Gait parameter control timing with dynamic manual contact or visual cues

Abstract

We investigated the timing of gait parameter changes (stride length, peak toe velocity, and double-, single-support, and complete step duration) to control gait speed. Eleven healthy participants adjusted their gait speed on a treadmill to maintain a constant distance between them and a fore-aft oscillating cue (a place on a conveyor belt surface). The experimental design balanced conditions of cue modality (vision: eyes-open; manual contact: eyes-closed while touching the cue); treadmill speed (0.2, 0.4, 0.85, and 1.3 m/s); and cue motion (none, ±10 cm at 0.09, 0.11, and 0.18 Hz). Correlation analyses revealed a number of temporal relationships between gait parameters and cue speed. The results suggest that neural control ranged from feedforward to feedback. Specifically, step length preceded cue velocity during double-support duration suggesting anticipatory control. Peak toe velocity nearly coincided with its most-correlated cue velocity during single-support duration. The toe-off concluding step and double-support durations followed their most-correlated cue velocity, suggesting feedback control. Cue-tracking accuracy and cue velocity correlations with timing parameters were higher with the manual contact cue than visual cue. The cue/gait timing relationships generalized across cue modalities, albeit with greater delays of step-cycle events relative to manual contact cue velocity. We conclude that individual kinematic parameters of gait are controlled to achieve a desired velocity at different specific times during the gait cycle. The overall timing pattern of instantaneous cue velocities associated with different gait parameters is conserved across cues that afford different performance accuracies. This timing pattern may be temporally shifted to optimize control. Different cue/gait parameter latencies in our nonadaptation paradigm provide general-case evidence of the independent control of gait parameters previously demonstrated in gait adaptation paradigms.

gait speed emerges from the complex coordination of many degrees of freedom controlling body segments of various momentums. Although gait parameters underlying gait speed such as stride length and cadence typically change in a correlated fashion to adjust speed (cf. Morris et al. 1998), evidence supports their independent control (Klarner et al. 2013; Malone et al. 2012; Prokop et al. 1997). Such individual control processes within the larger scheme of gait control have been teased out by adaptation paradigms that selectively perturb dynamics of specific effectors such as by adding weights (Smith et al. 2013) or using separate treadmills for each leg (Choi and Bastian 2007; Ogawa et al. 2014; Prokop et al. 1995) or sensory-motor relations (such as with prism goggles; Alexander et al. 2011). Behavioral adaptations observed in these studies have revealed independent elements of gait control specific to environmental (Yen et al. 2012) and behavioral contexts (Alexander et al. 2011; McNeely and Earhart 2010), limbs (Choi and Bastian 2007; Prokop et al. 1995; Smith et al. 2013), muscles (Lam et al. 2006; Ogawa et al. 2014), and kinematic aspects of the gait cycle (Blanchette and Bouyer 2009; Malone 2012). The feedback or anticipatory mode of these controls is deduced from behavioral adaptations observed during or after perturbed conditions, respectively. Such evidence has demonstrated anticipatory or feedback control of gait in particular muscles (Lam et al. 2006; Ogawa et al. 2014) or in the same muscle at different times of the gait cycle (Blanchette and Bouyer 2009). Collectively, this evidence speaks to the independent control of gait parameters underlying the flexibility of gait control.

The aim of the present study is to get a clearer picture of the temporal aspect of the control of specific gait parameters to achieve a specific desired gait speed. Specifically we asked, what are the timing relationships (how much anticipatory or reactive) between changes in external cues and in individual gait parameters used to control gait speed? It has been suggested that sensory feedback is weighted variably throughout the gait cycle (Logan et al. 2010; Patla and Vickers 1997). We follow on this idea of changing sensory weights by determining at what times during each step does a sensory cue weigh the most with respect to individual gait parameters related to gait speed.

In our experiment, participants adjusted their gait speed on a constant-velocity treadmill to maintain a constant distance with a fore-aft oscillating cue (either seen or touched). Cue motion toward or away from the participant signaled to walk slower or faster. Thus the cues signaled walking speed. These predictable, changing cues represent a dynamic environmental variable, rather than a perturbation as used in prior work. With such cues we can determine the instantaneous time of the ongoing environmental cue for which kinematic gait parameters are adjusted to achieve the gait speed dictated by the cue. The timing of cue values relative to individual gait parameters will indicate an anticipatory or feedback control mode. Inertias and sensory flow associated with treadmill vs. typical overground gait notwithstanding, the present paradigm reveals temporal patterns of spontaneous unencumbered gait speed adjustments, independent of adaptation/(re)learning mechanisms associated with many prior gait perturbation paradigms.

Our second broad aim was to address how alterations in timing patterns might relate to speed control accuracy. We controlled for cue oscillation frequency and mean gait speed by testing cues oscillating at four frequencies while participants walked at four treadmill speeds. We also compared two types of cues, manual contact vs. vision, to determine whether the coordinated control of gait parameters generalizes across these cues. These modalities have been studied extensively in the context of balance control. Static posture can be entrained to subliminal motion of visual arrays (cf. Jeka et al. 2000). Manual contact too light to provide significant mechanical stabilization engages reflexive postural responses to stabilize static posture as well as vision (Jeka and Lackner 1995). The similar influence of environmental cues on gait is demonstrated by the use of visual (Morris et al. 1996), auditory (Nieuwboer et al. 2007), or haptic (Rabin et al. 2015) cues implemented to overcome the shuffling and stalled gait of people with Parkinson's disease. Comparisons among these conditions will determine how cue/gait parameter temporal patterns generalize across gait speeds, cue frequencies, and cue modality and also how changes in cue/gait parameter temporal patterns speed relate to changes in speed control accuracy among these experimental manipulations.

To characterize the strength and timing of gait parameter/cue motion relationships we employed a novel modified correlation function analysis. Instead of correlating continuous time series, we used a series of discrete gait parameters (step length, peak toe velocity, step-cycle duration, and single-support and double-support durations) from each trial's step sequence. These were correlated with a corresponding sequence of discrete cue velocities taken at a given time lag within each step period. The correlation function maximum represents the trial's cue/gait relationship strength, and a maximum's associated time lag indicates the time at which the cue cues the gait parameter underlying gait speed.

METHODS

Participants.

Eleven healthy people employed at New York Institute of Technology College of Osteopathic Medicine (8 male, ages 20 to 40) participated after giving written informed consent. The New York Institute of Technology Institutional Review Board approved all procedures.

Task.

Participants were instructed to maintain a constant interval between their hips and a marker on the surface of a conveyor belt for 25 s (Fig. 1). The instruction was given in terms of the participants hips to encourage participants to vary their gait speed appropriately rather than change the orientation of their upper body during the task. The conveyor belt (18-cm wide, 6.1-m long; McMaster Carr, Dayton, NJ) was driven by Parker Compumotor G3 Dynaserv motor and amplifier (Parker Hannifin, Rohnert Park, CA), controlled with Lab View software (National Instruments, Austin, TX).

