INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

It is widely accepted that the coordination of multiple degrees of freedom involved in locomotion is constrained by the CNS through a small number of behavioral units (Bizzi et al. 2000; Grillner 1981; Lacquaniti et al. 1999; Pearson 1993; Rossignol 1996; Winter 1995). According to an influential scheme, different forms of locomotion are produced by means of flexible combination of unit burst generators, each unit controlling a set of synergistic muscles acting around a particular joint under the modulatory influence of peripheral and supraspinal inputs (Grillner 1981; Orlovsky et al. 1999; Pearson et al. 1998; Rossignol 1996). Local control of individual joints is combined to generate appropriate patterns for the whole limb (as well as for inter-limb coordination). There is neurophysiological evidence that whole limb parameters are encoded at several different levels of integration in the CNS. Thus in lower vertebrates, control of the foot appears to be based on a few spinal inter-neuronal modules that can be flexibly combined to generate appropriate motor patterns (Bizzi et al. 2000). Also, position and direction of foot movement are encoded in DSCT neurons of the anesthetized cat spinal cord, independent of the specific combination of joint angles and level of muscle activity (Bosco and Poppele 2001). Accurate foot placement in freely walking cats is controlled by the motor cortex possibly via the propriospinal system (Beloozerova and Sirota 1993; Drew 1993).

For human locomotion, Winter (1980, 1989, 1991) has demonstrated the existence of a law of kinetic co-variance that involves a tradeoff between the hip and knee torques, such that the variability of their sum is less than the variability of each joint torque taken separately. There is also ample evidence for kinematic control of the lower limbs. Kinematic control is reflected at the level of single joints, limb segments, and endpoint. Thus the reproducibility of the patterns of changes of both joint angles and limb segment angles has been documented for different walking speeds (see Borghese et al. 1996; Mah et al. 1994; Murray 1967; Winter 1983, 1991). Moreover, it has been shown that there exists a specific law of inter-segmental coordination: the temporal changes of the elevation angles of lower limb segments co-vary along a plane at different speeds (Bianchi et al. 1998; Borghese et al. 1996), during both forward and backward walking (Grasso et al. 1998), and when walking with different bent postures (Grasso et al. 2000b). A related planar co-variation has been found in the case of whole-body motion (Hasan and Thomas 1999). As for the control of the limb endpoint, the foot, Winter (1992) analyzed the trajectory of the heel and toe at natural walking cadence. He found that toe clearance is consistently small, with low inter-trial variability and heel-contact velocity is negligible vertically and small horizontally. Intra- and inter-subject variability of horizontal and vertical motion of the heel and toe are quite low. Sensitivity analysis revealed that small angular changes at each lower limb joint can account independently for toe clearance variability, whereas velocity changes at the hip and knee most likely account for altering heel-contact velocity.

Here we address the issue of the role of load-related contact forces at the ground in the process of limb trajectory formation and control. Ground contact forces reflect the net vertical and shear forces acting on the contact surface and result from the sum of the mass-acceleration products of all body segments while the foot is in contact with ground (Winter 1991). Contact forces represent high-order control parameters for both posture (Horak and Macpherson 1996; Lacquaniti and Maioli 1994; Macpherson 1988) and locomotion (Alexander 1991; Winter 1991, 1995). In this paper, we shall focus mainly on the net vertical forces that reflect gravitational and inertial loading forces. It is known that load plays a crucial role in shaping patterned motor output during stepping (Duysens et al. 2000; Harkema et al. 1997; Scott and Winter 1993). Thus transient loading of the limb enhances the activity in antigravity muscles during stance and delays the onset of the next flexion (Duysens and Pearson 1980). Also, it has been demonstrated that positive force feedback, in conjunction with negative length feedback and intrinsic muscle properties, can provide effective, stable load compensation (Prochazka et al. 1997a,b). Load feedback can act either directly through spinal reflex pathways or indirectly through the locomotor generator network involved in the production of extensor activity (Pearson 1993, 1995; Zehr and Stein 1999). In daily life, the net vertical load may decrease suddenly (as when we walk grasping a hand-rail or supporting ourselves on somebody's else arm) or more progressively (as when we lose weight over days or weeks). It is important to understand the mechanisms of adaptation to both sudden changes and longer-term changes of load.

Simulated reduced gravity represents a well-controlled technique to study body-weight unloading. This affects walking mechanics and energetics in a complex manner (see Davis and Cavanagh 1993; Edgerton and Roy 2000). The mean amplitude of activity of ankle extensor muscles (Dietz and Colombo 1996; Finch et al. 1991; Harkema et al. 1997), vertical contact forces (Flynn et al. 1997; Griffin et al. 1999), mechanical work (Cavagna et al. 2000; Griffin et al. 1999; Newman et al. 1994), and overall metabolic energy consumption (Farley and MacMahon 1992; Newman et al. 1994) all decrease proportionally to unloading. Also, the lower the gravity value, the lower is the energetically optimal speed, that is, the speed at which there is a minimum energy expenditure (Cavagna et al. 2000; Griffin et al. 1999), and the more restricted is the range of walking speeds, that is, the speed values at which walking is preferable to running from an energetic standpoint (Cavagna et al. 2000; Davis and Cavanagh 1993; Kram et al. 1997; Newman et al. 1994).

In striking contrast with the wide changes of the kinetic parameters, a number of global kinematic parameters (stride frequency and length, stance and swing relative durations, vertical displacement of the center of body mass) have been shown to change to only a very limited extent (<10%) over a fourfold reduction of gravity (Donelan and Kram 1997; Farley and MacMahon 1992; Finch et al. 1991; Griffin et al. 1999; Newman et al. 1994). However, these previous studies did not investigate limb kinematics in sufficient detail to address the question of how the kinematic control adapts to reduced gravity.

