Virtual reality (VR) is an emerging technology which offers access to a variety of yet partly unexplored possibilities in the field of physical rehabilitation, both from a clinical and research perspective (Canning et al., 2020). VR can be defined as a real-time computer-generated simulation of a three-dimensional environment that replaces the natural sources of stimulation of the real world by offering artificial visual, auditory, and even haptic stimuli (Steuer, 1992). VR technologies are rapidly advancing and nowadays allow users to experience virtual environments, where one can freely move and walk, via inexpensive, off-the-shelf head-mounted-displays (HMDs).

Driving factors for using VR in rehabilitation are that it offers great potential to increase intrinsic motivation, promote a more joyful and motivational exercise experience, and it was also shown to reduce the perception of pain and discomfort (Benham et al., 2019; Chen et al., 2021). As already demonstrated by Chen et al. (2021), VR might increase adherence in exercise, which is a key factor for an effective and efficient therapy. It is, therefore, not surprising that there is a constantly growing body of literature showing that VR can be a safe (Held et al., 2018) and effective tool in supporting physical rehabilitation. There are several recently published studies, including systematic reviews and meta-analysis, which underpin the potential VR offers in gait rehabilitation. It was demonstrated that VR can support gait rehabilitation for patients with Parkinson’s Disease (Mirelman et al., 2011; Lu et al., 2021), retrain gait symmetry (Shideler et al., 2021), restore function after stroke (Ahn and Hwang, 2019; Palacios-Navarro and Hogan, 2021; Peng et al., 2021), support neuro-psychomotor rehabilitation of children with cerebral palsy (de Oliveira et al., 2016), improve balance and gait in older adults (de Vries et al., 2020; Lee, 2020; Willaert et al., 2020; Delgado and Der Ananian, 2021), and is also being used to support gait training after amputation (Darter and Wilken, 2011).

In the field of gait rehabilitation, to date, VR is primarily used on treadmills. However, due to the rapidly advancing technology of HMDs (Saldana et al., 2020) there is also a growing interest in using VR to aid gait rehabilitation during overground scenarios. This would allow people to naturally walk and navigate through a virtual environment. These scenarios potentially offer yet mostly unexplored possibilities for gait rehabilitation, such as training of activities of daily living (Ahn and Hwang, 2019), obstacle crossing (Weber et al., 2021), among other environments that might not easily or safely be replicated in clinical settings. VR also allows the studying of effects of various conditions which are usually difficult to evaluate such as visual impairments, dual tasks, or environmental effects like diffuse lighting conditions, crowded places, and different room sizes (Almajid et al., 2020; Osaba et al., 2020; Taneda et al., 2021).

In light of this rapidly advancing field and the accompanying possibilities it offers for research and therapy, there is an urgent need to understand how VR environments impact individual gait patterns during overground walking. Otherwise, it will not be possible to put VR overground scenarios as a purposeful utility into both clinical and research practice for gait rehabilitation. While there are a few studies available describing the effect of walking in a VR on the treadmill (van der Krogt et al., 2014; Chan et al., 2019; Varas-Diaz et al., 2020), there is only very limited and partly contradictory research available which specifically addressed VR overground walking. For example, existing studies only partly reported adaptions to walking speed, cadence, step width, and step length next to increased variability of the same parameters (Janeh et al., 2018; Canessa et al., 2019; Martelli et al., 2019). One study reported kinematic gait adjustments in terms of increased center of mass excursion (Varas-Diaz et al., 2020). Besides the contradictory nature of the existing literature, it is limited to simple analysis of spatio-temporal gait parameters or center of mass excursion. To date, there is no information available on the effect of VR on standard full-body 3D gait analysis such as gait kinematics and kinetics and their variability. However, such analysis is necessary if the intention is to understand potential effects of VR environments on individual gait patterns during overground walking.

Therefore, the first goal of this work was to address that limitation by evaluating if overground walking in a VR environment in the context of a standard 3D gait analysis setting has a relevant effect on spatio-temporal parameters, full-body gait kinematics and kinetics, as well as on the variability of these variables.

