As virtual reality (VR) technology continues to expand with more affordable head mounted displays and improved quality of the user experience, the accessibility of this once novel tool is steadily rising. This has led more researchers and clinicians to adopt VR as a useful tool to study human gait. A key component of VR lies in its ability to control immersion through both non-immersive and immersive environments. Non-immersive VR typically utilizes treadmills and different types of screens (monitors, projectors etc.) to display an environment in the real world. Immersive VR utilizes a head mounted display (HMD) to create fully realized environments within the headset for subjects to walk through and explore. It is clear to users the differences of these non-immersive and immersive VR environments. The experiential quality of ‘being there’ using a HMD in immersive environments contrasts with the video-game like experience of looking at a screen or monitor in non-immersive environments. Consequently, sensory and perceptual experiences are often rated higher and closer to the real world by users who are in an immersive VR environment compared to a non-immersive VR environment(1).Furthermore, virtual environments can be specifically tailored to the individual and increase the motivation for completing rehabilitation and research paradigms(2). The flexibility and immersive qualities of VR make it a useful tool to create environments that are ecologically valid to better understand the impacts of environments and cognitive interferences on movement outcomes such as gait.

Over the past decade, VR has been increasingly used to study the gait related impacts of postural threat, anxiety, dual-tasking, and obstacle avoidance in both young healthy and clinical populations(3–5). VR has also been used as a tool for gait training, in both young and older adults(6). A review examining gait training in VR and non-VR environments among clinical populations found positive results for gait training under both VR and non-VR contexts. The study also found that training with VR led to better gait outcomes in post-stroke patients compared to training without VR, highlighting the efficacy of using VR as a rehabilitative tool that could have better results than traditional gait training (6). Possible explanations for these positive findings were attributed to VR increasing patients’ motivation and enjoyment during training, especially in immersive environments which can generate a stronger feeling of being physically present (6). While these studies do show positive impacts of VR based training on gait, the fidelity of studying human gait in virtual reality remains debated, since few studies that have been conducted have found mixed results.

Treadmill-based VR studies have shown healthy adult subjects in VR walked with reduced stride length, increased step width, and increased stride velocity variability compared to treadmill walking in the real world (7, 8). While some overground walking studies show similar findings of reduction of stride and step lengths in VR compared to real-world (9, 10) other overground VR walking studies show no such differences in stride and length between the VR and real-world conditions(11, 12). These mixed results may be due to the modality of the gait task (treadmill vs overground), as the differences in walking on a treadmill could affect placement of the foot compared to real-world. Differences could also arise from a limited sense of ‘immersiveness’ seen in treadmill tasks compared to overground walking using HMDs (12). Alongside step and stride length measures, clinically important measures such as gait speed are also influenced by virtual environments compared to the real world.

Gait speed is an important indicator of overall quality of life and the ability to meet the task demands of the activities of daily living (ADL)(13). As such, many VR based gait training studies have compared gait speed within VR and real-world environments to evaluate the efficacy of VR based training (6). Comparing gait speeds, most studies show a decrease in gait speed and cadence(9–11). This reduction in cadence and gait speed is also observed in treadmill studies(7, 8). One study showed no difference in gait speed between the real-world and VR environments(12). A possible explanation for this result could be due to the procedural aspects of this study where the lack of randomization between VR and real-world walking contributed to subjects being familiar with the walking environment in the real world and not being perturbed by the VR environment.

Similar to gait speed, gait variability is also an important indicator of mobility, especially in older adults as it has been associated with reductions in confidence during walking and has been shown to be a predictor for future falls(14–16). Examining gait variability, studies have shown increases in stride velocity %CV and step width %CV in VR environments compared to real-world (7, 9, 17). However, other metrics of gait variability, such as step length %CV and step time %CV were not different across environments (9) .One study showed no differences in any gait variability (%CV) between VR and real-world environments although they did report differences in non-linear measures of variability (18). It is important to note the scarcity of gait variability measurements in VR versus real-world comparisons and few studies have examined gait variability beyond stride velocity and step width variability. This paucity limits the understanding of how virtual environments may influence gait variability outcomes compared to real-world walking and implicates whether changes in gait behaviour in a virtual environment can translate into meaningful changes in real-world walking as well.

