VR-skateboarding is a training approach that combines VR and treadmill technology. It involved using a skateboard that was integrated with a split-belt treadmill (QQ-mill, Motekforce Link, Netherlands), as shown in Figure 1; Supplementary file S1. The skateboard was placed on the stationary belt of the treadmill, while the moving belt was set to a comfortable walking speed for the participant. Comfortable walking speed was measured using a 10-m walk test, which is the most common and reliable test (Bohannon, 1997; Bohannon and Williams Andrews, 2011; Cheng et al., 2020b). The leg that was placed on the skateboard was referred to as the “supporting leg,” while the leg that slide on the moving belt was referred to as the “moving leg.” For safety purposes, the skateboard wheels were fixed statically on the stationary belt of the treadmill. Handrails were also available at waist level for support during VR-skateboarding, if needed.

A virtual scenario for VR-skateboarding was created using Unity3D software (version 5.3.2, San Francisco, United States) and displayed using virtual reality head-mounted displays (HTC VIVE, HTC Corporation, New Taipei City, Taiwan). The virtual scenario depicted skateboarding on a city road, as shown in Figure 2. Three wireless inertial measurement unit sensors (HTC VIVE trackers, HTC Corporation, New Taipei City, Taiwan) were used in VR-skateboarding as follows: 1) two trackers were placed on the participant’s legs to track leg movements. The speed and distance travelled in the virtual scenario were adjusted based on the movement of the tracker on the moving leg; 2) one tracker was attached in front of the skateboard and used to control the left and right direction of skateboarding to avoid obstacles in the virtual scenario. The cumulative distance travelled was provided as real-time virtual feedback and a final score to motivate participants during VR-skateboarding.

The eligibility of participants was assessed based on inclusion and exclusion criteria. The inclusion criteria included being between the ages of 20 and 40 years and not having any symptoms such as leg pain or numbness. The exclusion criteria included having had previous surgery and having neurological disorders such as stroke, lumbar radiculopathy, or spinal cord injury.

For walking, participants were asked to walk on the split-belt treadmill at a comfortable walking speed for 1 min × 5 times. Then, participants were asked to perform VR-skateboarding using their non-dominant leg as the supporting leg on the skateboard. Participants were instructed to skate with their dominant leg as the moving leg at a comfortable walking speed for 1 min × 5 times. According to previous studies, treadmill speed has the potential to impact biomechanical characteristics during movement (Möckel et al., 2003; Matjačić et al., 2019). In order to eliminate confounding factors from treadmill speed, the treadmill was set to the same speed (i.e., comfortable walking speed) for both VR-skateboarding and walking.

A 3-dimensional motion capture system (VICON ver. 2.5, Oxford Metrics Ltd., Oxford, United Kingdom) with ten infrared cameras (VICON Bonita, Oxford Metrics, United Kingdom) was used to collect joint kinematic data at a sampling rate of 120 Hz. The system used 45 spherical retro-reflective markers (14 mm) placed over anatomical landmarks based on the Plug-In-Gait model (Cheng et al., 202b).

Surface electromyography (EMG) (TrignoTM, Delsys Inc., Boston, MA, United States) was used to collect muscle activity data (i.e., erector spinae: trunk extensor; gluteus medius: hip abductor; rectus femoris: knee extensor; and tibialis anterior: ankle dorsiflexor) (Wang et al., 2015). The sampling rate of the EMG was 960 Hz.

Two force platforms (QQ-mill, Motekforce Link, Amsterdam, Netherlands) were used to collect ground reaction force (GRF) data. The force platforms were able to sample at a frequency of 960 Hz using LabVIEW software (National Instruments, Austin, TX, United States).

Data from the motion capture system, EMG, and force platforms were processed using a custom program written in MATLAB R2020a software (MathWorks, Natrick, MA, United States). The GRF of the moving leg was used to identify the movement cycle (i.e., the stance and swing phases) of each stride. A GRF threshold of 10 N was used to identify the movement cycle (Baumgart et al., 2017). A total of 100 stable strides were selected for analysis (Kubinski et al., 2015). The stance phase occurs from heel-strike to toe-off, while the swing phase occurs from toe-off to heel-strike, as shown in Figure 3.

For joint kinematics, the data were filtered using a 2nd-order low-pass Butterworth filter with a cut-off frequency of 3 Hz. The filtered data was smoothed using a moving average (Chien and Hsu, 2018). The minimum, maximum, and range values of the trunk, hip, knee, and ankle joints during entire movement cycle were calculated (Smith et al., 2016).

