend was observed in the loading phase for the PL EMG amplitude and HGRF results.This indicated that after two sessions of VR training, the EG participants experienced earlier loading leg heel contact, allowing them to complete the step-aside movement sooner than the CG participants.Moreover, as shown in Figure 7A and B, the mean and standard deviation of the SOL and PL EMG amplitudes of the EG were lower than those of the CG.After training, the loading phase was sufficiently stabilized, eliminating the need for additional forceful contractions of the ankle muscles to maintain mediolateral ankle stability.Most participants reported muscle soreness in the bilateral PL, akin to the discomfort they experienced after resistance strength training.This may be attributed to frequent contractions of the PL muscles.Therefore, the observed decrease in PL and SOL muscle activation during loading leg heel contact may have resulted from the participants discovering and implementing a more efficient step-aside strategy after numerous repetitions of unpredictable step-aside movements.
The reduction in SART can be attributed to the impact of the "walker avoidance" game, aligning with the findings of Schmidt et al., who have revealed that tasks with similar timing and sequencing tend to exhibit positive transfer effects [10].The participants were involved in a scenario in which an arrow-shaped LED light was illuminated, followed by the enactment of a direction change using step-aside movements.This scenario was simulated by avoiding approaching virtual pedestrians who could change direction unpredictably.Similar to the findings of Levin et al., the participants in this study completed 125 training repetitions, which enhanced their avoidance capabilities [17].Simultaneously, the repeated utilization and activation of mediolateral ankle strategies may have encouraged participants to employ step-aside movements at any given moment.Initial encounters with a game often result in suboptimal performance.However, with increased instances of successful avoidance or collision, participants typically developed proficiency.Collisions with virtual pedestrians were relatively common during the early stages of the game.However, as the game progressed, the probability of successfully avoiding collisions increased, resulting in a statistically significant decrease (37.08%) in the number observed 1 week later.In the later stages of the game, the participants could determine the positions where a collision would or would not occur by sensing the distance between themselves and the midline and sideline.This perception was crucial because the scenes and models in the virtual environment were presented on a 1:1 scale with the real environment.Consequently, the sense of distance acquired in the game can be directly applied to real life.The participants developed an understanding, such as, "If I take a step this large, I will (or will not) collide with pedestrians."After the participants took a step, the midline and sideline positions provided visual feedback.Repeated successful avoidance (or collisions) informed the participants whether the step they intended to take was in a safe (or unsafe) position.Through repeated attempts and learning, the participants determined the safest way to avoid obstacles (step size, strength, and timing).This avoidance strategy becomes part of the muscle memory, allowing it to be instinctively applied in similar real-life scenarios, even benefiting from reflexive avoidance actions [18].
The "walker avoidance" game only required participants to employ step-aside movements while in a standing posture to avoid oncoming virtual pedestrians.The participants had the freedom to attempt each avoidance maneuver using their preferred methods, even during a collision.As the 125 virtual pedestrians approached each other, the participants prioritized skill development over precision and final performance.This approach introduces a relatively high level of motor variability, facilitating expedited learning of the most effective avoidance techniques during exploration [19].Furthermore, auditory feedback, characterized by a distinct and crisp beep following successful avoidance, not only serves as a motivational stimulus for participants but also plays a significant role in reinforcing the learning process [20], [21].
In addition, after the two training sessions, both pushing and loading legs exhibited enhanced agility during unpredictable step-aside movements while walking.Compared with the control and pre-training condition of pushing and loading legs, the efficiency of movement execution improved to a certain extent after VR training.The most notable outcome was observed in the EG participants, where motor learning resulted in the accomplishment of the same push and loading movements with relatively smaller SOL muscle activities [22], [23].By contrast, this phenomenon was not evident in the CG participants.
Previous studies have indicated that the efficacy of VR training in skill learning, retention, and transferability largely depends on the demands of the games involved.Groups with specific needs tended to benefit only when they were provided with tailored rehabilitation games [24].This study demonstrated the effectiveness of shortening SARTstanding/SART-walking and enhancing the stability, efficiency, and agility of step-aside movements through utilizing a newly developed "walker avoidance" game.This study introduces a novel, efficient, and convenient digital therapeutic rehabilitation approach for knee and ankle rehabilitation groups without balance issues or cybersickness.These include individuals recovering from total knee/ankle replacement, ankle sprain injury, anterior/posterior cruciate ligament injury, and lower extremity fractures.
However, owing to the differences among patients, including the elderly population and young participants, the results of this study may not be generalizable to other demographic groups.Additionally, this experiment could not fully replicate all real-life scenarios; hence, the findings may not be universally applicable to other obstacle avoidance scenarios that differ from the "walker avoidance" game.

V. CONCLUSION
In conclusion, our newly developed "walker avoidance" game has effectively reduced the reaction time in young adults when faced with unpredictable directional changes.This improvement could be achieved by simply wearing a VR-HMD device and playing the game for less than five minutes in a space no larger than 5 m 2 .These enhancements were not only limited to the standing position but could also be efficiently transferred to the walking position, improving step-aside obstacle avoidance abilities.Furthermore, the participants exhibited increased agility in the pushing and loading legs after two training sessions spaced one week apart, which was facilitated by motor learning.This study demonstrated the effectiveness of this targeted VR training program in improving motor function, which introduced a novel digital therapy approach to rehabilitation.It offered innovative perspectives and an approach for clinical rehabilitation while also providing new ideas for the VR content development industry.

The
Effects of a VR Training Program for Walker Avoidance Skill Improvement: A Randomized Controlled Trial I. INTRODUCTION

Fig. 1 .
Fig. 1.Randomized controlled trial flow diagram for the experimental group (EG) and control group (CG).SART-standing, step-aside reaction time in the standing position; SART-walking, step-aside reaction time during walking; VR, virtual reality.

Fig. 2 .
Fig. 2. "Walker avoidance" game setup.(A) Top view of the pedestrian pathway and event points in Unity 3D.(B, C, D) Pedestrian appearance and direction change process in the game.(E) Picture of a participant playing "walker avoidance" game with the Oculus Quest headset in VR condition (F) player-pedestrian collision scene in the game.

Fig. 5 .
Fig. 5. Mean and standard deviation of pushing leg soleus (SOL) electromyography (EMG) amplitudes between the experimental group (EG) and control group (CG).The gray shading indicates the region of interest.The yellow shading depicts the results of the statistical parametric mapping (SPM) post-hoc two-sample t-tests, indicating statistically significant major regions during (A) 0.89-1.07s, (B) 0.78-0.94s, and (C) 0.77-0.98s after pushing leg heel contact.p < 0.001 ( * ) indicates significant differences; EG/CG Pre/Post SOL-A/B, EG/CG SOL EMG data that were measured pre/post-training in session A/B.

Fig. 6 .
Fig. 6.Mean and standard deviation of pushing leg forefoot ground reaction force (FGRF).(A) The yellow shading depicts the results of the statistical parametric mapping (SPM) one-way repeated-measures analysis of variance, indicating statistically significant major regions during 0.88-1.05s. (B, C, D) The gray shading indicates the region of interest.The yellow shading depicts the results of the SPM post-hoc twosample t-tests, indicating statistically significant major regions during (B) 0.88-1.06s, (C) 0.83-0.90s, (D) 0.61-0.67s, and 0.83-1.02s after pushing leg heel contact.p < 0.001 ( * ) indicates significant differences.