Introduction
Bipedal walking is a remarkable ability of humans that requires highly complex neural control to effectively adapt in response to environmental challenges (Jahn et al., 2008; Queralt et al., 2008; Takakusaki, 2013; Tard et al., 2015; Corporaal et al., 2018; Nordin et al., 2019; Pozzi et al., 2019). Impairment of gait adaptation is common in older adults, and among the first indications of gait derangements in neurological diseases. This significantly increases the risk of falls (Caetano et al., 2016), resulting in fractures (Stalenhoef et al., 2002; World Health Organization, 2007), loss of independence (Tinetti et al., 1994; Stalenhoef et al., 2002; World Health Organization, 2007), poor quality of life, and high mortality (World Health Organization, 2007; Osoba et al., 2019).
Many studies have investigated overground gait adaptation in response to obstacles in healthy young and older adults (Sparrow and Tirosh, 2005; Weerdesteyn et al., 2018). However, precise measures of gait patterns in response to real world (RW) demands are scarce (Weerdesteyn et al., 2018), primarily due to the lack of setups in gait laboratories that can fully replicate everyday life environments (Sparrow and Tirosh, 2005).
Previous works used two main approaches to study gait modulation, with fixed (Vallis and McFadyen, 2005; Da Silva et al., 2011; Jansen et al., 2011; Yamada et al., 2011) or mobile obstacles (Gérin-Lajoie et al., 2005; Cinelli and Patla, 2007, 2008; Da Silva et al., 2011; Olivier et al., 2012, 2013; Basili et al., 2013; Huber et al., 2014; Knorr et al., 2016; Vassallo et al., 2017). Fixed obstacles have the advantage of easier standardization across trials and subjects, but they may induce anticipation and pre-planning (Yamada et al., 2011) and do not allow adequate study of the gait modulation that occurs in an outdoor environment, where moving obstacles are prevalent (Sparrow and Tirosh, 2005). With respect to fixed obstacles, moving obstacles cause larger changes in the gait pattern (Gérin-Lajoie et al., 2005), requiring higher mental processing costs (Cutting et al., 1995; Gérin-Lajoie et al., 2005) and being more challenging for people at high risk of falling (Osoba et al., 2020). Gait pattern changes include both gait trajectory (Gérin-Lajoie et al., 2005; Cinelli and Patla, 2007; Basili et al., 2013; Olivier et al., 2013; Vassallo et al., 2017) and velocity (Cinelli and Patla, 2008; Basili et al., 2013; Olivier et al., 2013; Huber et al., 2014; Knorr et al., 2016). In the presence of sufficient space, directional adjustments are preferred (Huber et al., 2014), but braking strategies (i.e., speed modulation) can also be present with obstacle crossing angles of 45° and 90° (Huber et al., 2014). Time constraints, including different obstacle velocities, can also affect gait adaptation. In fact, the (medio-lateral) safety margins for collision avoidance (Cinelli and Patla, 2007, 2008) and the step length (Da Silva et al., 2011) depend on the speed of the obstacle. These results highlight the importance of standardizing obstacle presentation to evoke similar kinematic responses across trials and subjects.
Some previous studies have used a person trained to walk with specific trajectories and speeds as the moving obstacle (Olivier et al., 2012, 2013; Basili et al., 2013; Huber et al., 2014; Knorr et al., 2016). This has the advantage of closely replicating an everyday situation but increases the variability in obstacle presentation, which could not be standardized in these studies. Other studies used robots (Vassallo et al., 2017), mannequins (Gérin-Lajoie et al., 2005; Cinelli and Patla, 2007, 2008), or remote-controlled objects (Da Silva et al., 2011) to improve the accuracy of obstacle presentation, but with some limitations. In particular, the movement of the obstacles was not dynamically adjusted to the behavior (trajectory or velocity) of the subject but fixed and arbitrarily chosen (Cinelli and Patla, 2007, 2008; Vassallo et al., 2017), based on normative data (Da Silva et al., 2011) or on the velocity of the subject during unperturbed walking (Gérin-Lajoie et al., 2005). In addition, in all but one study (Vassallo et al., 2017), the obstacle trajectory was fixed and did not adjust for the ongoing walking pattern of the subject.