For muscle activities, the EMG data were filtered using a 2nd-order Butterworth filter with bandpass and notch filters at 30–350 Hz and 60 Hz, respectively (Adewuyi et al., 2016). The filtered data was full-wave rectified (using a root mean square) and smoothed (using a moving average) (Tabard-Fougère et al., 2018). The EMG was normalized using the resting EMG for each muscle (Wang et al., 2015). The EMG data was also time-normalized from 0 to 100 percent for each phase (Androwis et al., 2018). The integrals of normalized EMG (i.e., trunk extensor, hip abductor, knee extensor, and ankle dorsiflexor) were then separately reported for the stance and swing phases, as well as for the entire movement cycle (Vigotsky et al., 2017; Wu et al., 2019).

For GRF, the data were filtered using a 2nd-order low-pass Butterworth filter with a cut-off frequency of 5 Hz. The filtered data was smoothed using a moving average (Cheng et al., 2020a). The peak values of GRF in entire movement cycle were computed, while the impulse values of GRF were separately computed for the stance and swing phases, as well as for the entire movement cycle (Golyski et al., 2018; Lee et al., 2020; Jafarnezhadgero et al., 2021).

Statistical analysis was performed in Predictive Analytics Software Statistics 18.0 for Windows (SPSS, Chicago, IL, United States). The normality of all variables was determined using the Shapiro–Wilk test. Nevertheless, the data were not normally distributed. Thus, the non-parametric Wilcoxon signed-rank test was used to compare the variables between VR-skateboarding and walking. The p-value was set at 0.05 as statistically significant.

The joint kinematic results for both VR-skateboarding and walking are illustrated in Figure 4; Table 1.

During VR-skateboarding, participants exhibited greater minimum and maximum trunk angles and a wider range of movement in their trunk compared to walking (z = −3.92, p < 0.01; z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively). This indicated that participants bent their trunk forward more and moved in a wider range during VR-skateboarding when compared to walking.

In the supporting leg, participants demonstrated increased minimum and maximum hip (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) and ankle (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) angles during VR-skateboarding compared to walking. However, participants had reduced range of movement in the hip and ankle joints (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) during VR-skateboarding compared to walking. The knee also showed increased minimum angle (z = −3.92, p < 0.01) but reduced maximum angle and range of movement (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) during VR-skateboarding. The findings suggest that, in the supporting leg, VR-skateboarding entailed greater flexion in the hip and ankle joints and a smaller range of movement compared to walking. Additionally, during VR-skateboarding, participants demonstrated decreased flexion in the knee joint and a reduced range of movement in this joint.

In the moving leg, the hip joint angles showed a lower range of motion (z = −3.92, p < 0.01) during VR-skateboarding compared to walking, with both the minimum and maximum angles being higher (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) in VR-skateboarding. The knee joint also showed a lower range of motion during VR-skateboarding (z = −3.92, p < 0.01), with the minimum angle being higher (z = −2.72, p < 0.01) and the maximum angle being lower (z = −3.92, p < 0.01). Whereas, the ankle joint demonstrated a greater range of motion during VR-skateboarding (z = −3.80, p < 0.01), with the maximum angle being higher (z = −3.92, p < 0.01) and the minimum angle showing no significant difference (z = −1.30, p = 0.19) compared to walking. VR-skateboarding involved greater hip flexion and a smaller range of movement in the moving leg compared to walking. However, it also entailed decreased knee flexion and a reduced range of movement, as well as increased ankle dorsiflexion and a greater range of movement.

The muscle activity results for both VR-skateboarding and walking are illustrated in Figure 5; Table 2.

Muscle activity of the trunk extensor in the stance phase, swing phase, and entire movement cycle was higher during VR skateboarding than during walking (z = −3.92, p < 0.01; z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively). This indicated that VR-skateboarding appeared to involve higher muscle activity in the trunk extensor muscles compared to walking.

In the supporting leg, muscle activity of the knee extensor was higher in the stance phase, swing phase, and entire movement cycle (z = −3.92, p < 0.01; z = −3.92, p < 0.01; z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) during VR-skateboarding than during walking. However, muscle activity in the hip abductor and ankle dorsiflexor was lower in the stance phase (z = −3.80, p < 0.01; and z = −3.92, p < 0.01, respectively), swing phase (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively), and entire movement cycle (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) during VR-skateboarding than during walking. The results suggest that VR-skateboarding entailed distinct muscle activity patterns compared to walking, specifically higher activity in the knee extensor but lower activity in the hip abductor and ankle dorsiflexor.