Here we studied the changes of vertical contact forces, lower limb kinematics, and electrical muscle activity (EMG) at different speeds and gravitational loads. Healthy subjects were asked to walk on a motorized treadmill while the percentage of body weight unloaded (BWS) was modified in steps between 0 and 100% (simulating gravity levels between 1 and 0 g) by means of a well-characterized unloading system (Gazzani et al. 2000). We found that changing BWS between 0 and 95% resulted in drastic changes of kinetic parameters (contact forces and EMG patterns) but in limited changes of the inter-segmental kinematic coordination. In particular, the planar co-variation was always obeyed, and the variability of foot trajectory was systematically small, except at 100% BWS when subjects stepped in air with no ground contact. In the latter case, variability increased considerably and were restored to lower values when minimal surrogate contact forces were provided during the "stance" phase. Therefore we conclude that minimal vertical contact forces are sufficient for accurate foot trajectory control.

The present findings also are relevant as normative data to the field of locomotor training of patients with spinal cord injury (SCI) because protocols involving various types of BWS are currently being used in several centers for this purpose (e.g., Barbeau and Rossignol 1994; Dietz et al. 1995; Harkema et al. 1997; Wernig et al. 1995). Preliminary results have been reported in abstract form (Grasso et al. 1999, 2000a; Ivanenko et al. 2000b).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

Eight healthy subjects [4 males and 4 females, between 22 and 40 yr of age, 65 ± 13 (SD) kg, 1.70 ± 0.10 m] volunteered for the experiments without being paid. The studies conformed with the Declaration of Helsinki, and informed consent was obtained from all participants according to the protocol of the Ethics Committee of the Santa Lucia Institute.

Experimental setup

TREADMILL. The experiments were carried out on a treadmill at different speeds and with different BWS. For all but one experimental protocols, a treadmill (EN-MILL 3446.527, Bonte Zwolle BV, Netherlands) operating at a constant belt speed set by the control panel was used. The walking surface of this treadmill is 1.5 m long, 0.6 m wide, and 0.15 m high above the ground. For one specific protocol (position-related treadmill speed, see following text) a different treadmill (Woodway XELG 70, Germany, 2.7 m long, 0.7 m wide, 0.6 m high) was used. In this case, the instantaneous velocity of the treadmill was recorded via an optical encoder (resolution, 0.005 m/s) and controlled by a computer using a feedback from the subject's position: forward displacements from the initial position increased the treadmill velocity proportionally, whereas backward displacements decreased it. To measure subject's position, a lightweight stiff thread was attached at the waist, and it was kept in tension by a constant force (~2 N) produced by a torque motor (type JR24M4CH, ServoDisc, PMI, Commack, NY). A linear potentiometer on the motor shaft measured the changes of the subject's position (that occurred when the velocity of the subject differed from that of the treadmill belt). Position was sampled at 30 Hz with an accuracy of 2 mm, and treadmill speed was updated at the same frequency (after low-pass filtering to avoid jerks). The servo-system provides an efficient and stable control of the treadmill speed in the frequency range of 0-2 Hz. The feedback constant (G = 1.1) and treadmill acceleration (1 m s²) were selected so as to make subjects feel comfortable when they changed their walking speed.

BODY WEIGHT SUPPORT. BWS was obtained by means of a pneumatic device that applies a controlled upward force at the waist, close to the center of body mass (WARD system) (Gazzani et al. 2000). The BWS mechanism consists of a mechanical gear driven by a pneumatic cylinder, equipped with safety stops (Fig. 1). It is held in a cart that slides forward and backward over a track formed by a double steel beam, mounted in the middle of the upper side of a parallelepiped steel frame. Low-friction sliding of the mechanism ensures that only vertical forces are applied to the subject. The subject is supported in a harness, pulled upward by a steel cable connected to the piston of the pneumatic cylinder. Total vertical excursion admitted is 1 m, so that the device, without any regulation, adapts itself to the subject's height or helps raising him over ground for air-stepping (100% BWS). WARD exerts the preset unloading force independent of the position of the center of body mass, thus simulating a reduced-gravity environment. A load cell (Leane FN3030) is positioned between the harness and the cable to measure the actual delivered force. The desired unloading force (expressed as percent of subject's weight) is set on the control computer that accordingly adjusts the pressure inside the pneumatic cylinder. Preset BWS values were applied using a ramp-up (20 N/s, ~30 s to reach 100% BWS), hold (20-100 s), and ramp-down (~30 s) profile of unloading force. This procedure resulted in minimal discomfort. The overall constant error in the force applied to a subject and dynamic force fluctuations monitored by the load cell have been estimated to be <5% of body weight (Gazzani et al. 2000). Force fluctuations are essentially negligible at low BWS levels but become significant at high BWS (95-100%).

View larger version (30K): [in this window] [in a new window]   Fig. 1. Experimental setup. Subjects walked on a treadmill at different speeds and with different levels of body weight support (BWS). They were supported in a harness, pulled upwards by a cable connected to a pneumatic device that exerted the preset unloading force. Limb kinematics was recorded by monitoring the coordinates of 5 markers at the following landmarks: the midpoint between the anterior and the posterior superior iliac spine (ilium, IL), greater trochanter (GT), lateral femur epicondyle (LE), lateral malleolus (LM), and fifth metatarso-phalangeal joint (VM).

CONTACT FORCES RECORDING. Local loading of the plantar side of the foot was measured by means of a custom-made sole interposed between foot and shoe. This device consists of a matrix of 14 × 7 piezo-resistive elements, spaced along the length and width of the foot, respectively. The local vertical force sensed by each (2.5 × 2.5 mm) element was recorded at 65 Hz. The sensors were statically and dynamically calibrated with the use of a force platform (Kistler 9281B) (see Giacomozzi and Macellari 1997). Threshold and resolution were 7 and 5 N, respectively. Peak resultant forces derived from the pressure sensors were linearly correlated with the corresponding values measured by the force platform (y = 0.97x 18, r² = 0.90).

LIMB KINEMATICS RECORDING. Three-dimensional (3D) motion of selected body points was recorded at 100 Hz by means of the Optotrak system (Northern Digital, Waterloo, Ontario, Canada; ±3 SD accuracy better than 0.2 mm for x, y, z coordinates) during ~20-100 s depending on walking speed. Five infrared emitting markers were attached on the right side of the subject to the skin overlying the following landmarks: the midpoint between the anterior and the posterior superior iliac spine (ilium, IL), greater trochanter (GT), lateral femur epicondyle (LE), lateral malleolus (LM), and fifth metatarso-phalangeal joint (VM; see Fig. 1). When subjects wore shoes (see following text), VM marker was attached to the shoe after verifying the correspondence on the bare foot.