VR offers tremendous possibilities in the design of VR environments. These environments can resemble real environments, in the case of a gait analysis setting this is most likely the gait laboratory, but they also allow to create environments with a freely chosen ‘physical dimension’ (i.e., the size of the virtual environment in real-world units) and design. From a gait analysis perspective, the walkway length is always an important factor, as the walkway length influences self-selected walking speed (Najafi et al., 2011; Stuck et al., 2020), and thus it will affect several important biomechanical variables. However, little is known if the physical dimension of the laboratory itself influences the individual gait pattern, independent by the length of the walkway. That information could be very important, especially when one wants to compare gait analysis data from different laboratories with different physical room dimensions but most likely comparable walkway lengths. Reasons for comparing or even combining data from different laboratories can be manifold, but one emerging topic is the field of data science and in particular big data and machine learning applications (Phinyomark et al., 2018). Furthermore, this information could be important to consider during the design of virtual environments for gait rehabilitation and analysis.

Thus, the second goal of our work was to investigate if the “physical dimension” of VR environments (i.e., the size of a virtual room) influences self-selected walking speed and subsequently relevant biomechanical gait variables during overground walking in VR.

Four out of the 21 participants reported to never have used VR-HMDs before, all others have used them only once or twice.

Regarding our first primary outcome, walking speed, the repeated measures ANOVA indicated significant differences for the four walking conditions [F (3, 60) = 24.678, p < 0.001, eta² = 0.553]. Subsequent post hoc dependent t-tests indicated a significant [t (20) = 6.18, p < 0.001, Cohen’s d = 1.3] decrease of 0.09 ± 0.06 m/s in walking speed when walking in the RLab compared to the VRLab, but no differences between VRLab+ and VRLab− (Figure 3). Regarding the other spatio-temproal parameters the repeated measures ANOVA and subsequent post-hoc tests indicated significant differences (p < 0.001) in all spatio-temporal parameters between RLab and VRLab except for step width and no differences between VRLab+ and VRLab−. See Supplementary Figure S1; Supplementary Tables S1, S2 in the supplement. Following significant results of the repeated measures ANOVA (see Supplementary Table S3 in the supplement), subsequent post-hoc tests also identified the coefficient of variation for step width (p < 0.001) and foot off (p < 0.001) as statistically increased when walking in the VRLab compared to the RLab. Step time variability slightly missed the alpha level (p = 0.086). No other differences were found. See Supplementary Figure S2; Supplementary Tables S3, S4 in the supplement.

Regarding our second primary outcome, GaitSD, the ANOVA indicated significant differences for the four walking conditions [F (3, 60) = 11.253, p < 0.001, eta² = 0.360]. Post hoc analysis with a Bonferroni adjustment indicated a significant [t (20) = -4.831, p < 0.001, Cohen’s d = 1.05] increase of on average 57% in GaitSD (Figure 4) from the RLab (2.8 ± 0.8) to the VRlab (4.4 ± 1.6), but no differences between VRlab+ (4.4 ± 1.6) and VRlab− (4.4 ± 1.9).

SPM indicated some small but significant differences in lower and upper body kinematics, joint moments, and power. However, significant differences exceeding the relevance threshold were only found for decreased plantar-flexion (−3 degrees) and sole angle during push-off (−4 degrees), as well as a decreased ankle plantar-flexion power generation during push-off (−0.4 W/kg) when walking in VR compared to the real-world. The ankle plantar flexion moment during push-off (−0.1 Nm/kg) also demonstrated a reduction slightly greater than the relevance threshold, but was not significant. Details can be found in Supplementary Figures S3–S6 in the supplement.