Virtual reality (VR) holds great promise for overcoming many of these limitations. Experimental conditions in immersive VR are ecologically valid, realistic, highly controlled, and replicable in a safe environment (Bailenson et al., 2003). A VR setup allows accurate and real-time measurement of the position of the subject and the obstacle (Loomis et al., 1999; Bailenson et al., 2003) for standardization in its presentation. A VR setup can also be enriched with multiple cognitive and motor tasks (dual-task paradigm; Janeh et al., 2019) and perceptual loads (Martelli et al., 2019), requiring additional resources for planning and sensorimotor integration (Mirelman et al., 2011) that can aid a more comprehensive study of gait adaptation (Gérin-Lajoie et al., 2005; Konczak et al., 2009). Obstacle avoidance tasks in VR have shown great potential also for rehabilitation purposes in parkinsonian patients (Mirelman et al., 2011), post-stroke patients (Jaffe et al., 2004), and patients with cerebral palsy (Gagliardi et al., 2018). In most of these studies, however, the use of a treadmill limited the level of immersiveness, which can be resolved by implementing overground walking with a head-mounted display (HMD; Winter et al., 2021).
In recent years, several studies have been successful in developing VR paradigms capable of inducing gait modulation with virtual objects (Fajen et al., 2003; Fink et al., 2007; Gérin-Lajoie et al., 2008; Cirio et al., 2013; Argelaguet Sanz et al., 2015) or virtual persons (Argelaguet Sanz et al., 2015; Lynch et al., 2018; Olivier et al., 2018). Overall, these studies showed similar gait adaptation strategies in VR and RW, with the former characterized by higher obstacle clearance (Fink et al., 2007; Gérin-Lajoie et al., 2008; Argelaguet Sanz et al., 2015; Olivier et al., 2018) and slower velocity (Fink et al., 2007; Argelaguet Sanz et al., 2015). These differences may be due to uncertainties in obstacle localization, possibly caused by excessive attentional demands required by the VR environment (Gérin-Lajoie et al., 2008), absence of body rendering (Fink et al., 2007), and diminished field of view (Fink et al., 2007; Gérin-Lajoie et al., 2008). This latest hypothesis was, however, questioned by Jansen and coll., who showed kinematic gait changes during static obstacles avoidance only for a field of view as small as 40°×25° (Jansen et al., 2011), and by Knapp and Loomis, who found no underestimation of distances in relation to a decreased field of view (Knapp and Loomis, 2004).
All these studies have shown the great potential of VR in the study of gait modulation, but they are not without limitations. First, most of them used CAVE-like systems (Cruz-Neira et al., 1992) with joystick navigation, due to limited walking space (Lynch et al., 2018; Olivier et al., 2018). These devices are very expensive and require trained personnel, thus reducing their use in clinical and rehabilitation facilities. Second, studies of gait modulation in VR focused primarily on validating experimental setups previously used in RW rather than developing new ones. Static obstacles were preferred over moving obstacles, with the aim of understanding the impact of different characteristics of virtual obstacles on walking behavior (Bailenson et al., 2003; Argelaguet Sanz et al., 2015) or different avoidance strategies between VR and RW (Fajen et al., 2003; Fink et al., 2007; Gérin-Lajoie et al., 2008; Argelaguet Sanz et al., 2015). The few studies employing moving obstacles in VR (Lynch et al., 2018; Olivier et al., 2018) used joystick navigation and did not adjust the movement of the obstacle to the movement of the subject. The potential of VR in replicating everyday environments and standardizing the presentation of obstacles has yet to be fully exploited.
Ours is a proof-of-concept study that aimed to demonstrate the feasibility of using a fully-immersive VR environment to study overground gait adaptation and obstacle avoidance in a highly standardized manner. We tested this protocol on a small group of young healthy subjects and described biomechanical features of overground gait modulation for collision avoidance. We employed an HMD to ensure immersiveness and facilitate future clinical applications. A virtual agent (VA) was preferred over a virtual object to replicate one of the most common scenarios in daily life, which is walking while another pedestrian crosses the path (Basili et al., 2013; Olivier et al., 2013; Huber et al., 2014; Knorr et al., 2016). A full-bodied VA was shown to induce larger gait adaptation with respect to inanimate objects (Argelaguet Sanz et al., 2015; Lynch et al., 2018). For the first time, the movement of the object (i.e., the VA) was standardized based on the ongoing movement of the participant to ensure a constant perturbation across subjects and trials. The speed of the VA was defined so that participants were induced to modulate their gait to let the VA pass first. Indeed, when two pedestrians cross their paths, the one way contributes more to collision avoidance (both in terms of walking trajectory and speed changes) than the one passing first (Olivier et al., 2013; Knorr et al., 2016). This setup was designed for future studies on patients with Parkinson’s disease, where specific gait disturbances such as gait freezing predominantly occur during gait pattern modulation (e.g., confrontation with obstacles; Pozzi et al., 2019).