EMG RECORDING. EMG activity was recorded by means of surface electrodes from the gluteus maximus (GM), vastus lateralis (VL), rectus femoris (RF), biceps femoris (long head, BF), tibialis anterior (TA), and lateral gastrocnemius (GCL) using optic fiber BTS TELEMG. EMG signals were preamplified (100 times) at the recording site, digitized, and transmitted to the remote amplifier via 15-m optic fibers (Grasso et al. 2000b). These signals were high-pass filtered (artifact rejection was obtained by cascading 3 1st-order high-pass at 1 Hz followed by 1 1st-order high-pass at 10 Hz), low-pass filtered (200-Hz, 4-pole Bessel low-pass), and sampled at 500 Hz. Sampling of kinematic and EMG data were synchronized.

Protocols

Subjects always walked while wearing the harness even when no BWS was applied. Their arm posture differed according to the instruction. In one case (referred to as "standard arm posture") subjects were asked to place the abducted arms on horizontal rollbars located at the side of the treadmill at breast height. In another case, they were asked to keep the arms folded on the chest, and in still another case to swing them freely along the body side. When the arms were allowed to swing, proximal markers (IL and GT) were masked to the Optotrak recording system during portions of the gait cycle. Subjects walked either barefoot or with their gym shoes on. Contact forces could be measured only in the latter case. In all subjects, each experimental condition was tested in separate blocks of trials with all of the above arm postures and while walking either barefoot or with shoes. Most results illustrated in this paper refer to standard arm posture, barefoot walk, but we verified thoroughly that there was no systematic difference across testing conditions on a number of kinematic parameters. In all conditions subjects were required to look straight ahead without paying attention to the contact with the treadmill belt.

Before the recording session, subjects practiced for a few minutes in walking on the treadmill at different speeds, first without BWS and then with different levels of BWS. All subjects showed rapid adaptation to these conditions (Donelan and Kram 1997). In all experiments, each trial for a given condition included at least eight consecutive gait cycles (typically 15). Presentation order of speeds and BWS was randomized across sessions and experiments. The following two protocols were performed in all eight subjects. Four of them were tested repeatedly (3-5 different times) at distances of days or months.

SPEED EFFECTS. With BWS at 0, treadmill speed was set at one of the following values: 0.4, 0.7, 1.1, 2, 3, 5, and 6.5 km/h.

BWS EFFECTS. BWS was set at 0, 35, 50, 75, 95, or 100% of body weight. Treadmill speed was set at 0.7, 1.1, 2, 3. or 5 km/h for each BWS value. We limited the high range of tested speeds because it is known that the range of walking speeds that can be comfortably used at reduced gravity is more restricted than at 1g, and subjects may find preferable to switch to running at lower speed values than at 1 g (Kram et al. 1997). Griffin et al. (1999) have previously shown that subjects are able to maintain a walking pattern 6.3 km/h for 0 to 75% BWS.

ONE-LEGGED STEPPING. In this and the following protocol, the treadmill was shifted laterally below the subject, in such a way that one leg was on the treadmill and the other leg was outside. Four subjects were asked to perform stepping movements with the right leg only, while supporting and balancing themselves on the stationary, fully extended left leg (aside the treadmill). BWS was set at 0. Two different conditions were tested in different trials: subjects stepped on the treadmill at 3 or 5 km/h and they stepped in the air, their right foot moving just above but never contacting the treadmill (as in the case of 100% BWS).

BETWEEN-LIMBS DISSOCIATED CONTACT FORCES IN 2-LEGGED STEPPING. Two subjects, unloaded at 85 or 90% BWS, stepped at 3 km/h with the left leg making contact with the treadmill and the right leg moving freely in air (aside the treadmill).

AIR-STEPPING WITH SURROGATE CONTACT FORCES. Four subjects stepped with both legs at 100% BWS while sensing minimal contact forces at the feet during "stance." This was obtained in two different ways during the same experiment. 1) A parallelepiped (13 cm high, 25 cm long, 10.5 cm wide) of lightweight (54 g), low-stiffness (2.5 N/cm) foam rubber was taped under the subjects' feet. Fully unloaded subjects stepped in air, their height being adjusted so that the lower surface of the foam lightly touched the treadmill belt during "stance." Treadmill speed was 3 km/h. 2) Two soft blocks of foam rubber (each one 15 cm high, 76 cm long, 45 cm wide, 2 N/cm stiff) were placed one on top of the other above the treadmill belt, like a pillow. Fully unloaded subjects stepped in air, their height being adjusted so that they lightly slipped over the stationary pillow (the treadmill was off in this condition).

EFFECT OF INSTRUCTION. To test the role of the specific instruction given to the subject on limb kinematics, the following tasks were performed in two subjects. 1) Mental arithmetics. The subject, unloaded at 0, 95, or 100% BWS, stepped at 3 km/h while he counted down loudly (by subtracting 7 iteratively, starting from an arbitrary number provided by the experimenter). This condition was compared with a control condition without mental arithmetics. 2) Self-supported air-stepping. In this and the following conditions, the subject was not suspended to the WARD system, but he fully supported his weight by firmly pushing with the extended arms against the lateral rollbars (see preceding text) and raising himself above the treadmill. In one task, the subject was asked to step in the air, as if he walked on ground (as in the standard air-stepping, see above). 3) Self-supported air-stepping with ankle dorsi-flexion. In another task, the subject was asked to step in the air while trying to touch the ground with the heels (no actual contact occurred at any time because the subject was raised above the treadmill). This was obtained by dorsi-flexing the feet. 4) Self-supported air-pedaling. In this task the subject was asked to reproduce in the air the cyclical movement involved in bicycle pedaling.

WALKING WITH POSITION-RELATED TREADMILL SPEED. Seven subjects participated in this protocol. BWS was set at 0 for all tests (as in the case of the Speed effect protocol, see preceding text). The subject was placed at the center of the treadmill belt, then the initial treadmill speed was set at a value between 1.8 and 5.4 km/h (in 0.9-km/h increments) and the subject started walking. The treadmill velocity was controlled proportional to the changes of subject's position relative to the starting point (by 1.1 m/s per meter of displacement). Thus the subject could vary his walking speed at will and still remain within the central area of the treadmill.