Fig. 1.

Cue-modality conditions. Subjects walked on a constant-speed treadmill in line with the end of the conveyor belt, either touching it with one hand or looking at it. Subjects adjusted their walking speed so their trunk would the pattern of oscillation (0.1 m amplitude) of the cue mounted on the conveyor belt.

Experimental conditions varied.

Experimental conditions varied cue modality (vision, manual contact; see Fig. 1); treadmill speed (0.2, 0.4, 0.85, and 1.3 m/s); and cue motion [none, ±10-cm amplitude (20 cm peak-to-peak) at 0.09, 0.11, and 0.18 Hz; see cue kinematics shown in Fig. 5]. Cue oscillation periods were >5 s, greater than the mean gait step-cycle period. During visual cue trials, arms were crossed across the participants' chest to control for arm swing precluded by the constraint of touching the cue in the manual contact trials (Fig. 1, left). Participants had peripheral view of the laboratory, decorated with assorted stationary kinematics-science apparatus and middle-aged scientists. During manual contact trials eyes are closed to prevent visual feedback. Participants were advised that their hand should not slide on the conveyor belt surface and that the manual contact cue was meant to be felt rather than something to lean on. Participants were encouraged to configure their touching arm in a manner comfortable for them. Participants performed one 25-s trial in each condition. Trial order was pseudorandomized individually per subject to control for possible learning effects. The experiment required about an hour for each participant.

Fig. 5.

Cue tracking accuracy of one representative participant. Time series of trunk position (solid line) and cue position (dotted) superimposed to illustrate the fidelity of the former to the latter, for all 32 conditions. Columns organize 4 treadmill speeds, within which haptic feedback trials are left and visual feedback trials are right. Rows correspond to 4 cue-oscillation frequencies.

Measurements.

A seven-infrared-camera system (Vicon, Denver, CO) recorded the three-dimensional motion of markers attached to third thoracic (T3) and fourth lumbar (L4) vertebrae, bilaterally on the posterior calcaneus (heel) and the distal second metatarsal (toe) (100 Hz). Low-pass filtered (20 Hz) data were analyzed with Matlab (Boston, MA).

Step-cycle identification.

Our analysis was on a step-by-step (toe-off to contralateral toe-off; Ivanenko et al. 2004) basis, rather than a full gait cycle (2 steps) because we presumed the continuous cue velocity change required observable differences between steps. Toe-off was estimated to occur at the local maximum of the horizontal component of acceleration before peak velocity of the toe marker; heel-strike at the time of a local maximum in the vertical component of acceleration following peak velocity of the heel marker (Hreljac and Marshall 2000).

Gait parameter identification.

For each step cycle we determined: the step length (sagittal distance between toe markers), peak horizontal toe velocity, step duration, and single- and double-support durations. All temporal computations and outcomes are in units of seconds, rather than step-cycle percentage, because of the nonlinear relationship of gait cycle support phases and gait speed (Perry and Burnfield 1995). Swing and double-support phases of the step durations shorten with increased gait speed but not in equal proportion: the faster the gait, the less relative time spent in double support and the greater the percentage in single support (Murray et al. 1966).

Gait speed control strategy analysis.

To gauge the relationship between step-cycle parameter changes and cue motion we correlated the series of gait parameter values for all a trial's steps with corresponding series of instantaneous cue velocities for all a trial's steps, each taken at the same relative time within the step cycle. Figure 2 shows an example of gait parameter series for a step sequence (peak velocity, filled circles) along with a corresponding instantaneous cue velocity series at zero lag (unfilled circles) and an example of an instantaneous cue velocity series at a nonzero lag (Xs). By determining the time during the step cycle at which the instantaneous cue velocity series correlates highest with the gait parameter series, we can characterize a timing relationship (how much anticipatory or reactive) between the gait parameter changes to control gait speed and external velocity cues.

Fig. 2.

Relating gait parameters to cue velocity. In this example, toe velocity data (thin solid and dashed lines) and cue velocity (thick dotted line). A sequence of peak toe velocities for each step is plotted in filled circles. A corresponding series of cue velocities at zero lag is plotted as unfilled circles; one nonzero lag (+0.35 s) series is plotted as Xs.

Not all of the gait cycle parameters of interest are instantaneous (e.g. double-support duration does not “occur” at once; stride length is not a temporal parameter at all), but the time at which the cue velocity is most associated with them may be. By necessity the cue velocity time associated with a gait parameter is discussed in terms of a time relative to a gait cycle event. Our temporal analysis is time based, rather than frequency or step cycle percentage based. However, the sinusoidally changing, nonuniform gait speed may “distort” step-cycle subintervals nonlinearly more so for higher cue frequencies. To minimize this effect and facilitate averaging across conditions, we determined the instantaneous cue velocity associated with various gait parameters relative to a temporally near event in the step cycle associated with each gait parameter, rather than a common event in the step cycle.

Thus we correlated each trial's step-series gait parameter values with series of corresponding instantaneous cue velocities. For each trial we repeated the correlation, shifting the delay at which the cue velocity series was taken relative to the appropriate step-cycle event. The absolute maximum of these correlations among time shifts of the cue velocity series represents the time at which cue velocity is either sampled as feedback or anticipated to control the correlated gait parameter. The maximum correlation was calculated as r = max[r(l)], where r(l) is the cross correlation of the gait parameter step series and instantaneous cue velocity series as a function of the time lag l of cue velocity relative to the appropriate step cycle event associated with the gait parameter. r(l) is calculated:

$$r\left(l\right)=\frac{1}{N}{\displaystyle {\sum }_{l=0}^N}{v}_{\text{cue}}\left(t_i+l\right)\times p\left(i\right),$$

where i is a step and N is the total number of steps, p(i) represents the step-cycle parameter (step length, step duration time, and single-support or double-support duration) value of step i, v_cue(t_i + l) is the cue speed at time l relative to t_i, a specific time during step-cycle i. For the step length/cue velocity correlation t_i is the heel-strike (double-support initiation); for the peak toe velocity/cue velocity correlation t_i is the moment of maximum toe velocity (shown in Fig. 2, filled circles); for correlations involving timing parameters t_i is the toe-off corresponding to the initiation of step duration and single support or conclusion of double support. To avoid biases related to changing values of v_cue, v_cue(t_i + l), and p(i) were normalized relative to the respective maxima for each l.

Gait speed control accuracy analysis.

