The control of balance during walking for humans is an important problem because the failure to maintain balance leads to falls and often injury. In the United States alone, costs from fall-related injuries amount to 20–30 billion USD annually (Burns et al., 2016). The upright human body is mechanically unstable, with a high center of mass and a relatively small base of support. Balancing this body is already a challenge during standing, requiring continuous regulation of muscle activity by the central nervous system (Morasso and Schieppati, 1999). As a result of intensive research over the last decades, we understand this neural control of balance during standing reasonably well (e.g., Winter, 1995; Peterka, 2002; Maurer et al., 2006; Kiemel et al., 2011).

Balance control during walking has been studied much less in comparison. One reason for this is that walking is highly nonlinear, with the configuration of the body changing substantially as contact with the ground is established and lost at different points during the gait cycle. This implies that at different points in the gait cycle, the central nervous system has different options available for how to generate force along the body and against the ground to affect balance. One well-studied mechanism is to shift the location of the foot placement when taking a step, which changes the pull of gravity on the body during the subsequent swing (Hof, 2008; Wang and Srinivasan, 2014; Bruijn and van Dieën, 2018; Reimann et al., 2018a). Another mechanism is to use ankle musculature to pull on the body during single stance (Hof et al., 2010; Hof and Duysens, 2018; Reimann et al., 2018b). Compared to foot placement shift, the ankle mechanism is limited in effect by the relatively small area of contact under the stance foot. Moments across the ankle joint both pull the body sideways and the foot up, so excessively large moments will result in the foot rolling over. On the other hand, the ankle mechanism has the advantage of being able to act much faster throughout single-stance and even double stance, whereas the foot placement mechanism can only be used when taking a step. For optimal benefit, there is preliminary evidence that humans flexibly coordinate these two balance mechanisms (Fettrow et al., 2018), and possibly others, to adaptively respond to different challenges of balance during walking. A recent study by Vlutters et al. (2019) shows that the (sagittal) ankle mechanisms is activated even when the ankle joint is blocked by an orthosis, indicating that it might be controlled by a dedicated reflex.

Responses to balance-related perturbations generally depend upon the phase of the gait cycle in which they are applied.Reflexes in the lower leg, probed by cutaneous electric stimulation, are highly modulated during walking, and even reverse direction (Yang and Stein, 1990; Zehr and Stein, 1999). Changes of foot placement in response to electric stimulation of the vestibular system are also highly phase-dependent (Bent et al., 2004). In contrast, responses in the upper body to these stimulations do not, or only weakly, depend on the phase of the perturbation. Logan et al. (2014) found similar results in response to visual perturbation, where the lower body response was strongly phase-dependent and the upper body response only weakly. But during normal walking, the only way to affect the lateral translational motion of the upper body is to generate a force against the ground, using muscles along the lower body. In combination, these results pose the question of how the central nervous system achieves a phase-independent response of the upper body using a phase-dependent response of the lower body.

We hypothesize that the central nervous system flexibly coordinates different balance mechanisms in the lower body in a phase-dependent way to stabilize the movement of the whole body in a phase-independent way. Here we test this hypothesis experimentally by perturbing humans walking on a treadmill with Galvanic vestibular stimulation at three different points in the gait cycle, and analyzing how the response in the whole body center of mass (CoM) and the functional force response depend upon the phase of the perturbation. We define the functional response as the displacement between the center of pressure (CoP) and the CoM. This variable is proportional to the acceleration of the CoM when assuming a single-link inverted pendulum model of the body biomechanics (Hof et al., 2005), which has proven to be reasonable approximation in a variety of cases (Kuo, 2007) and has been shown to be related to active balance control (Vlutters et al., 2016). We expect that the functional balance response to these stimuli does not depend on the phase of the perturbation. We focus our analysis to the frontal plane, which is more challenging for balance control compared to the sagittal plane (O'Connor and Kuo, 2009).

Twenty young, healthy subjects (10 female) volunteered for this study. Subjects were between 19 and 38 years old (22.2 ± 4.57), 170.7 ± 8.4 cm tall and weighed 70.5 ± 13.5 kg. Subjects provided informed verbal and written consent to participate. Subjects with self-reported history of neurological disorder or surgical procedures involving the legs, spine or head were excluded. The experimental design was approved by the Temple University Institutional Review Board (#22499). We used a statistical power analysis based on pilot data from another experiment (Reimann et al., 2018b) to determine the number of subjects required to reliably detect functionally relevant differences in the use of ankle roll and foot placement, set to 1 mms integrated CoP-CoM displacement and 5 mm foot placement change (1−β = 0.9).