Regarding the subset of the SSQ, most of the participants indicated that they have not noticed any or only slight symptoms. One participant (5%) reported severe eye strain along with moderate discomfort, and nausea. Five participants (24%) reported moderate difficulties in focusing and three reported moderately blurred vision (14%). For details see Figure 6. Regarding how participants perceived walking in the VR, two participants (10%) reported feeling moderately uncomfortable, 11 (52%) reported that they perceived a difference in their walking pattern while navigating through the VR, seven (33%) reported that it felt different compared to the real-world, six (29%) reported that the HMD limited them during walking, and two (10%) were moderately afraid to bump into real objects. 19 participants (90%) would like to see VR further developed in gait analysis settings (Figure 5).

This study investigated if walking in an overground VR environment provided by an HMD has a relevant effect on selected 3D full-body gait variables in healthy individuals. We also investigated if the “physical dimension” of the VR environment influences self-selected walking speed and asked the participants to report how they perceived walking in the VR.

Participants demonstrated slight gait adaptions in spatio-temporal, kinematic, and kinetic variables during walking in the VR compared to walking in the real world. Walking speed was on average reduced by 0.09 ± 0.06 m/s (−7.3%) in the VR. Earlier research has already partly demonstrated that overground walking in VR can have an effect on spatio-temporal parameters. Martelli et al. (2019) found a decrease in stride length (∼ −4%). Janeh et al. (2017) found a decrease of walking speed (∼ −8%), step length (∼ −6%), and an increase of double support (∼ +7%) when walking in VR. Canessa et al. (2019) reported a decrease in cadence (∼ −10%). There is another particular study which analyzed whether overground VR environments that replicate freezing-of-gait (FoG) provoking situations for people with Parkinson’s disease would exacerbate gait adaptions relevant for FoG (Yamagami et al., 2020). While they have used similar motion capture techniques as in our study, they only reported a subset of spatio-temporal parameters along with step length variability. They found similar effects, such as reduced walking speed (∼ −13%), step length (∼ −11%) along with increased step width (∼ +8%) compared to the physical laboratory. However, they assessed the effect of VR overground walking in patients with impaired motor control and thus, their results cannot be directly compared to our sample of healthy individuals. In comparison to that earlier research, we have analyzed a complete set of standard spatio-temporal parameters used in clinical settings. Our results clearly support the initial findings of the preceding research with healthy participants and demonstrate that individuals change to a pattern of decreased walking speed, cadence, step/stride length, along with an increase in step/stride time, increased % double and decrease % single support (see Supplementary Tables S1, S2 for details in the supplement). In terms of statistically relevance, our results presented moderate to high effect sizes, but from a clinical perspective walking speed was slightly below our relevance threshold of 0.1 m/s (Bohannon and Glenney, 2014). Yet, in eight out of 21 participants that threshold was reached with differences ranging from 0.1 m/s to 0.23 m/s, and 0.15 ± 0.05 m/s on average.

Next to spatio-temporal adaptions, SPM indicated slight changes for lower and upper body kinematics as well as lower body joint moments and powers (see Supplementary Figures S2–S5 in the supplement). However, in almost all cases effects were small, even tough relevance thresholds were reached for plantar-flexion and sole angle during push-off, ankle plantar flexion moment during push-off, and ankle plantar-flexion power generation during push-off. These changes can be attributed to the decrease in walking speed in the VR.

Earlier research has linked HMD wear to increased neck flexion which consequently could place the musculoskeletal system of the head and neck under increased levels of stress. Knight and Baber (2007) evaluated the effects of HMDs on neck posture during HMD simulated paramedic training activities. In their study paramedics presented a greater percentage of postures with the neck flexed by greater than 20 degrees compared while immersed in VR with an HMD compared to the same condition without wearing an HMD. The HMD condition also had a greater percentage of postures that involved the neck being rotated or laterally flexed. They assumed that these changes could be either related to issues of comfort and fit, e.g., to reduce slippage, the wearers may modify their posture to better balance the HMD on the head. Their second explanation was that wearers modify their head posture to overcome a reduced field of view due to the HMD. Restricted field of view is associated with increased head movements and are assumed to have a disruptive effect on the efficient use of coordinated head and eye movements to acquire spatial information (Venturino and Wells, 1990). Our data did not indicate any changes to neck posture which is an important finding as altered neck position could limit the applicability of HMDs in clinical settings. To this end we can only make assumptions why we did not observe changes in head posture. One potential reason could be associated with the different technical advancements of the HMDs. The HMD used by Knight and Baber (2007) dates back several years whereas we have used a state-of-the-art HMD offering a much higher image resolution, a wider horizontal and vertical field of view, and a better head strap. These factors will presumably contribute to a more comfortable and natural VR experience.