The spatial fidelity of gait to cue position (mean error) was quantified as the mean unsigned error of trunk position relative to cue position:

$$\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}\left|\left(x_{{\text{subject}}_i}-\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}{x}_{{\text{subject}}_i}\right)-\left(x_{{\text{cue}}_i}-\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}{x}_{{\text{cue}}_i}\right)\right|,$$

where x_subject is the T3 marker position, i is a data sample (100 Hz), and N is the total samples (2,500).

The temporal fidelity was determined from the time lag associated with the maximum correlation of trunk and time-shifted cue positions. The maximum correlation was r = max[r(l)], where r(l) is the cross correlation of the trunk position and time-shifted cue position time series as a function of the time lag l of cue position. r(l) was calculated:

$$\text{r}\left(\text{l}\right)=\frac{1}{\text{N}}{\displaystyle {\sum }_{\text{i}=0}^{\text{N}}}\frac{{\text{x}}_{\text{cue}}\left(\text{i}+1\right)}{\left|{\text{x}}_{\text{cue}{}_{\text{max}}}\right|}\times \frac{{\text{x}}_{\text{subject}}\left(\text{i}\right)}{\left|{\text{x}}_{\text{subject}{}_{\max }}\right|},$$

where i is a data sample (100 Hz) and N is the total samples (2,500), x_subject(i) is the signed fore-aft deviation of the trunk at time i from its mean position, x_cuemax is the maximum distance of the cue from its mean position, x_cue(i + l) is the signed fore-aft deviation of the cue at l time steps from i from its mean position, and x_subjectmax is the maximum distance of the subject from their mean position.

The T3 marker position was used as the main accuracy outcome because L3 also reflects movement such as hip motion related to gait itself, rather than position relative to the cue. However, to determine whether fore-aft degrees of freedom of the trunk were employed in any conditions, we compared the fore-aft variation of T3 relative to L4,

$$\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}\left|\left(T{3}_i-\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}T{3}_i\right)-\left(L{4}_i-\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}L{4}_i\right)\right|,$$

across conditions. Any leaning (or change in relative T3/L4 fore-aft positions) in either direction throughout the trial increases this value. To better characterize the extent to which the control of upper body position was controlled by gait (as was the experimental intention) rather than ‘chased’ by the lower extremities, we also calculated the correlation and time lag between T3 and L4, [r(l)], where r(l) is the cross correlation of the T3 position and time-shifted L4 time series as a function of the time lag l of cue position. r(l) was calculated:

$$r\left(\text{l}\right)=\frac{1}{N}{\displaystyle {\sum }_{i=0}^N}\frac{\text{T}3\left(i+\text{l}\right)}{\left|\text{T}{3}_{\max }\right|}\times \frac{\text{L}4\left(i\right)}{\left|\text{L}{4}_{\max }\right|},$$

where i is a data sample (100 Hz) and N is the total samples (2,500), x_subject(i) is the signed fore-aft deviation of the T3 at time i from its mean position, T3_max is the maximum distance of T3 cue from its mean position, T3(i + l) is the signed fore-aft deviation of T3 at l time steps from i from its mean position, and L4_max is the maximum distance of the subject from their mean position.

Statistical analysis.

To determine whether times during the step cycle associated with various gait parameters were differed between gait parameters, as well as whether they were affected by the experimental conditions, a 2 × 4 × 3 × 4 repeated-measures ANOVA tested main effects of cue (vision, manual contact); treadmill speed (0.2, 0.4, 0.85, and 1.3 m/s); and cue speed variation (none, ±0.01, ±0.0125, and ±0.02 m/s) on all individual outcome measures and parameters (step length, step duration, double support, and peak toe velocity) on times of instantaneous cue velocity most correlated with gait parameters. 2 × 4 × 4 repeated-measures ANOVAs evaluated significance of cue (vision, manual contact); treadmill speed (0.2, 0.4, 0.85, and 1.3 m/s); and cue speed variation (none, ±0.01, ±0.0125, and ±0.02 m/s) on all individual outcome measures. Stationary-cue results were omitted from cue motion correlation analyses. Tukey post hoc tests evaluated significance of pair-wise comparisons among walking speeds and cue variations.

RESULTS

Our results showed each gait parameter was maximally correlated with different instantaneous cue velocities. These instantaneous cue velocities for the manual contact cue were systematically delayed relative to those for the visual cue. Also, the manual contact cue afforded accurate gait speed control. To set the stage for these results, we will briefly describe how gait parameters differed for different treadmill speeds and were adjusted for different cue frequencies.

Varying mean walking speed: treadmill speed effects on gait parameters.

These results are summarized in Table 1A. Faster walking during higher treadmill speeds was achieved by larger step lengths and toe velocities and briefer step durations (all main effects P < 0.001; all pairwise comparisons P ≤ 0.007). Differences in double support were significant between all four treadmill speeds (P < 0.001). Differences in single-support duration were significant only for the fastest treadmill speed (P < 0.001). Faster gait also reduced within-trial variances (SDs) of step length and step-cycle and double- and single-support durations but increased variance of peak toe velocity (all ANOVAS, pairwise P < 0.001).

Table 1.

Within-condition results and statistical significance: gait parameter means and within-trial variances