Subjects were able to successfully complete the experimental task of walking under intermittent Galvanic stimulation. There was no instance of a subject falling or utilizing the safety harness. Subjects walked with an average velocity of 0.97 ± 0.15 m s⁻¹ across the unperturbed steps. Subjects responded to the vestibular stimulation by swaying in the direction against the fall stimulus as expected (Reimann et al., 2017).

We used Galvanic vestibular stimulation to perturb the sense of balance in walking humans at three different phases in the gait cycle. We analyzed the whole-body response to this sensory perturbation in the form of changes of the center of mass (CoM) movement, and the displacement of the center of pressure (CoP) relative to the CoM, which is one aspect of the ground reaction forces generating the whole-body movement changes. Based on previous studies, we expected that the whole-body response would be phase-independent, and hypothesized that the functional response of the CoP-CoM displacement would also be phase-independent.

Our results broadly support and extend the finding that vestibular information is used to maintain balance during walking and that the balance response to vestibular stimuli depends on the phase of the gait cycle (Bent et al., 2004; Iles et al., 2007; Dakin et al., 2013). Bent et al. (2004) reported that upper-body kinematic responses to GVS were independent of the phase of the gait cycle but that the foot-placement response was phase dependent. Dakin et al. (2013) found that there is vestibular influence on motor output across many muscles throughout the gait cycle, but that the contribution of any individual muscle to the overall response is highly phase-dependent.

Subjects responded to the fall stimulus by shifting their CoP in the direction of the perceived fall, i.e., the cathode, which resulted in a body sway in the opposite direction, toward the anode, as expected. The onset time of this functional response did not depend on the phase of the perturbation (see Figure 3 and Table 2). This indicates that although different balance mechanisms are recruited at different points in the gait cycle, the neural controller tends to respond as fast as possible with any available mechanism. For the EARLY and MID perturbations, the foot placement following the perturbation was shifted in the direction of the perceived fall, but the first CoP shift occurred during single stance, indicating that the controller recruits the immediately available mechanism first and does not wait until other, potentially stronger mechanisms become available.

Contrary to our expectations and to results by Bent et al. (2004), the onset time of the whole-body response did depend on the phase of the perturbation. Responses to the LATE perturbations were consistently faster than to the EARLY and MID perturbations, with no significant difference between the latter two conditions (see Figure 2 and Table 1).

The magnitude of both the initial CoP shift in the direction of the perceived fall and the whole-body sway in the opposite direction depended upon the phase of the perturbation. The whole-body response was consistently smaller for the LATE perturbation compared to EARLY and MID (see Figure 2 and Table 1). The functional response, in contrast, was consistently larger for the LATE perturbations (see Figure 3). This is not consistent with Bent et al. (2004), who found that the size of the upper-body response did not depend on the timing of the stimulus. This may be because we used a smaller GVS stimulus than (0.5 mA vs. 1–1.5 mA, Bent et al., 2004) and also different stimulus timing, with our LATE stimulus falling in between their mid-stance and toe-off stimuli.

On the whole, there were few significant differences between EARLY and MID perturbations, but both of these were consistently significantly different from the LATE perturbations. The CoM responses were faster and smaller, whereas the CoP shift responses were larger, but with unchanged onset time. In other words, the onset time of the functional response is phase-independent, but that of the whole-body response is not, and the magnitude of both responses is phase-dependent, but in opposite directions. This difference in response pattern is peculiar, because as noted earlier, all changes of whole-body movement have to be generated by pushing against the ground using the lower body. So how can a smaller CoP shift generate a larger CoM movement change?