Data analysis

The time-varying coordinates of the center of pressure and the resultant vertical component of in-shoe contact forces were derived from the force distribution map on the plantar foot. EMG data were numerically high-pass filtered at 20 Hz to remove motion artifacts, rectified and then low-pass filtered with a zero-lag Butterworth filter with cutoff at 25 Hz. Data from each trial were ensemble-averaged after time interpolation over individual gait cycles to fit a normalized 1,000-points time base. The body was modeled as an interconnected chain of rigid segments: IL-GT for the pelvis, GT-LE for the thigh, LE-LM for the shank, and LM-VM for the foot. The elevation angle of each segment in the sagittal plane corresponds to the angle between the projected segment and the vertical. These angles are positive in the forward direction (i.e., when the distal marker is located anterior to the proximal marker). Motion in the sagittal plane only was considered here, as it represents the major component in normal gait (Mah et al. 1994). The limb axis was defined as GT-LM. Gait cycle was defined as the time between two successive maxima of the limb axis elevation angle (Borghese et al. 1996). The time of maximum and minimum elevation of the limb axis roughly corresponds to heel-contact and toe-off (stance to swing transition), respectively (Borghese et al. 1996). They were used to identify "stance" and "swing" phases. In previous experiments in which a force platform (Kistler 9281B) was used to monitor the contact forces during ground walking, we found that the kinematic criterion predicts the onset and end of stance phase with an error >2% (Borghese et al. 1996; Grasso et al. 1998). This was confirmed in the present experiments using in-shoe force monitoring. Stride length was estimated as the VM horizontal excursion (x) over the gait cycle. Stance phase relative duration was expressed as percentage of the gait cycle. Speed in air-stepping was computed as 2x divided by gait cycle duration. Data from all gait cycles of each trial were pooled together for further analysis.

INTERSEGMENTAL COORDINATION. The inter-segmental coordination was evaluated in position space as previously described (Bianchi et al. 1998; Borghese et al. 1996). Briefly, the changes of the elevation angles of the thigh, shank and foot generally co-vary throughout the gait cycle during normal walking. When these angles are plotted one versus the others in a three-dimensional graph, they describe a path that can be fitted (in the least-square sense) by a plane over each gait cycle. The first two eigenvectors of the covariance matrix of the ensemble of time-varying elevation angles lie on the best-fitting plane of angular covariation, whereas the third eigenvector is the normal to the plane and defines the plane orientation. In addition, the time course of each elevation angle was expanded in Fourier series, and the percent variance accounted for by the first harmonic (fundamental component) was computed.

VARIABILITY OF KINEMATIC PARAMETERS. Variability analysis across consecutive gait cycles was performed separately over the swing phase and over the stance phase. Ensemble averages with SD of kinematic variables were computed at each point of the normalized time base.

REFERENCE FRAMES FOR ENDPOINT TRAJECTORY. Both the forward and vertical velocity of the center of body mass (CM, close to the ilium) fluctuate over a gait cycle, and its position undergoes small horizontal and vertical oscillations during treadmill locomotion (Thorstensson et al. 1984). CM vaults like an inverted pendulum over the supporting limb, decelerating in the rising phase and accelerating in the falling phase. Thus there is the issue of the reference frame in which foot trajectory should be described. We analyzed path and motion of two different markers (LM, close to the heel, and VM, close to the toes) relative to: the instantaneous IL position, and a fixed laboratory position. In the latter case, foot position was corrected by the mean horizontal position of IL cycle by cycle, to account for possible drifts of the subject along the treadmill during the recording epoch. Main results will be illustrated with this latter method.

ENDPOINT PATH. The shape of the path described by either LM or VM was compared across conditions by computing the overall horizontal and vertical excursion and the mean area of the LM and VM trajectory in the laboratory reference frame over each trial. The latter parameter was derived by computing the area of the polygon defined by the x, y coordinates of LM (or VM) marker for each cycle and by taking the average value over all cycles of each trial.

ENDPOINT VARIABILITY. LM and VM trajectories were first re-sampled in the space domain by means of linear interpolation of the x, y time series (1.5-mm steps) over all consecutive gait cycles. Two different estimates of the spatial variability of foot path were derived from these data. 1) Spatial density was calculated for each trial as the number of points falling in 0.5 × 0.5 cm² cells of the spatial grid divided by the number of step cycles. The density of each cell was depicted graphically by means of a color scale (empty cells were excluded in the plot). In addition, a mean density was computed over the swing (or stance) phase as the sum of all cell densities divided by the number of nonempty cells. 2) Normalized tolerance area was derived as follows (see Fig. 9A). The mean length L of foot trajectory over the swing phase (or over the stance phase) across all consecutive gait cycles in each trial was calculated as the corresponding path integral. Every 10% of the horizontal excursion we computed the two-dimensional 95%-tolerance ellipsis of the points within the interval (see Lacquaniti et al. 1987, 1990; McIntyre et al. 1998 for related techniques). The typical number of points in each interval ranged from 300 to 1,000 (depending on the walking speed and the number of gait cycles). The areas of all tolerance ellipses were summed and normalized by L. This total area provides an estimate of the mean area covered by the points per 1 cm of path.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Global gait parameters

Figure 2 shows the mean values (over all trials and subjects) of stance relative duration (A), stride length (B), and vertical excursion of the foot (C). Mean values (±SD) of trials performed at full body load (0 BWS) are plotted as a function of treadmill speed in the left column panels, whereas mean values of trials performed at different speeds are plotted as a function of BWS in the right column panels. At 0 BWS, global gait parameters exhibited the well known monotonic relationship with increasing speed (see Grillner 1981; Winter 1991). Thus the stride length significantly increased with speed (on average, by 6.2 cm per km/h increment, r² = 0.99), whereas the stance relative duration decreased (by 3%, r² = 0.81). At each tested speed, increasing the level of BWS from 0 to 95% (included) resulted in a small decrease of both stride length and stance relative duration (on average, by 0.66 cm and by 0.73% per 10% increment of BWS, respectively). Although small, these changes were statistically significant [F(4,28) = 4.6 and 33.3, P = 0.0054 and P < 0.00001, respectively, within-subject ANOVA]. Also, these gait parameters exhibited essentially the same relationship with speed for all tested levels of BWS except in air-stepping (100% BWS). In this latter condition, instead, the parameters did not change significantly with speed. On average, stance relative duration was 53 ± 3%, a value shorter than at any other BWS value, whereas stride length was 63.6 ± 0.7 cm. In air-stepping therefore, speed increments were achieved simply by means of proportional decrements of gait cycle duration.