A					B					C
Treadmill Speed	Mean	SD	Within Trial SD	SD	Cue Frequency	Mean	SD	Within Trial SD	SD	Cue	Mean	SD	Within Trial SD	SD
Step length, m
0.2 m/s	0.217*	0.048	0.066*	0.009	None	0.433*	0.040	0.025*	0.005	Manual	0.394	0.039	0.041	0.005
0.4 m/s	0.311*	0.021	0.049*	0.007	0.09 Hz	0.414	0.035	0.038*	0.007	Visual	0.397	0.045	0.047	0.009
0.85 m/s	0.457*	0.061	0.035*	0.008	0.11 Hz	0.410	0.033	0.049*	0.007
1.3 m/s	0.596*	0.064	0.027*	0.005	0.18 Hz	0.405	0.027	0.065*	0.008
ANOVA	F = 513.5; P < 0.001		F = 115.0; P < 0.001		ANOVA	F = 14.55; P < 0.001		F = 196.5; P < 0.001
Pairwise	*P < 0.001		*P < 0.001		Pairwise	*P < 0.001		*P < 0.001
Peak toe velocity, m/s
0.2 m/s	0.701*	0.060	0.545*	0.060	None	1.404	0.050	0.681	0.052	Manual	1.394	0.037	0.688	0.037
0.4 m/s	1.011*	0.052	0.604*	0.051	0.09 Hz	1.383	0.045	0.673	0.040	Visual	1.392	0.044	0.687	0.042
0.85 m/s	1.629*	0.057	0.741*	0.056	0.11 Hz	1.387	0.030	0.682	0.034
1.3 m/s	2.232*	0.048	0.859*	0.040	0.18 Hz	1.398	0.037	0.712*	0.034
ANOVA	F = 2401.9; P < 0.001		F = 118.13; P < 0.001		ANOVA			F = 8.25; P < 0.001
Pairwise	*P < 0.001		*P < 0.001		Pairwise			*P < 0.001
Step-cycle duration, s
0.2 m/s	1.087	0.246	0.225	0.125	None	0.830*	0.117	0.063*	0.025	Manual	0.768	0.099	0.076	0.033
0.4 m/s	0.772	0.061	0.071	0.024	0.09 Hz	0.643*	0.143	0.071	0.032	Visual	0.767	0.094	0.092	0.047
0.85 m/s	0.565	0.076	0.026	0.004	0.11 Hz	0.702	0.071	0.093	0.042
1.3 m/s	0.490	0.052	0.016	0.004	0.18 Hz	0.739	0.114	0.110	0.071
ANOVA	F = 67.57; P < 0.001		F = 29.23; P < 0.001		ANOVA	F = 12.58; P < 0.001		F = 5.02; P = 0.006
Pairwise	*P < 0.001		*P < 0.001		Pairwise	*P < 0.001		*P = 0.006
Double support, s
0.2 m/s	0.647*	0.156	0.225*	0.111	None	0.371*	0.053	0.057*	0.019	Manual	0.330	0.048	0.075	0.030
0.4 m/s	0.325*	0.037	0.067*	0.021	0.09 Hz	0.276	0.076	0.069	0.030	Visual	0.324	0.039	0.088	0.043
0.85 m/s	0.164*	0.042	0.021*	0.003	0.11 Hz	0.283	0.042	0.091	0.037
1.3 m/s	0.110*	0.031	0.013*	0.003	0.18 Hz	0.315	0.044	0.108	0.066
ANOVA	F = 96.88; P < 0.001		F = 36.65; P < 0.001		ANOVA	F = 10.63; P < 0.001		F = 5.94; P = 0.003
Pairwise	*P < 0.001		*P < 0.001		Pairwise	*P < 0.001		*P = 0.003
Single support, s
0.2 m/s	0.467	0.105	0.063*	0.018	None	0.459	0.080	0.030	0.010	Manual	0.438	0.069	0.029*	0.007
0.4 m/s	0.473	0.092	0.035*	0.012	0.09 Hz	0.444	0.077	0.029	0.008	Visual	0.443	0.076	0.034*	0.010
0.85 m/s	0.431	0.059	0.016*	0.004	0.11 Hz	0.436	0.071	0.030	0.007	ANOVA			F = 8.61; *P = 0.015
1.3 m/s	0.392*	0.044	0.011*	0.003	0.18 Hz	0.424*	0.064	0.037	0.014
ANOVA	F = 10.85; P < 0.001		F = 85.72; P < 0.001		ANOVA	F = 10.19; P < 0.001		F = 2.96; P = 0.048
Pairwise	*P < 0.001		*P < 0.001		Pairwise	*P < 0.001

^*P values in each section/column define all of the above asterisked numbers in that section/column.

Varying within-trial walking speed: cue oscillation effects on gait parameters.

These results are summarized in Table 1B. Cue oscillation increased within-trial variances (standard deviations) of step length and step and double-support durations but not single-support duration (ANOVAs, P ≤ 0.006). Toe velocity variance was significantly increased only by the highest cue frequency (ANOVA, P < 0.001; pairwise P ≤ 0.011). Cue oscillation also increased mean step length, and decreased double-support and step duration (all ANOVA and pairwise, P < 0.001). The highest cue frequency had a shorter single-support duration (ANOVA and pairwise, P < 0.001).

Correlation/timing analysis reveals different temporal relationships to cue velocity among gait parameters.

Figure 3 shows gait parameter series relative to cue kinematics in corresponding manual contact (left) and visual cue (right) trials of one representative participant (0.4 m/s treadmill speed; 0.18-Hz cue oscillation). The peak toe velocity series (Fig. 3A, filled circles on the solid and dashed lines) follows the most correlated set of instantaneous cue velocities (unfilled circles on the filled dotted line) by ∼0.3 s. However, the timings of cues most correlated to other gait parameters differ across cue modalities. The instantaneous manual contact cue velocity series most correlated to step lengths (Fig. 3B, left, unfilled circles on dotted line) precedes the heel-strike associated with that step length (Fig. 3B, left, filled circles), while the most correlated visual cue velocity series (Fig. 3B, right, unfilled circles on dotted line) nearly coincides with the corresponding heel-strike (Fig. 3B, right, filled circles). Similarly, the instantaneous manual contact cue velocity series most correlated to double-support durations (Fig. 3C, left, unfilled circles on dotted line) precedes the toe-off associated with that double-support duration (Fig. 3C, left, filled circles), while the most correlated visual cue velocity series (Fig. 3C, right, unfilled circles on dotted line) nearly coincides with the corresponding toe-off (Fig. 3C, right, filled circles). This reflects a trend across participants of most-correlated manual contact cue velocity series occurring earlier than the visual cue velocities (ANOVA P = 0.034). [See Table 2 for means ± SD within conditions of cue modality (A), treadmill speed (B), and cue frequency (C) and Fig. 4 for a timeline of step-cycle events and temporal relationships within cue modalities.]

Fig. 3.

Cue velocity and gait parameter time series of one representative participant. These data come from trials with 0.4 m/s treadmill speed and 0.18-Hz haptic (left) and visual (right) cue oscillation. The cue velocity (dotted lines) is shown in the same time scale, but on amplitude scales normalized to peak toe velocity (A), step length (B), or double-support duration (C). Step lengths (C, filled circles) are plotted at time of the associated heel strike. Double support durations (D, filled circles) are plotted at the time of the toe-off concluding double support.

Table 2.