One possible explanation is that these differences in the whole-body movement are generated by combinations of forces that do not affect the location of the CoP, so the proportional relationship between CoP and CoM movement based on the single-link inverted pendulum assumption does not hold. One example of such forces is a hip-roll mechanism, where the upper body is rotated around the hip, generating a shear force that pushes the CoM sideways without shifting the CoP (Horak and Nashner, 1986; Reimann et al., 2018a). While this hip mechanism is usually discussed in standing or the single-stance period during walking, a biomechanical equivalent exists during double stance. This would involve differences in the trunk roll angle response early after the stimulus. Figure 4 shows that there are indeed differences, but they are neither large nor conclusive. Explaining such details satisfactorily generally requires a detailed model of the whole sensorimotor control loop, including neural dynamics, muscle physiology, biomechanics, and interaction with the environment. While such models exist for standing (e.g., van der Kooij et al., 1999; Peterka, 2000), there is no comparable model to explain balance control during walking. The available models of walking are not sufficiently detailed to explain the results presented here (Taga, 1998; Geyer and Herr, 2010; Song and Geyer, 2015).

We hypothesized that the CNS generates a functional, phase-independent motor response to balance perturbations by recruiting different balance mechanisms in a phase-dependent way. Based on this hypothesis, we predicted that the CoP shift relative to the CoM in response to a sensory fall stimulus would not depend on the phase-shift in the gait cycle in which that stimulus occurred. This prediction was partially supported by the data, in that the onset of the response was phase-independent, but also partially contradicted, in that the magnitude was phase-dependent. It is important to point out that lack of statistical significance of phase for the functional response is not sufficient evidence of phase-independence. It is possible that there is a phase-dependent difference, but this study was not sufficiently powered to detect it. Indeed, visual inspection of Figure 3B indicates a similar onset time pattern between LEFT and RIGHT, but more data is necessary to tell if this is an actual effect.

Our results indicate that the CNS recruits available balance mechanisms as fast as possible to respond to a sensed threat to upright stability, but the strength of the response is initially limited during single stance and only realized after a period when other mechanisms also contribute. The reason for this limitation is likely biomechanical. Responses to the EARLY and MID perturbations begin during single stance, where only the ankle roll mechanism is available. The degree to which ankle roll can shift the CoP is limited by the surface size of the foot sole (Hof et al., 2010), which is a possible explanation for the reduced magnitude of the CoP shift in EARLY and MID compared to LATE perturbations.

All stimuli were triggered on the right heel-strike, but the direction of the fall stimulus, determined by the polarity of the GVS current, was randomized. The direction of the stimulus had relatively little effect. Statistical analysis resulted in significance of direction after Bonferroni correction only for the magnitude of the trunk roll response. Visual inspection of the results supports this. For the whole body response (Figure 2), functional response (Figure 3) and the lower body response (Figure 5, the patterns of both onset time and magnitude of the response are very similar between LEFT and RIGHT stimuli. Only magnitude of the trunk roll angle response (Figure 4) in the LATE condition seems to be a distinct departure from this pattern. We observed a similar lack of dependence on direction in our previous study using visual perturbations (Reimann et al., 2018b). This lack of effect is slightly surprising, since one might expect that a perceived fall toward the stance foot is less threatening to overall stability than away from the stance foot.

Such a limitation of balance responses during single stance might be the underlying reason that drives some populations with balance deficits to adapt their gait patterns in characteristic ways. For example, people with Parkinson's disease tend to have shorter single-stance times than age-matched controls for similar walking speed (Morris et al., 1994). This might be because they have problems using the ankle roll mechanism for balance control and reduce the duration of the single stance period as a coping strategy.

The details of which balance mechanisms are recruited at which point in the gait cycle are still not well understood. One striking observation from our data is that for the LATE perturbation, the onset of the average CoP shift appears to align with the transition to double stance (Figure 3). But since there is no change in foot placement (Figure 5), this CoP shift cannot be generated by the well-understood foot placement mechanism. One possibility is that the transition to double stance allows recruitment of the push-off mechanism to shift weight between the two stance legs by modulating the push-off force of the trailing limb around that transition (Kim and Collins, 2015; Reimann et al., 2018b). Understanding this phenomenon requires a detailed kinematic analysis that we will perform in future work.

The datasets generated for this study are available on request to the corresponding author.

This studies involving human participants were reviewed and approved by University of Delaware IRB. The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

HR, TF, and JJ designed the experiment. HR, TF, and ET performed the experiment. HR, TF, DG, ET, and JJ analyzed the data. HR, TF, DG, and JJ wrote the manuscript.