View larger version (28K): [in this window] [in a new window]   Fig. 2. Mean values (over all trials and subjects) of stance relative duration (A), stride length (B), and vertical excursion of the foot (C). Left: values for trials performed at full body load (0 BWS) plotted as a function of treadmill speed. Right: values for trials performed at different speeds plotted as a function of BWS. In C, the values of the overall vertical (y) excursion of the distal-most marker (VM) are plotted along with those for a more proximal marker located at the ankle (LM).

Changing arm posture or walking with shoes instead of barefoot did not affect significantly any of the preceding parameters in any tested condition.

Pattern of vertical contact forces

Local loading of the plantar side of the foot was measured in the trials in which subjects wore shoes. The time course of the net vertical component of ground reaction forces is plotted as a function of the spatial coordinates of the foot in Fig. 3A, and the corresponding trajectory of the center of pressure (superimposed on the foot outline) is plotted in Fig. 3B. These data were obtained by averaging all gait cycles performed by one subject at 3 km/h at different levels of BWS; qualitatively similar data were obtained in all other subjects.

View larger version (29K): [in this window] [in a new window]   Fig. 3. A: time course of the net vertical component of in-shoe reaction forces plotted as a function of the spatial coordinates of the foot. Note change in vertical scale in the 2 bottom panels. In the "foam" condition, subjects stepped in air (100% BWS) while a foam-rubber taped under the feet lightly touched the treadmill belt during "stance". Fifteen gait cycles performed by one subject at 3 k/ h were averaged for each condition. B: trajectory of the center of pressure superimposed on a foot outline (corresponding to the outer elements of the pressure-sensitive matrix interposed between foot and shoe). C: mean values of the first (F1) and second (F2) peaks derived from force records such as those of A.

At BWS between 0 and 75% (Fig. 3B, left), the center of pressure was located initially at the heel, then shifted anterior and slightly lateral in the mid-foot, to end its trajectory medially under the first metatarsal head. The following events are recognizable in the force profiles for the same range of BWS (Fig. 3A, top 4 panels). Early stance lasts over the first 20% of total stance duration from heel-contact to the first main peak of vertical force due to loading with the body weight. Mid-stance spans the interval (~55% of stance duration) between the two main peaks of force. The trough is due to partial unloading of the limb by the vertical impulse generated in the previous push-off of the contralateral limb. Late stance coincides with the final push-off phase of the ipsilateral limb; it lasts from the second peak of force to toe-off. Early and late stance correspond to the double support phases. At 95% BWS, no measurable contact force was detected at the heel; the force was applied initially at the mid-foot, then shifted anterior and medial where the main peak was reached (Fig. 3B, right, and A, bottom left).

In general, increasing BWS level from 0 to 95% (included) resulted in a proportional decrement of force amplitude as one would expect. Mean values of the first (F1) and second (F2) force peaks are plotted in Fig. 3C. Both peaks decreased linearly with increasing BWS (r² = 0.99). On average, these peaks deviated by 7.5 ± 4% from the values predicted by assuming that they scale by the same amount as the nominal BWS.

In the standard condition with 100% BWS (not shown), subjects stepped in the air, their feet oscillating back and forth just above but never contacting the treadmill. Therefore no contact forces were recorded. The pattern of reaction forces recorded with surrogate contact (foam condition) will be presented in a later section.

For BWS <95%, the qualitative changes of vertical contact forces with speed resembled those previously described at 0 BWS (Borghese et al. 1996; Grasso et al. 2000b; Winter 1991). Thus the first peak and the trough became more pronounced with increasing speed (on average, their amplitude was significantly correlated with speed, r² = 0.87). For BWS 95%, contact force measurements became less reliable due to the limited resolution of the in-shoe recording device and to the force fluctuations of the unloading system (see METHODS).

EMG activity

Figure 4 illustrates the time course of EMG activity of lower limb muscles over the normalized gait cycle at different BWS. The ensemble averages of all gait cycles performed by two different subjects at 3 km/h are plotted in A and B, respectively. Figure 5 shows the mean EMG activity (over all trials and subjects) plotted as a function of treadmill speed and BWS (same format as in Fig. 2). Mean EMG was computed over all gait cycle (Fig. 5) and also separately over stance and swing phases.

View larger version (57K): [in this window] [in a new window]   Fig. 4. Time course of electromyographic (EMG) activity of lower limb muscles vs. the normalized gait cycle at different BWS. The ensemble averages of all gait cycles performed by 2 different subjects at 3 km/h are plotted in A and B respectively. GM, gluteus maximus, VL, vastus lateralis, RF, rectus femoris, BF, biceps femoris, TA, tibialis anterior, GCL, lateral gastrocnemius.

View larger version (22K): [in this window] [in a new window]   Fig. 5. Mean EMG activity was computed over the gait cycle and averaged across all trials and subjects. For each muscle, left-most column: values for trials performed at 0 BWS plotted as a function of treadmill speed. Right-most column: values for trials performed at different speeds plotted as a function of BWS.

At 0 BWS, the pattern of muscle activity was roughly comparable to that reported in the literature as were the changes in mean amplitude with increasing speed (see Grasso et al. 1998; Winter 1991; Winter and Yack 1987). Note that there is considerable inter-subject variability of EMG patterns especially for thigh muscles. Thus the main peaks of BF activity occurred in late swing/early stance in subject A, but in early/mid-swing in subject B. Mean activity of all tested muscles tended to increase exponentially with speed (on average, r² = 0.97, see Fig. 5, left-most panels in each pair), though the increment was not always monotonic. Thus in VL and RF mean activity was smallest at 2 km/h.