Within-condition results and statistical significance: gait parameter-cue velocity correlations and timings

A					B					C
Treadmill Speed	Correlation with Cue Velocity	SD	Time of Most Correlated Instantaneous Cue Velocity, s	SD	Cue Frequency	Correlation with Cue Velocity	SD	Time of Most Correlated Instantaneous Cue Velocity, s	SD	Cue	Correlation with Cue Velocity	SD	Time of Most Correlated Instantaneous Cue Velocity, s	SD
Step length (i)
0.2 m/s	0.816	0.044	−0.247*	0.143	None	N/A		N/A		Manual	0.781	0.040	−0.066*	0.155
0.4 m/s	0.798	0.045	−0.142	0.177	0.09 Hz	0.698*	0.051	−0.199	0.141	Visual	0.784	0.044	−0.202*	0.143
0.85 m/s	0.790	0.038	−0.095	0.156	0.11 Hz	0.795*	0.047	−0.123	0.200	ANOVA			F = 6.26; *P = 0.031
1.3 m/s	0.727*	0.087	−0.052	0.124	0.18 Hz	0.855*	0.023	−0.080	0.106
ANOVA	F = 6.07; P = 0.002		F = 6.75; P = 0.001		ANOVA	F = 77.29; P < 0.001
Pairwise	*P = 0.002		*P = 0.001		Pairwise	*P < 0.001
Peak toe velocity (ii)
0.2 m/s	0.847	0.055	0.079	0.160	None	N/A		N/A		Manual	0.850	0.026	0.129*	0.068
0.4 m/s	0.844	0.039	0.053	0.184	0.09 Hz	0.758*	0.051	0.039	0.142	Visual	0.828	0.048	0.017*	0.147
0.85 m/s	0.846	0.044	0.113	0.095	0.11 Hz	0.856*	0.036	0.095	0.139	ANOVA			F = 10.24; *P = 0.009
1.3 m/s	0.819	0.059	0.047	0.095	0.18 Hz	0.903*	0.020	0.085	0.080
					ANOVA	F = 66.82; P < 0.001
					Pairwise	*P < 0.001
Step-cycle duration (iii)
0.2 m/s	−0.534	0.163	0.107	0.409	None	N/A		N/A		Manual	−0.624*	0.110	0.155	0.258
0.4 m/s	−0.582	0.190	−0.081	0.293	0.09 Hz	−0.490*	0.125	0.064	0.379	Visual	−0.523*	0.115	0.045	0.273
0.85 m/s	−0.572	0.142	0.192	0.262	0.11 Hz	−0.602	0.101	0.136	0.238	ANOVA	F = 8.86; *P = 0.014
1.3 m/s	−0.607	0.110	0.184	0.261	0.18 Hz	−0.628	0.106	0.102	0.198
					ANOVA	F = 4.44; P = 0.025
					Pairwise	*P = 0.025
Double support (iii)
0.2 m/s	−0.647	0.152	0.238	0.313	None	N/A		N/A		Manual	−0.692*	0.041	0.209	0.198
0.4 m/s	−0.691	0.074	0.069	0.196	0.09 Hz	−0.539*	0.086	0.010	0.221	Visual	−0.603*	0.064	0.060	0.211
0.85 m/s	−0.641	0.060	0.021	0.238	0.11 Hz	−0.667*	0.050	0.200	0.172	ANOVA	F = 19.93; *P = 0.001
1.3 m/s	−0.611	0.072	0.210	0.194	0.18 Hz	−0.736*	0.037	0.194	0.174
					ANOVA	F = 34.28
						P < 0.001
					Pairwise	*P < 0.001
Single support (iii)
0.2 m/s	0.441*	0.295	0.026	0.259	None	N/A		N/A		Manual	0.054	0.288	0.013	0.218
0.4 m/s	0.177*	0.310	0.045	0.303	0.09 Hz	0.077	0.219	−0.032	0.274	Visual	0.098	0.272	−0.110	0.174
0.85 m/s	−0.135*	0.310	−0.036	0.370	0.11 Hz	0.073	0.271	0.021	0.240	0.11 Hz	0.073	0.271	0.021	0.240
1.3 m/s	−0.178	0.318	−0.229	0.176	0.18 Hz	0.078	0.292	−0.135	0.248	0.18 Hz	0.078	0.292	−0.135	0.248
ANOVA	F = 23.61; P = 0.001
Pairwise	*P < 0.001

i) Time of instantaneous cue velocity preceding heel strike; negative value corresponds to time following heel strike; ii) time of instantaneous cue velocity preceding peak toe velocity; iii) time of instantaneous cue velocity preceding toe-off.

^*P values in each section/column define all of the above asterisked numbers in that section/column.

Fig. 4.

Timing of step-cycle events and gait parameter cues. Vertical lines denote times of events plotted along a horizontal time line. SD error bars of these events are horizontal. Events to the left precede events to the right. Grey area represents single-support phase. Step-cycle events [peak toe velocity (t = 0 s), heel-strike, and subsequent toe-off] are shown in black. Timing of instantaneous cue velocities shown are associated with peak toe velocity (relative to previous toe-off, “X”), step length (relative to previous heel-strike, “○”), double-support (relative to following toe-off, “●”), and step duration (relative to following toe-off, “§”). Horizontal error bars are SD among participants.

Grand mean correlations of cue velocity were strong with step length (r = 0.78 ± 0.03), peak toe velocity (r = 0.84 ± 0.03), and step-cycle (r = −0.56 ± 0.08) and double-support (r = −0.65 ± 0.04) durations but not with the single-support duration (r = 0.076 ± 0.278). The gait parameters correlated maximally with cue velocity at different times (ANOVA P < 0.001), during the step cycle. Peak toe velocity correlated maximally with a cue velocity preceding it by 0.07 ± 0.10 s (Fig. 4, “X”), near the start of single support. The instantaneous cue velocities maximally correlated with step length, double-support, and step durations all occur during double support. Step length correlated with a cue velocity at 0.13 ± 0.10 s (Fig. 4, “○”) following heel-strike-that is, during the double-support phase subsequent to foot placement defining the step length. The mean instantaneous cue velocities most correlated to step duration (0.10 ± 0.25 s before toe-off; Fig. 4, “§”), and double-support duration (0.13 ± 0.15 s before toe-off; Fig. 4, “●”) precede the toe-off concluding these durations.

Step and double-support durations were more highly correlated to the manual contact cue velocity than visual cue velocity.

Step duration was more correlated with manual contact cue velocity (r = −0.624 ± 0.110) than visual cue velocity (r = −0.523 ± 0.115, P = 0.014; Table 2C; note the sinusoidal patterns in step-cycle time series in Fig. 3). Likewise, the duration of the double-support portion of the step cycle correlated more highly with manual contact cue velocity (r = −0.692 ± 0.041) than visual cue velocity(r = −0.603 ± 0.064; P = 0.001).

Timings of cue velocity which were most correlated to step duration (0.155s ± 0.258 vs. 0.045s ± 0.273 before toe-off) did not differ significantly between feedback modes (P = 0.19; Table 2C; Fig. 4, “§”). The timing of velocity most correlated to double-support durations did not differ significantly either (0.209 ± 0.198 s vs. 0.060 ± 0.211 s before toe-off; P = 0.12; Fig. 4, “●”).

Step lengths and peak toe velocities correlated maximally with a greater temporal lead of the manual contact cue velocity than visual cue velocity.