Combined with mixed findings and the scarcity of relevant gait variability outcomes, further research is needed to quantify how gait changes in overground walking in an immersive VR environment compared to real-world walking. Thus, this study aimed to investigate the influence of an immersive VR environment on overground gait. To address the scarcity of research on gait variability a wide array of outcome measures including multiple measures of variability were selected for analysis. Based on previous overground studies, it was hypothesized that subjects would demonstrate decreased gait velocity, step length and increased step width in the VR environment compared to real world. Furthermore, it was also hypothesized that subjects would display increased gait variability in the VR environment compared to real world. Addressing the question of gait differences between VR and real-world environments will provide a much-needed view on how virtual environments can affect gait and how well a virtual environment represents real-world gait If studying and evaluating gait in virtual environments does not adequately represent real-world behaviour, then there could be a consequence in the understanding the efficacy of virtual training and rehabilitation. Both positive and negative outcomes of virtual training could be associated with the training methods and protocol rather than on the environment itself that alters gait. This can inform how VR may be adopted as a tool for gait training in the future, which specific types of VR environment may be more suitable, and how clinicians and researchers can better contextualize changes in gait in the virtual setting compared to the real-world.

Methods

Twelve young healthy adults (n = 12, 9F, age = 22.8 ± 3.48 years, height = 168 ± 8.01 cm) were recruited as participants from the University of Waterloo graduate and undergraduate student population (Table 1). Exclusion criteria included any previous difficulty experienced while using VR (either in the past or during familiarization trials) such as nausea, severe fatigue, light-headedness and other discomfort that could potentially impact gait. For safety reasons, participants that had non-corrected visual impairments, balance impairments, medical implants, musculoskeletal disorders, neurological disorders (stroke, migraine, seizures), and heart conditions were excluded from the study. Recent history of physical injury was also an exclusion criterion where participants with injured limbs, concussion or usage of assistive walking devices were not selected to participate in the study. This study was approved by the University of Waterloo Research Ethics Board (REB#43902). All participants provided written informed consent to the researchers prior to participation.

After obtaining consent and completing a demographics questionnaire (see Table 1), participants were instructed to walk in one of two counter balanced blocks. The blocks represented walking either in the virtual environment (VE) or the real environment (RE). After the end of one block, participants were given time to rest before starting the next block of walking. Walking was performed over a 6m Zeno™ Walkway gait carpet, with 10 passes of walking collected during each block, totalling to 120m of walking per participant. Participants were instructed to start 1.5 m prior to the start of the carpet and to walk an additional 1.5 m past the end of the carpet on every pass. A virtual replication of the lab was created using Unity3D to minimize the effect of a novel VR environments on walking performance. Prior to data collection, participants were given time to familiarize themselves with the environment (Fig. 1.A) and were instructed to take as much time as they needed to feel comfortable walking in the virtual environment. During the VE condition, participants wore a wireless HTC Vive Pro Eye as they walked. Participants were instructed to walk at their natural self-selected pace in both the VE and RE (Fig. 1.B).

Demographic information can be found in the table below.

Shows the mean and standard deviation of the demographic makeup of the subjects.

Table 2. Summary results of gait parameters across environment conditions

Shows the mean (SD) of each gait parameter across the two environment conditions alongside the respective p-value of the pairwise Student’s t-tests or Wilcoxon-rank tests. Significant differences are denoted by bolded values. ⁺Step Length Variability was calculated using n = 11 due to an outlier caused by equipment error.

Due to a smaller than 20 sample size, effect sizes were calculated using Hedges’ g using the Hedges’ correction to the Cohen’s d value. Large effect sizes were observed (g > 0.80) in cadence, step length, velocity, double limb support, step time, and step width variability across the two environment conditions. No significant differences were observed between step width, step length variability, step time variability, and stride velocity variability.

Alongside pair-wise t-tests, correlations of the gait parameters that showed significant differences between the two conditions were also conducted. This was to examine whether gait behaviour in both environments were consistent with one another (i.e. does having a naturally faster cadence/velocity in real life also result in faster cadence/velocity in the virtual environment). Results are shown in Fig. 3. (A-F) reporting the Pearson’s correlation coefficient of cadence (r = 0.71), double support % (r = 0.92), steps length (r = 0.91) and velocity (r = 0.84), step width variability (r = 0.92) and step time (r = 0.67) between the two VE and RE environment conditions. All parameters except step time showed strong correlations (r > 0.70) between the two environment conditions. Double support time (%) (Fig. 3. B), step width variability (Fig. 3. C), and step length (Fig. 3. E) showed the strongest correlations between the two environments, r = 0.92, r = 0.92, and r = 0.91 respectively. Weakest correlation was observed between step time, showing a moderate correlation of r = 0.67 (Fig. 3. F).