Increasing BWS level from 0 to 95% resulted in considerable changes of the pattern and amplitude of the EMG activity. The simplest kind of change with BWS can be seen in GM and GCL (hip and ankle extensors, respectively): their activity decreased significantly and monotonically with increasing BWS at all tested speeds [F(4,28) = 31.6 and 26.5, P < 0.00001], consistent with their antigravity function. The changes of the knee extensors VL and RF, however, were less intuitive and depended on speed. At low speeds (0.7-1.1 km/h), their activity did not change significantly with BWS. At intermediate speeds (2-3 km/h), it increased significantly (P < 0.00001) with BWS during all gait epochs we considered (overall gait cycle, separate stance, and swing phases). At higher speed (5 km/h), it decreased significantly (P < 0.005). Even more complex changes with BWS can be observed in the pattern of activity of hamstring muscles (BF) during both stance and swing: the main peaks of activity occurred around heel-contact for BWS between 0 and 50%, but they occurred in mid-stance and late stance at higher BWS. Overall changes of BF mean activity with BWS were not systematic across speeds except for the transition from 75 to 95% BWS when the mean activity increased significantly (P < 0.005) during all gait epochs. The changes of TA (ankle dorsi-flexor) were not systematic and depended on speed. At low speeds (0.7-1.1 km/h), TA activity did not change significantly with BWS. At intermediate speeds (2-3 km/h), it increased significantly (P < 0.005) with BWS, and at still higher speed (5 km/h), the changes were nonmonotonic, mean amplitude increasing when BWS increased from 0 to 75% and dropping at the transition from 75 to 95% BWS.

EMG patterns in air-stepping were quite different from those at lower BWS in some muscles (BF, TA, GCL), whereas they were roughly comparable with those at high BWS levels in other muscles (GM, VL, RF). In general, the pattern of activity in "antagonist" muscles (VL, RF vs. BF; GCL vs. TA) tended to be roughly reciprocally organized, although there was considerable inter-subject variability. For instance, six subjects (such as that of Fig. 4A) exhibited a high level of activity in TA and GCL muscles, whereas the other two subjects (such as that of Fig. 4B) had only modest muscle activity in the same muscles. Mean activity changed little with speed.

Limb kinematics

The average waveforms of the elevation angles of the lower limb segments are plotted versus the gait cycle in Fig. 6 (data from the same 2 subjects as in Fig. 4). The inset in each panel shows the 3D gait loops obtained by plotting the elevation angles one versus the others. Figure 7A reports the mean values (over all trials and subjects) of the percent variance (PV) accounted for by the first harmonic of each elevation angle as a function of speed and BWS. Figure 7B reports the mean PV accounted for by the three eigenvalues of the covariance matrix of the elevation angles (see METHODS). PV1 and PV2 are related to the global form of the 3D gait loop, and their magnitude relative to that of PV3 provides an index of planarity of the loop (when PV1  PV2 PV3) (see Borghese et al. 1996).

View larger version (30K): [in this window] [in a new window]   Fig. 6. Ensemble averages of the elevation angles of the lower limb segments plotted vs. the normalized gait cycle (data from the same 2 subjects as in Fig. 4). Insets: 3-dimensional (3D) gait loops obtained by plotting the elevation angles 1 vs. the others.

View larger version (24K): [in this window] [in a new window]   Fig. 7. A: mean values (over all trials and subjects) of the percent variance (PV) accounted for by the 1st harmonic of each elevation angle as a function of speed and BWS (same format as in Fig. 5). B: mean PV accounted for by the 3 eigenvalues of the covariance matrix of the elevation angles. PV1 and PV2 are related to the global form of the 3D gait loop, and PV3 provides an index of planarity of the loop (0% corresponds to an ideal plane).

At 0 BWS, the kinematic waveforms were very similar to those previously reported, and the 3D gait loop lied close to a plane during both stance and swing (Bianchi et al. 1998; Borghese et al. 1996; Grasso et al. 1998, 2000b). Increasing the treadmill speed or the level of BWS resulted in qualitatively similar changes of limb kinematics. Angular waveforms tended to become progressively more sinusoidal (as shown by the increase of PV of the 1st harmonic, Fig. 7A), and the 3D gait loop became narrower (PV1 increase and PV2 decrease, Fig. 7B). Note that the changes of all these parameters with BWS were almost continuous and monotonic throughout the 0-95% BWS range, except for the changes of PV of the 1st harmonic of thigh angle that were not related to BWS systematically across different speeds. Planar co-variation was obeyed at all tested speeds and BWS (on average, PV3 was 0.99 ± 0.16%). Planar co-variation was also obeyed under conditions that changed the nature of the mechanical interaction between the subject and the treadmill belt. In the standard experiments, the treadmill operated at a constant belt speed and therefore the subjects had to go along with the belt. In a different set of experiments (position-related treadmill speed, see METHODS) instead, at 0 BWS, the subjects could vary their walking speed at will and still remain within the central area of the treadmill because treadmill velocity was automatically adjusted to the instantaneous position of the subject. Limb kinematics was comparable to that described in the preceding text for the standard conditions.

In air-stepping, the angular waveforms were almost perfectly sinusoidal in all subjects at all speeds (on average, PV of the 1st harmonic equal to 98.6 ± 0.8%). This also can be appreciated by the almost perfect planarity of the 3D gait loop (PV3 = 0.09 ± 0.08%, note that the composition of 3 sinusoidal waveforms yields a plane). There was also a change in the inter-segmental coordination pattern as the phase lead of thigh oscillations relative to shank (and foot) oscillations was systematically greater than at all other BWS levels. Thus the 1st harmonic of the thigh angle led that of the shank by 43 ± 15° and by 94 ± 15° at 95 and 100% BWS, respectively. This is also reflected by the markedly different orientation of the 3D gait loop in these two conditions (Fig. 6).

Changing arm posture or walking with shoes instead of barefoot did not affect significantly eigenvectors and eigenvalues of the covariance matrix of the elevation angles in any tested condition. For 0 BWS, these data confirm the previous findings of Borghese et al. (1996) who also compared different walking styles.