Interestingly, participants also demonstrated an increased variability in spatio-temporal parameters and a markedly increase in the lower extremity gait kinematic pattern (in terms of the GaitSD) when walking in the VR. Martelli et al. (2019) already reported an increase in variability of step width (+13%) and a trend in step length variability (+24%). Also Yamagami et al. (2020) have reported an increase in step length variability (+67%) in their patients, however as mentioned above in patients with Parkinson’s disease. In line with this earlier research, we found an increase of 38% in step width variability, next to 40% in foot off variability. Our results extend that earlier results by showing that the increased variability can also be clearly seen in lower body gait kinematics (GaitSD). The demonstrated variability of spatio-temporal parameters during walking in the real world compare well to the optimal thresholds for movement performance in healthy controls described by Ravi et al. (2020). As our results show a significant increase in variability for some of the spatio-temporal parameters when walking in VR compared to the real world this could be an indication for instability induced by the VR during walking. Even tough there is also a well-documented link between reduced walking speed and increasing gait variability (Chien et al., 2015) our observed effects, especially for variability in kinematics, seem not in proportion to the expected increase due to slower walking speeds. This further corroborates the idea that the VR environment induced instability. A similar increase in variability was already reported in early research on gait adaptation for walking in VR environments on treadmills more than a decade ago (Hollman et al., 2006; Hollman et al., 2007). According to that research, individuals reduced stride length, increased step width and demonstrated pronounced variability in walking speed and step width. This was interpreted as a sign of a more conservative or cautious gait provoked by instability due to the VR environment. It was assumed that the instability may be caused by a mismatch of perception and optical flow or the fidelity of the VR system (Banton et al., 2005; Durgin et al., 2005; Sloot et al., 2014). Although we have used a much more advanced VR technology in terms of HMDs, our results point in the same direction. This seems a bit disappointing as one would assume that with better immersion technologies the walking pattern will tend to resemble a normal pattern. One reason why this was not the case might be the reduced field of view HMDs offer. Peripheral vision plays an important role in movement control and postural control and thus its limitation can influence movement (Sivak and MacKenzie, 1990; Iosa et al., 2012). Another reason might be occasionally occurring inaccuracies in alignment of the virtual to the real world caused by relative movements of the HMD to the head or blurred vision due to problems with focusing. This is also in line with the results of the questionnaires where some participants reported both, difficulty in focusing and blurred vision. These points should be addressed in future developments as they might help to increase the immersion grade while reducing unwanted effects on the gait pattern.

Regarding the results of the questionnaire, several participants reported that they perceived at least a moderate difference in their gait while walking and some also indicated that the HMD was a limiting factor during navigating through the VR. Compared to the VR-HMD, augmented reality headsets (AR-HMDs) using optical see-through displays such as Microsoft’s HoloLens 2 seem to offer advantages in that direction as AR-HMDs allow for full field of view of the real environment with digital elements superimposed to the real-world. Using AR-HMDs during overground walking could be a direction for future research. On the other hand, AR-HMDs still have the major drawback that superimposed virtual information can only be visualized with an even more narrow field of view compared to VR-HMDs. However, compared to the frequently used projection technologies for VR (such as curved screens combined with treadmills), a recent study clearly highlighted the advantages of VR-HMDs. Elor et al. (2020) compared the experience of exergaming between HMDs and a Cave Automated Virtual Environment (CAVE) and found that the HMD excelled in in-game performance, biofeedback response, and player engagement for both healthy individuals and users with cognitive disabilities. In addition, AR and VR as well HMDs are technologies that still undergo a tremendous development, so we can expect them to become even more capable, lighter, and robust in the near future.