In the moving leg, VR-skateboarding was associated with lower muscle activity in the hip abductor during the stance phase (z = −3.17, p < 0.01) and entire movement cycle (z = −2.68, p < 0.01) compared to walking, but no significant difference was observed in the swing phase (z = −0.85, p = 0.39). Additionally, VR-skateboarding involved lower muscle activity in the knee extensor and ankle dorsiflexor in the stance phase (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively), swing phase (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively), and entire movement cycle (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively) compared to walking. These findings suggest that muscle activity in the hip abductor, knee extensor, and ankle dorsiflexor of the moving leg was lower during VR-skateboarding compared to walking, except for the hip abductor in the swing phase, which showed no difference.

Our results showed that the average stance phase during walking was 63.58% ± 0.24%. During VR-skateboarding, the average stance phase was 51.91% ± 1.74%. Hence, the stance phase was shorter during VR-skateboarding than during walking (z = −3.92, p < 0.01). The GRF results for both VR-skateboarding and walking are illustrated in Figure 6; Table 3 and Table 4.

In the supporting leg, the peak GRF during VR-skateboarding was lower than during walking (z = −3.92, p < 0.01). However, the impulse GRF during VR skateboarding was higher in the stance phase (z = −3.92, p < 0.01) and entire movement cycle (z = −3.92, p < 0.01) but lower in the swing phase (z = −3.92, p < 0.01) compared to walking. These results indicated that during VR-skateboarding, the force loading on the supporting leg and weight distribution in the swing phase were less compared to walking, but there was a greater distribution of weight during the stance phase and throughout the entire movement cycle.

In the moving leg, the peak and impulse GRF during VR-skateboarding was lower than walking (z = −3.92, p < 0.01; and z = −3.92, p < 0.01, respectively). The results showed that during VR-skateboarding, the force loading on the moving leg and weight distribution in the swing phase were lower compared to walking.

The clinical significance of this study is the potential use of VR-skateboarding as a rehabilitation training approach to improve balance. When performing VR-skateboarding, participants had greater range of movement and flexion in their trunk compared to when walking, as well as increased muscle activity and weight distribution in the supporting leg. These differences in biomechanics contributed to increase joint and muscle coordination during VR-skateboarding. Additionally, VR-skateboarding might also be considered a type of closed kinetic chain exercise, where the foot is fixed. Our results indicated that participants consistently kept the foot of the supporting leg on the skateboard. Previous studies have revealed that closed kinetic chain exercises can promote muscle co-contraction to stabilize the trunk and legs. This muscle co-contraction also aids in improving proprioception, or the ability to sense the position and movement of one’s body in space, subsequently improving balance. Furthermore, the use of VR technology may have influenced the observed differences in joint kinematics, muscle activity, and weight distribution. These findings suggest that VR-skateboarding may be an effective training approach for improving balance in individuals undergoing rehabilitation.

In addition, the consideration of the biomechanical characteristics of VR-skateboarding is important for software engineers creating new features and implementing multidisciplinary approaches. By understanding the biomechanics, software engineers can design features that are ergonomic and user-friendly, ensuring that the products they develop are comfortable, safe, and appropriate for balance training. This is particularly important for products or systems that would be used by a wide range of individuals with different physical abilities and characteristics, as it allows for the creation of solutions that are accessible and inclusive. Incorporating a multidisciplinary approach also enables software engineers to consider the diverse needs and perspectives of different populations, such as those with unilateral leg symptoms, to create well-rounded and effective solutions. Overall, the incorporation of biomechanical characteristics in the development process can lead to the creation of innovative and highly functional products and systems that meet the needs of a diverse range of users.

This study had a few limitations. First, for safety reasons, participants held onto a handrail while VR-skateboarding. This may have partially supported their body weight and potentially affected joint kinematics, muscle activity or weight distribution. However, we believed that handrail use is necessary in patient populations to prevent accidents during training. Second, this study conducted the experiment on the same population and at the same speed (i.e., comfortable walking speed). Moreover, participants were all healthy individuals with no leg abnormalities. Therefore, the results of VR-skateboarding in this study should be interpreted with caution when applied to patient populations or different speeds. Third, participants performed VR-skateboarding and walking with bare feet in this study. According to previous studies, wearing shoes can change biomechanical characteristics, particularly reducing force loading (Zhang et al., 2013; Udofa et al., 2019). Hence, this factor should be taken into consideration when applying our findings to VR-skateboarding while wearing shoes.