The primary objective of this study was to investigate the influence of an immersive VR environment on overground gait. This study showed that there were indeed differences in gait between a real- (RE) and virtual environments (VE) similar to past work (7, 9, 11, 12, 17, 19). The results supported the first hypothesis, showing an ~ 8.5% decrease in gait velocity and ~ 5.2% decrease in step length in the VE compared to the RE. Furthermore, a ~ 3.5% reduction in cadence, alongside a ~ 6.6% increase in double-stance (%) time, and a ~ 3.2% increase in step time in VE might suggest that participants adjust their gait behaviour in VE compared to RE. One interpretation is that reductions in pace may reflect a cautious gait slowing, however there no changes observed for step width during the VR compared to the RE.

An alternative interpretation may relate to the sensory differences in VE compared to RE which may contribute to the adoption of a more cautious gait in a VE. Research has shown that modulation in optic flow can impact spatiotemporal measures of gait (20, 21).The virtual environment presents a manipulated and different optic flow streamed to the HMD compared to real life, which could present itself as a sensory challenge, requiring more cautious steps and overall gait behaviour to overcome. Although significant advances in virtual reality technology have been made since Hollman et al. (2007) and the fidelity, resolution, and refresh rate of VR headsets are far better, these differences in gait still persist between real and virtual environments.

Our second hypothesis predicted that gait variability would be greater under VE compared to RE. This hypothesis was partially supported by the results since step width variability showed a ~ 20% increase in VE compared to RE. Although previous treadmill-based studies have shown difference in gait variability between virtual and real environments, particularly in stride velocity variability, no such differences were observed in this study [Hollman 2006]. Overground VR studies have also shown only step width variability changing between VE compared to RE (9, 17). Our findings suggest that HMD based VEs may only impact step width variability during walking and other measures of variability (stride length, stride velocity, step time, stance time etc.) may be consistent between environments.

Despite absolute differences in gait outcomes, the correlations (Fig. 3) show a more nuanced influence of real and virtual environments on gait. Strong correlations (all correlation values > 0.70, with the exception of step time) between gait parameters while navigating both the VE and RE show that the individual’s behaviour between these two conditions is relatively consistent. For example, if a participant had a relatively high cadence, step length, velocity, or high double support stance time in the RE, they were also likely to show similarly high values of these parameters while walking in the VE. Thus, gait behaviour was consistent between conditions within an individual.

A critical component missing while walking in a virtual environment is the visual information of the individual’s limbs as they walk. In this environment, visual information of the legs, arms, and torso are not present in the virtual environment which may contribute to the adoption of cautious gait in VE. Research has shown that vision plays a strong role in foot placement and step placement. Occluding visual information of the foot showed an increase in foot placement error on a target(22). Thus, a more hesitant or cautious gait may be adopted in the VE, where visual feedback of the lower limbs was not present, to minimize placement error. Research has also described how swing limb trajectory during gait is influenced by exproprioceptive information from the visual information of the limbs (23). Thus, an occlusion of this exproprioceptive information may influence stepping trajectory while in the virtual environment, resulting in altered gait. Future work investigating the influence of visual feedback in virtual environments may be valuable in improving the use of VR when assessing gait.

An important limitation in all VR related work falls on the quality, fidelity, and overall “look” of the environment itself. Although high quality assets and meticulous detail was used to develop and replicate the physical environment in VR, there is still a noticeable difference between the virtual environment and the real-life environment. This discrepancy may contribute to participants walking differently by virtue of being in a ‘separate novel place’ from the real world, despite the efforts in trying to minimize differences. As more assets continue to be made available, future work could improve on the quality of these assets to make the environment look more life-like, minimizing the difference between virtual and real environments.

The VIVE Pro HMD have a field of view of 110^o which is lower than the typical healthy human eye field of view 200^o and thus could also impact gait performance in the virtual environment (19, 24). Furthermore, this could be exacerbated by the influence of an added weight of the HMD on the head and neck during walking which is present in the VE condition but was not present in the RE condition.

Reduction in gait speed, cadence, step length and an increase in double support time (%), step width variability, and step time was observed when walking in a virtual environment compared to a real-life environment. Although these gait differences are present, the relative behaviour of gait is maintained in both the virtual and real-life environments such that a fast walker in real life also tends to walk fast whilst in a virtual environment and vise versa.