Our second aim was to investigate if the “physical dimension” of the VR environment influences self-selected walking speed. Research has shown that walkway length influences self-selected walking speed (Najafi et al., 2011; Stuck et al., 2020), and thus it will affect several biomechanical variables. For example the recent systematic review by Stuck et al. (2020) found an overall median absolute difference of 0.04 m/s with a trend towards higher gait speed measured for longer walkway lengths compared to shorter ones. Our results did not indicate any differences in motor control adaptation between the small and the large VR environment. These results could be relevant for future studies developing virtual environments for gait rehabilitation. In addition, these results might also allow for the careful conclusion that data from different laboratories, which have similar walkway lengths but a different room-size might not be biased by the latter. This could be important especially when one wants to combine gait analysis data from different laboratories with different physical dimensions but comparable walkway lengths. Reasons for combining data from different laboratories can be manifold, but one emerging topic is the field of data science and in particular big data and machine learning applications (Phinyomark et al., 2018).

In summary, our most important findings are that people walk slightly slower and show increased gait variability when navigating in overground scenarios through virtual environments, independent by the size of the virtual room. These gait alterations need to be considered when researchers use VR environments in their investigations. In consequence, an important future direction of research should also be directed towards the identification of factors which potentially can contribute to a truly natural walking experience in VR.

There are some limitations in our work which need to be recognized. Firstly, our VR system had a limited size of the tracking area. Therefore, we had to confine the walkway length to about seven meters which is rather small and will presumably result generally in a slightly lower walking speed compared to natural walking. Secondly, our investigation is limited to healthy individuals, tough of varying age. Our study needs to be replicated with elderly people as well as with various patient groups, such as individuals with neurological disorders. Thirdly, the virtual representation of each participant in the VR was limited to their feet. Wirth et al. (2021) have recently shown that walking on a treadmill without self-representation and with different avatar self-representations, i.e., a point-cloud, a silhouette, and a humanoid representation, did not affect the gait pattern. However, the missing self-representation was rated significantly lower in terms of experienced degree of realism and the possibility to act. This could be an important factor for VR-based applications in rehabilitation as one driving factor of VR is to increase intrinsic motivation and promote a more joyful and motivational exercise experience (Chen et al., 2021). Fourthly, all walking conditions were assigned in complete random order. This means that some of our participants (N = 15/21) had one of the VR environments as their first condition. They might have developed a cautious gait already during that first condition which could have biased the following real-world walking condition to a certain extent. Lastly, since in our study the period of wearing the HMD was relatively short in comparison to the typical time spent in a rehabilitation session our findings should be cautiously used to inform the design of prolonged VR-assisted rehabilitation programs.

We have investigated if walking in an overground VR environment with an HMD has a relevant effect on full-body 3D gait biomechanical variables. Our results clearly demonstrate that people adopt a slightly slower walking pattern with small accompanying kinematic and kinetic changes. Most of the kinematic and kinetic differences were too small to be regarded as relevant, but reduced walking speed (−7.3%), increased lower body kinematic variability (+57%) along with increased percent double support time (+4.3%), and increased step width variability (+38%), among others, indicate gait adaptions toward a slightly more conservative or cautious gait due to instability induced by the VR environment. The “physical dimension” of the VR environment does not seem to have an impact on the gait pattern during a standard gait analysis protocol. We propose that these effects need to be taken into account in the design of VR-based overground training devices.

VR and HMDs are technologies that have been and still are undergoing a tremendous development, so we can expect them to become an even more capable tool in future to serve as a meaningful device in clinical practice and research. Our study lays the foundation for upcoming developments in the field of VR-assisted gait rehabilitation as it describes how that technology in overground walking scenarios impacts our gait patterns. This information is of high relevance when one wants to develop purposeful rehabilitation tools in future.