Shape of endpoint trajectories

Figure 8 shows the superimposed trajectories of the ilium and foot (computed in the laboratory reference frame, see METHODS) over consecutive step cycles performed by one subject at different speeds (A) and BWS levels (B). Foot trajectories resembled horizontally elongated loops with the high portion during swing. The vertical excursion of the foot increased systematically with increasing speed, in relation with the increment in stride length (see Fig. 2B). For example, the overall vertical excursion of the distal-most marker (VM) and a more proximal marker located at the ankle (LM) was 5.4 ± 1.1 and 11.9 ± 2.2 cm, respectively, at 0.4 km/h, while it was 10.2 ± 0.9 and 16.9 ± 1.4 cm, respectively, at 6.5 km/h (see Fig. 2C, left). The inter-subject variability of VM vertical excursion was significantly (P < 0.005) smaller than that of LM, for all tested speeds.

View larger version (25K): [in this window] [in a new window]   Fig. 8. Shape and variability of endpoint path in 1 subject. The trajectories of the ilium (IL marker) and foot (VM) over 12 consecutive step cycles have been superimposed for trials performed at different speeds at 0 BWS (A) and different BWS levels at 3 km/h (B). C: plots of spatial density of VM path (same trials as in A and B, left and right columns, respectively). Spatial density was integrated over the swing phase: the lower the density (toward the blue in the color-cued scale), the greater the variability. Plots are anisotropic, vertical scale being expanded relative to horizontal scale. Note that density is roughly comparable in all conditions except in air-stepping (100% BWS), where it is much lower.

As for the effect of body weight unloading (Fig. 8B), at any given speed the path shape was relatively conserved from 0 to 95% BWS, but it became considerably expanded in the vertical direction at 100% BWS (see Fig. 2C, right). At each speed, the mean area of the LM and VM trajectories decreased slightly (by 20 and 8 cm² per 10% BWS increment, corresponding to 5.2 and 2.8% of the mean value at 0 BWS, P < 0.05) from 0 to 95% BWS, but it became considerably larger in air-stepping. For example, at 3 km/h, the mean area of the VM and LM paths was 299 ± 64 and 412 ± 75 cm², respectively, at 0 BWS, 213 ± 71 and 200 ± 96 cm² at 95% BWS and 1,109 ± 412 and 875 ± 411 cm² at 100% BWS.

Endpoint spatial variability

As it can be appreciated in Fig. 8, A and B, step-by-step reproducibility of foot path was high both across speed and unloading changes, except 100% BWS. We used two different estimates of the overall spatial variability (computed in the laboratory reference frame), the spatial density and the total normalized tolerance area (see METHODS); they gave qualitatively similar results. The spatial density of VM path during the swing phase is plotted in Fig. 8, C and D, for the same data depicted in Fig. 8, A and B. The spatial density at each discretized locus is depicted graphically by means of a color scale (empty cells were excluded in the plot). In this representation, the lower the density (toward the blue in the color scale) and the larger the integrated zone of the color areas, the greater the variability of foot path from cycle to cycle. Density was roughly comparable in all conditions except in air-stepping, where it was consistently lower. To compare conditions, a mean density was computed as the sum of all cell densities over the swing phase divided by the number of nonempty cells. For example, at 3 km/h, the mean density during the swing phase was significantly (paired t-test, P < 0.0003) lower at 100% BWS (0.8 ± 0.2) than at 95% BWS (2.0 ± 0.4).

Figure 9 reports the values of the total normalized tolerance area (computed in laboratory-fixed coordinates over the swing phase) for both VM and LM in each subject at different BWS (with a schematic of derivation in A, see METHODS). Note that a greater tolerance area indicates greater variability. Path variability was roughly comparable at all BWS (Figs 9, B-F), but 100% BWS (Fig. 9G, note change in scale) when it was much greater. Moreover the variability of VM path was systematically lower than that of LM at all speeds and BWS [F(1,7) = 83.4, P = 0.00004, within-subject ANOVA], but 100% BWS when it did not differ systematically between the two markers.

View larger version (40K): [in this window] [in a new window]   Fig. 9. Total normalized tolerance area in laboratory-fixed coordinates integrated over the swing phase. A: schematic of derivation (see METHODS). B-G: mean values of the tolerance area for both VM (dark) and LM (light) in each subject (s1-s8) for trials at a fixed speed (1.1 km/h) but different BWS. Note that a greater tolerance area indicates greater variability.

The values of total normalized VM tolerance area (VMA) over the swing phase have been averaged over all subjects and plotted in Fig. 10A as a function of speed and BWS. The results presented for one speed in Fig. 9 are now extended to all tested speeds. At 0 BWS, VMA was low at all tested speeds, with a minimum at 3 km/h. Increasing BWS from 0 to 95% (included) resulted in minor changes of mean VMA. This parameter increased slightly (on average, by 2.4% per 10% increment of BWS), but significantly with BWS [F(4,28) = 9.75, P < 0.0001]. On average, at 95% BWS, VMA was 1.3 times greater than that at 0 BWS. VMA was much larger in air stepping than in all other conditions in every subject, independent of speed. For example, at 3 km/h, the mean VMA was 2.3 ± 0.6 cm²/cm at 0-95% BWS and 11.6 ± 3.2 cm²/cm in air stepping (P < 0.0001, paired t-test). The results did not depend on the choice of the reference frame used to analyze foot trajectories: at 3 km/h, the mean VMA in IL-centered coordinates (see METHODS) was 2.0 ± 0.4 cm²/cm at 0-95% BWS and 11.1 ± 3.0 cm²/cm in air stepping. Moreover, the values of VMA integrated over the stance phase followed exactly the same trend with speed and BWS as those reported for the swing phase.

View larger version (25K): [in this window] [in a new window]   Fig. 10. Mean parameters (over all trials and subjects) describing kinematic variability during swing as a function of speed and BWS. A: total normalized VM tolerance area. B: SDs of VM vertical position. C: SDs of VM horizontal position. D-F: SDs of the elevation angle of thigh, shank and foot, respectively.