The correlations of cue speed to step length with the visual cue (r = 0.784 ± 0.044) and with the manual contact cue (r = 0.781 ± 0.040) did not differ significantly (P = 0.686; Table 2C). The same was true for toe velocity-cue speed correlations (r = 0.850 ± 0.026 vs. 0.828 ± 0.048; P = 0.183). However, the instantaneous cue speed with which these parameters were most correlated occurred earlier with the manual contact cue than the visual cue. Step length was maximally correlated to an earlier instantaneous manual contact cue velocity (0.07 ± 0.16 s following heel-strike) than visual cue velocity (0.20 ± 0.14 s; P = 0.031; Fig. 4, “○”). The manual contact cue velocity most correlated to peak toe velocity also occurred earlier (0.13 ± 0.07 s preceding the peak toe velocity; Fig. 4) than the manual contact cue velocity (0.02 ± 0.15 s; P = 0.009; Fig. 4, “X”). Timings for cue times associated with step and double-support durations followed the same trend between cue modalities but were not statistically significant.

Effects of cue frequency and gait speed on gait speed control strategy.

The fastest treadmill speed had the lowest correlation of step length to the cue velocity (r = 0.727 ± 0.087; P ≤ 0.041 pairwise; Table 2A). The slowest treadmill speed had the latest occurring cue velocity most correlated to step length (0.25 ± 0.14 s following heel-strike; P ≤ 0.023 pairwise). Single-support duration, generally not very correlated to cue velocity, was more correlated during the lowest treadmill speed (r = 0.441 ± 0.295; ANOVA P < 0.001; pairwise P ≤ 0.019). Unusually, this correlation at this slow treadmill speed was positive, meaning that at 0.2 m/s, speeding up corresponded with longer duration single support.

Higher cue frequencies increased correlations of step length, peak toe velocity, and step and double-support durations to cue velocity (all ANOVA P ≤ 0.025; Table 2B). No time lags differed between cue frequencies.

Gait matched the manual contact cue motion better than visual cue motion.

The cue mode did not significantly affect means and within-trial variances of gait parameters (step length, peak toe velocity, and step-cycle, double-, and single-support durations; P > 0.05; Table 1C). However, the trunk motion matched the manual contact cue better than the visual cue. Figure 5 shows superimposed fore-aft position time series of the cue (dotted) and trunk (solid) of one representative subject for all conditions. Accuracy outcome means and standard deviations across subjects of all individual conditions are plotted in Fig. 6. Means, standard deviations and significant effects on tracking accuracy outcomes are summarized in Table 3, A and B. The mean cue-tracking error for all subjects was smaller with the manual contact cue (1.96 ± 0.278 cm) than with visual feedback (3.09 ± 0.555 cm; P < 0.001; Fig. 6A and Table 3A, right). Cue oscillations increased tracking error (P ≤ 0.001 pairwise). Visual cue motion increased the mean tracking error (+1.3 cm) more than manual contact cue motion (+0.3 cm) relative to stationary cue counterparts (feedback × cue P < 0.001).

Fig. 6.

Summary results of cue-tracking error and trunk orientation change. Within-trial mean tracking error (A), trunk/cue correlation (B), and associated time lags (C); within-trial mean T3-L4 change (D), T3/L4 correlation (E), and associated time lags (F). Histogram means and error bars (standard deviation) are across subjects (n = 13). Filled bars are haptic feedback; unfilled bars are visual feedback. Positive lag means cue (C) or T3 (F) leads. Trunk/cue correlations and time lags were not calculated for the no-oscillation condition (C and D) because cue amplitude equaled zero. T3/L4 time lags are omitted for the no-oscillation condition (F) since corresponding correlations are very small.

Table 3.

Within-condition results and statistical significance: tracking accuracy

A. Tracking Error, cm
Treadmill Speed	Mean	SD	Cue Frequency	Mean	SD	Cue	Mean	SD
0.2 m/s	2.781	0.790	None	1.824*	0.415	Manual	1.962*	0.278
0.4 m/s	2.354	0.462	0.09 Hz	2.646	0.489	Visual	3.088*	0.555
0.85 m/s	2.374	0.309	0.11 Hz	2.769	0.589	ANOVA	F = 58.00; *P < 0.001
1.3 m/s	2.592	0.543	0.18 Hz	2.861	0.469
			ANOVA	F = 16.59; P < 0.001
			Pairwise	*P < 0.001

B. Trunk/Cue Correlation and Time Lag
Treadmill Speed	Correlation	SD	Lag, s	SD	Cue Frequency	Correlation	SD	Lag, s	SD	Cue	correlation	SD	Lag, s	SD
0.2 m/s	0.953	0.015	−0.005	0.064	None	N/A		N/A		Manual	0.968*	0.010	0.015	0.057
0.4 m/s	0.957	0.020	−0.013	0.076	0.09 Hz	0.952*	0.010	−0.041*	0.075	Visual	0.946*	0.019	−0.017	0.097
0.85 m/s	0.961	0.013	0.000	0.072	0.11 Hz	0.962	0.013	0.009	0.059	ANOVA	F = 14.97; *P = 0.003
1.3 m/s	0.956	0.020	0.014	0.110	0.18 Hz	0.956	0.016	0.029	0.064
					ANOVA	F = 4.35; P = 0.027		F = 12.82; P < 0.001
					Pairwise	*P = 0.027		*P < 0.001

C. T3-L4 Fore-Aft Change, cm
Treadmill Speed	Mean	SD	Cue Frequency	Mean	SD	Cue	Mean	SD
0.2 m/s	1.368	0.189	None	1.331	0.186	Manual	1.266	0.045
0.4 m/s	1.359*	0.156	0.09 Hz	1.297	0.133	Visual	1.355	0.050
0.85 m/s	1.239*	0.133	0.11 Hz	1.302	0.106	ANOVA	F = 4.65; P = 0.057
1.3 m/s	1.277	0.156	0.18 Hz	1.314	0.146
ANOVA	F = 3.47; P = 0.028
Pairwise	*P < 0.035

D. T3-L4 Fore-Aft Correlation and Time Lag
Treadmill Speed	Correlation	SD	Lag, s	SD	Cue Frequency	Correlation	SD	Lag, s	SD	Cue	Correlation	SD	Lag, s	SD
0.2 m/s	0.900	0.029	0.009	0.071	None	0.680*	0.055	0.015	0.127	Manual	0.903	0.019	0.003	0.047
0.4 m/s	0.902	0.016	0.003	0.063	0.09 Hz	0.983	0.005	0.004	0.009	Vision	0.915	0.023	0.012	0.027
0.85 m/s	0.914	0.018	0.007	0.007	0.11 Hz	0.988	0.003	0.005	0.006
1.3 m/s	0.921	0.030	0.011	0.019	0.18 Hz	0.986	0.006	0.006	0.007
					ANOVA	F = 354; P < 0.001
					Pairwise	*P < 0.001

^*P values in each section/column define all of the above asterisked numbers in that section/column.