We had two main working assumptions: the first was that a highly realistic and immersive virtual environment would not alter the gait pattern. For this part of the study, our results should be considered preliminary, and we defer validation of our setup to future works with more subjects. The second hypothesis was that the presence of the VA would induce significant gait modulation, both in terms of stride velocity, length, and duration, and in terms of stride width, lateral trunk displacement, and lateral deviation of the gait trajectory. This second goal, especially for future clinical research applications, should be considered more relevant and the main purpose of this work.
Discussion
Our study showed that a fully-immersive VR environment is an effective setup to induce gait adaptation for obstacle avoidance. The consistent and replicable gait modulation induced by the VA in all participants indicates that this is a promising tool to study gait adaptation in a safe, highly-standardized, controlled, and lifelike environment.
The proposed VR environment did not induce changes per se in the basic kinematic features of gait. Still, we cannot rule out that the limited sample size may have prevented capturing significant differences, especially considering that previous studies described some alterations (e.g., stride length and velocity, cadence, heading angle) between walking in real and virtual environments (Hollman et al., 2006; Menegoni et al., 2009; Katsavelis et al., 2010; Janeh et al., 2017). Future studies are warranted to confirm these results in larger case series.
In all subjects, interaction with the VA induced significant changes in both gait trajectory (Table 3) and velocity, particularly the latter (Table 2 and Figure 3). This was expected, based on the crossing angle of the VA (Huber et al., 2014) and the presence of the walkway (Figure 2), and supports previous observations on the role of speed adjustments in obstacle avoidance (Cinelli and Patla, 2008; Basili et al., 2013; Olivier et al., 2013; Huber et al., 2014; Knorr et al., 2016). By defining a limited space for gait modulation, we made speed changes alone insufficient to avoid a collision with the VA (Huber et al., 2014), thus requiring parallel adjustments in gait trajectory (Huber et al., 2014) and step length (Gérin-Lajoie et al., 2005). The recovery of stride length, only partially accompanied by an increase in stride velocity, made the second modulator the longest in duration (Figure 3). Of note, changes between modulators were smooth, and values gradually restored to the unperturbed range during the second and third modulators (Figure 3).
Our VR paradigm also induced some additional mediolateral changes, which consisted of an increase in stride width and medio-lateral sway (Tables s 2 and 3; Gérin-Lajoie et al., 2005; Cinelli and Patla, 2007, 2008; Huber et al., 2014). The increase in stride width may reflect a strategy to ensure balance for the avoidance of the VA, which perturbs postural stability as suggested by the increased medio-lateral sway in the VA/A+ condition (Table 3). Changes in stride width, however, were inconsistent across subjects and these findings should be further confirmed in larger cohorts.
Finally, we noticed an increase in trunk inclination during the second and especially third modulator. This could be an attempt to maintain sufficient personal space relative to the VA (Gérin-Lajoie et al., 2005, 2006, 2008; Argelaguet Sanz et al., 2015; Lynch et al., 2018; Olivier et al., 2018), particularly during strides in which the VA was close to the participant (i.e., the second and third modulators).
One limitation of our study is the choice not to randomize between conditions (i.e., RW, VR/A, and VR/A+). The main reasons for this are that switching repeatedly from RW to VR can induce discomfort (e.g., dizziness and nausea), and requires additional time to remove and reposition the HMD, reducing subject compliance and the number of overall trials. Randomization between the VR/A- and VR/A+ conditions would have resulted in wait-and-see behavior, with additional gait changes given just by the expectation of whether the VA would begin moving. Instead, we wanted subjects to know that they needed to modulate their gait.
In conclusion, our VR setup was able to effectively induce timely gait modulation in a standardized, immersive, and realistic scenario that simulated a person crossing the path of the participant. Modulation involved both temporal and spatial adaptations of the gait cycle, as well as gait trajectory and trunk inclination. The use of this protocol in older subjects and patients with gait disorders could be useful to elucidate specific alterations in gait adaptation, and have diagnostic and therapeutic (physical therapy) value for future studies (Dockx et al., 2016; McCrum et al., 2017). In particular, we envision that the adaptive gait behavior induced by our VR paradigm may represent an ideal trigger for the occurrence of gait freezing episodes in patients with Parkinson’s disease and other neurological disorders (Fasano et al., 2017; Pozzi et al., 2019). This assumption is based on our experience and previous studies describing the occurrence of gait freezing episodes, mainly during modulation of gait when facing an obstacle under conditions of temporal or spatial constraint (Nieuwboer et al., 2001; Hausdorff et al., 2003; Pozzi et al., 2019).