Trunk motion was very highly correlated with cue motion (grand mean r = 0.957 ± 0.014; Fig. 6B), with no significant main effects of feedback modality, treadmill, or cue frequency (all P > 0.05; Table 3B). However, trunk/cue timing relationships differed among cue frequencies. For higher frequency cue motion, trunk motion lagged cue motion 0.009 ± 0.059 s (0.11 Hz) and 0.029 ± 0.064 s (0.18 Hz; Fig. 6C and Table 3B, middle). However, the trunk preceded the cue during the lowest-frequency cue oscillation (0.041 ± 0.075 s, ANOVA P < 0.001, P < 0.005 all pairwise).

With manual contact participants changed their upper body orientation slightly less than with the visual cue (manual contact: 1.266 ± 0.045 vs. vision: 1.355 ± 0.051 cm; P = 0.057; Fig. 6D and Table 3C). T3 and L4 motion was highly correlated (r > 0.98) during cue oscillations but not in the absence of cue oscillation (r = 0.182; P < 0.001; Fig. 6E and Table 3D, middle). The T3/L4 correlation was not significantly affected by the cue modality (P = 0.193). T3 changed position ahead of L4 during manual contact by 0.003 ± 0.047 s, but the lead was not significantly greater than with vision (0.012 ± 0.027 s; P = 0.41; Fig. 6F and Table 3D, right). There were no significant main effects in the timing of correlated T3 and L4 motion (all F < 0.8; P > 0.4).

Treadmill speed did not significantly affect tracking error (P = 0.147, Table 1A) or trunk/cue correlations (P = 0.588) or timing (P = 0.787, both Table 1B). Treadmill speed affected leaning, measured by the mean fore-aft T3-L4 relative position change (ANOVA P = 0.028, Table 3C) but not systematically. The highest and lowest mean lean magnitudes were in the intermediate treadmill speeds. Treadmill speed did not significantly affect T3-T4 correlations (P = 0.104) or timing (P = 0.966, both Table 3D, left).

DISCUSSION

Gait parameters are adjusted variously to control gait speed in anticipation of cue velocity changes.

The near-synchrony of ongoing cue and trunk position demonstrates that similar anticipatory control is engaged to match either manual contact or visual cues. The tendency for the trunk reversing direction before the cue during lower frequency oscillation likely represents judging the cue to reverse direction before it does during excursion extremes when cue motion is near or below perceptual thresholds. The lack of independence observed in the high correlations and brief time lags (<0.05 s) between markers at T3 and L4 vertebrae levels suggest participants matched cue motion mainly adjusting gait parameters rather than using the degrees of freedom of the upper body. This suggests that while some tracking error may be due to trunk control, more of it likely resides with gait control.

Our cue timing analysis pointed to different cue times for different gait parameters. The instantaneous cue velocity most correlated with step length occurs during double support following the heel-strike defining the step length. The cue velocity corresponding to peak toe velocity occurs during single support. The cue velocity associated with the toe-off concluding double-support and step durations happens during double support at a brief latency before the toe-off on average. Conversely stated, step length precedes its associated cue velocity, and temporal gait parameters (peak toe velocity and double-support and step durations) follow the cue velocity at respective brief intervals.

The different times at which the cue velocity correlates maximally with specific gait parameters may reflect variable weighting of sensory information during stepping to optimize control (Logan et al. 2010; Patla and Vickers 1997). This reweighting may be related to variations in feedback sensitivity. For example, step length and step duration may be adjusted based on cue velocities during double support because of better proprioception when the body is most stable (Floyd et al. 2014).

Let us consider these timing relationships in greater detail. The motion of the foot during gait is principally due to the combined effects of iliopsoas muscles (swinging leg hip flexion active during the 1st 25% of single support), gluteus maximus (hip extension supporting leg), and momentum from the previous step (Perry and Burnfield 1995). The peak toe velocity represents a late temporal estimate of its underlying neural activity. Therefore, the brief interval of peak toe velocity following its correlated instantaneous cue velocity suggests an anticipatory control. Although the speed of the swing leg during single-support duration is highly related to walking speed, the single-support duration is not (r < 0.1).

The heel-strike defining step length precedes its most correlated cue velocity occurring during the immediate double support. In other words, step length is determined during single support to match the cue in the subsequent double support. This places the step length cue time later than toe velocity cue, suggesting independent control of these two important parameters underlying gait speed. The very high correlations of these parameters to their anticipated cue velocities demonstrate the accuracy of this anticipatory control, and the low variance among participants shows the prevalence of this strategy.

The relationship of double-support duration with cue velocity is weaker and more temporally variable among participants than those of step length and toe velocity. The preceding anticipatory foot placement may be so accurate that adjusting double-support duration is less necessary, resulting in the weaker correlation with cue speed. This weaker relationship affected our timing results: some timing means were skewed by time lags associated with lower correlations, particularly on intermediate (more “normal”) treadmill speeds. Nevertheless, this shows that a strategy of rushing or delaying toe-offs to adjust gait speed is a secondary one, more prevalent during unusual walking speeds. The timing on average, like that of peak toe velocity, is a very short delay (∼0.1 s) relative to toe-off “defining” the double-support duration. The coincidence of the double-support duration cue and preswing muscle activity of the adductor longus, gracilis, sartorius, extensor digitorum, and hallucis longus(es) related to toe-off precludes a cue/double-support duration feedback relationship (Perry and Burnfield 1995).

The overall step duration's inverse relationship with cue velocity is driven mainly by the double-support phase. This is clearest during the slowest treadmill speed when single-support duration is positively (not inversely as usual) correlated with cue velocity. Although single-support duration was not correlated with cue velocity at other treadmill speeds, the briefer single-support observed during the highest cue frequency conditions may also represent a ubiquitous strategy in which single support is reduced to “catch up.” The positive single-support/cue velocity correlation emerges at the slowest treadmill speed because the anterior momentum, which propels the body forward at normal gait speeds, is so small at 0.2 m/s that higher speed achieved through longer step length requires more time. However, most other cue/gait temporal relationships of this slow treadmill speed aligned with the other speeds. One exception was the greater lead of cue velocity to step length, which clearly reflects an anticipatory control strategy.

Generalization of cue velocity/gait parameter timing relationships across cue frequencies and gait speeds suggests a common control scheme and validates control parameters.

The general timing pattern of instantaneous cue velocities associated with different gait parameters was preserved across manual contact and visual cues, with an en masse delay of gait parameters relative to cues in the manual contact condition. Some studies have characterized a connection of stride length and cadence (Egerton et al. 2011), while others have found evidence of their independent control (Klarner et al. 2013; Malone et al. 2012; Prokop et al. 1997). The preservation of relative cue timings across cue sensory modalities argues for a single coordination scheme within which temporally distinct gait parameters associate with different discrete times of the same environmental cue.

Significantly, the null effects of gait speed on cue/gait parameter timing relationships validate our analysis with regard to the use of time rather than percentage of gait cycle. As mentioned above, step-cycle phases do not scale in time: step-duration changes have little effect on single-support duration but a significant effect on percentages of single- and double-support phases (Murray et al. 1966). Likewise, the cue/gait parameter timing relationships would differ more between gait speeds in the phase domain than the time domain. This suggests that the control of these parameters is time based rather than phase based. We note a trend for larger cue/step length timing relationships for slower speeds, suggesting some phase relationship. However, range of cue/heel-strike intervals (0.25-0.05 s) is ∼40% of the range of step durations among gait speeds (1–0.5 s, see Table 1B), so this relationship cannot be phase invariant either.

By similar reasoning, null effects of cue frequency on cue/gait parameter timing relationships suggest that velocity is the cue parameter of interest, rather than position or acceleration. The velocity is a 90° phase-advanced equivalent of the position of our sinusoidal oscillating cues. Therefore, it may be valid to conclude that the cue position drives the various gait parameters with timing relationships adjusted to the 90° phase position-velocity difference. However, a 90° phase shift corresponds to different times for the cue frequencies, again leading to nonuniform cue/gait parameter temporal relationships across cue frequency conditions.

Conventionally, the heel-strike initiating double-support is considered the start of the gait cycle (e.g. Perry and Burnfield 1995). Yet, Ivanenko et al. (2004) have argued that the gait cycle origin is the propulsion rather than the conventional heel-strike, based on the consistent timing relative to toe-off of major muscle activation patterns relative to single-support onset. Our findings agree with the latter conception from a sensory-goal standpoint. The cue velocities during double-support phase are associated with gait events occurring during that phase and the preceding single-support phase, i.e., stride length (as shown in Fig. 4). Therefore, the external cues associated with gait events of single-support phase and the subsequent double-support phase are more similar temporally than those of double- and subsequent single-support phases, as the conventional gait cycle conception would have it.

The cues in the present treadmill paradigm signal error similarly to static balance cues.

The prominence of cues occurring during double support in the control of gait suggests that the use of subject-centered cues to match a desired velocity may be similar to the use of such cues to achieve zero velocity during balance. Holden et al. (1994) first demonstrated how haptic cues, in the form of nonsupportive manual contact with stationary surfaces, afford skin deformation integrated with body proprioception to stabilize balance. Haptic cues have shown an advantage over visual cues in balance control such that center-of-pressure sway during static stance was ∼30% lower with isolated haptic cues than either isolated visual or vestibular cues (Lackner et al. 1999). This advantage may be due to lower haptic sensory thresholds and/or briefer neural processing latencies.

Our experimental treadmill task is analogous to the static balance paradigms in that the task goal is to maintain a constant position relative to an external cue and any change in cue feedback signified error. The difference between this type of visual feedback from more typical “flow” feedback gathered while moving relative to stationary surroundings might contribute to the poorer visual control compared with manual contact. Typical visual flow feedback, subject to integration with proprioception, also has highly specific relationships to individual gait parameters underlying speed (Prokop et al. 1997). The manual contact cue in the absence of vision offers no analogous surrounding contextual information. It remains an open question whether our results would generalize to “flow” type haptic cues (e.g. running fingertips along a banister) or whether haptic cue discrimination is possible at regular walking speeds.

Higher accuracy with manual contact reflects advantageous cue/gait parameter timing.

Other than whether or not cues were oscillating, the only significant main effect on tracking accuracy was the cue modality. Smaller errors with manual contact cues demonstrate an advantage of manual contact over purely visual guidance. This advantage of manual cues may be related to the fact that the manual contact gait parameter cue timings occurred later in the step cycle than the visual cues. Manual contact cue/gait parameter timing relationships may be optimal as determined by mechanical links or related to haptic feedback control latencies. Manual contact affords a stabilizing mechanical linkage between the trunk and the cue. Although the arm motion to account for tracking errors is well within the constraining limits of arm position, participants may have employed this linkage as a constraint though which they may achieve spatial stability relative to the cue. The correspondence of earlier manual contact cue velocity with gait parameters may therefore represent an optimized timing pattern to track the cue with the benefit of the constraining touching arm link/constraint. The temporally advanced timing pattern of gait parameters with visual control may represent evidence of a tendency to use visual cues to plan further ahead.

The touching arm also affords tactile information and proprioceptive information, which have briefer processing latencies compared with vision (cf. Rabin et al. 2006). Such haptic sensory cues engage postural responses to stabilize balance more quickly than visual cues (cf. Rabin et al. 2006). Therefore, we would expect the greater accuracy of gait control with haptic cues to be reflected by briefer latencies of gait to manual contact cue velocity. However, the response to the anticipated haptic cue seems to take longer. The very brief delays of gait parameters and their associated instantaneous visual cue velocities suggest anticipatory relationships. However, the peak-toe velocity and toe-off concluding double support follow their associated haptic cues at 0.13 ± 0.07 and 0.21 ± 0.20 s, respectively, which may encroach on the range of functional feedback control. Such control may be intermittent rather than continuous (Loram et al. 2011) and apply to a subset of all steps. For example, feedback-driven EMG responses to mechanical perturbations during gait take ∼0.10–0.12 s to occur (Field-Fote and Dietz 2007). It is possible haptic processing is brief enough (cf. Rabin et al. 2006; Loram et al. 2011) to take advantage of, whereas visual feedback may take so long that is not as useful to control changing gait speed. The feedback relationship also accounts for the haptic cue's stronger relationship of step and double-support timing.

The similar temporal cue-gait relationships across haptic and visual cues suggest a common process across cue conditions by which gait speed is controlled. Yet, gait parameter changes correlate with earlier mechanical contact cues than visual cues. Our findings of feedback control of timing parameters, but not step length (with the appropriate cue), are supported by adaptation studies. Ogawa et al. (2014) reported split-belt gait adaptation aftereffects in ankle EMG at heel-strike (associated with step length) but not in the ground reaction force at toe-off (concluding double support). This is consistent with our findings of control of step length anticipating cue speed, and feedback control of double support. The cue/gait relationships we have identified during continuous gait speed adjustment may represent a general case test of the control mechanisms underlying adaptation aftereffects observed following split-belt treadmill adaptation.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

E.R. conception and design of research; E.R. and P.S. performed experiments; E.R. and P.S. analyzed data; E.R. and W.W. interpreted results of experiments; E.R. prepared figures; E.R. drafted manuscript; E.R. and W.W. edited and revised manuscript; E.R. approved final version of manuscript.