The indices we have considered so far describe the integrated variability of foot path. Additional information can be gained by examining the time-varying changes of trajectory variability. Figure 11 shows the values of VM and LM tolerance area discretized every 10% of the horizontal excursion for swing and stance. In the 0 to 95% BWS range, path variability was small at any time of the gait cycle, being generally lower during stance than during swing. Also, VM variability was always lower than LM variability except around heel touch-down (just prior to, during, and immediately afterwards). By contrast, at 100% BWS the variability of both VM and LM increased considerably as compared with lower BWS levels. Similar trends were observed at all other tested speeds.

View larger version (62K): [in this window] [in a new window]   Fig. 11. Time-varying changes of foot trajectory variability. Values of VM and lateral malleolus (LM) tolerance area have been discretized every 10% of the horizontal excursion for swing (A) and stance (B). Data obtained from all subjects at 3 km/h have been averaged and plotted as a function of BWS: the greater the tolerance area (toward the blue in the color-cued scale), the greater the variability. Interval 1 and 10 of the stance phase plots correspond to toe-off and heel-touch-down, respectively. Gait cycle begins at interval 10 of stance plots and proceeds in decreasing order for stance and increasing order for swing (see insets).

We also analyzed the variability of VM position separately in the vertical direction (VMy-SD, Fig. 10B) and in the horizontal direction (VMx-SD, Fig. 10C). These two parameters behaved differently with BWS changes. On average, VMy-SD followed the same trend as that described for the overall VMA, exhibiting only minor changes with changes in speed or BWS, except at 100% BWS when it jumped high. Moreover, as in the case of the tolerance area (Fig. 9, B-F), the variability of the vertical displacement was systematically lower for the VM marker than that for the LM marker at all speeds and BWS [F(1,7) = 187.1, P < 0.00001, within-subject ANOVA], except for 100% BWS when it did not differ systematically between the two markers.

As for VM variability in the horizontal direction (Fig. 10C), increasing BWS from 0 to 95% resulted in a significant (P < 0.0001) increase at low speeds (<2 km/h) but no significant change at higher speeds. VMx SD in air-stepping did not differ systematically from that at other BWS values for comparable speeds.

The variability in the lateral direction (VMz-SD, not shown in the figure) was ~1 cm and did not depend systematically on walking speed nor on BWS, including air-stepping. Changing arm posture or walking with shoes instead of barefoot did not affect significantly the variability parameters of the foot markers in any tested condition.

Variability of segment angles

The pattern of changes of foot trajectory variability with changes of speed and BWS can be compared with the corresponding pattern of variability of the time-varying changes of limb segment angles in Fig. 10. Mean SD values for thigh, shank, and foot angles are plotted in D-F, respectively. In general, SD values of segment angles (and joint angles, not shown) depended on both speed and BWS. In particular, increasing BWS from 0 to 95% resulted in a significant increment of SD values of thigh, shank, and foot angles (P < 0.0002). In the 0-95% BWS range, these values were always significantly higher at low speeds (0.4-1.1 km/h) than at high speed (2 km/h) (P < 0.00001).

In air-stepping, the SD values of the thigh and shank were greater than those at 95% BWS at high speeds (>2 km/h; P < 0.003) but did not differ significantly for low speeds (<2 km/h). SD values of the foot did not differ significantly from those at 95% BWS levels at any speed.

Kinematic variability was smallest ~3 km/h at all BWS except in air-stepping when no systematic dependence on speed could be found (Fig. 10).

One-legged stepping

We have seen that the integrated variability of foot trajectory undergoes abrupt, stepwise changes when subjects stepped in air as compared with any other tested load (Fig. 10A). To exclude that this result depends on the lack of passive mechanical stabilization of the body from the supporting limb and to clarify the role of ipsilateral versus contralateral contact, we performed additional experiments. In one set of experiments, subjects stepped with the right leg only, while supporting their weight on the stationary left leg (BWS was set at 0). Two different conditions from one such experiment are compared in Fig. 12A: the subject stepped on the treadmill (left) and she stepped in air, the foot moving just above but never contacting the treadmill (right). Once again in air-stepping, spatial variability (measured as density in A, and as tolerance area in B) increased substantially in all subjects compared with the control. Mean ± SD values of VM tolerance area over all subjects are plotted in Fig. 12B.

View larger version (15K): [in this window] [in a new window]   Fig. 12. One-legged stepping. Subjects stepped with the right leg only, while supporting their weight on the stationary left leg. A: stepping on the treadmill (left), and air-stepping (right) in 1 subject and corresponding values of spatial density during swing phase. B: mean ± SD values (over all subjects) of total VM tolerance area over swing.

Between-limbs dissociated contact forces in two-legged stepping

In this experiment, subjects stepped with the left leg on the moving treadmill and with the right leg free in air. In this manner, only the left limb sensed contact forces during the stance phase. Step-by-step spatial variability of the right foot trajectory did not differ significantly from that measured in two-legged air-stepping.

Air-stepping with surrogate contact forces

The experiments reported so far have indicated that the kinematics during both stance and swing is affected by the presence or absence of ground contact of the same limb during stance. However, the stance phase of ground stepping differs from its temporal equivalent in air-stepping not only because of the contact forces but also because of the shape of foot trajectory. Thus the foot trajectory is systematically flat over a large portion of stance when the foot contacts the ground, whereas it tends to be gently curved in the absence of contact (see Fig. 8B). Thus it could be argued that the major changes in foot variance we found in air-stepping do not depend on the absence of contact forces but depend on the fact that this task corresponds to a form of movement unrelated to ground stepping. To address this point, we performed a series of experiments in which subjects were unloaded at 100% BWS but sensed minimal contact forces during "stance." This was obtained in two different ways. In one protocol, the lower surface of a compliant foam-rubber, taped under the subjects' feet, lightly touched the moving belt of the treadmill during stance. In another protocol, 100% unloaded subjects stepped on a compliant (pillow-like) foam placed over the stationary treadmill belt.

The time course of the net vertical reaction force and the trajectory of the center of pressure during stance are plotted in Fig. 3A (bottom right) and Fig. 3B (right), respectively, for the protocol with foam rubber. As a result of the foam interface, small vertical contact forces (on average, 35 N, with a peak of about 50 N, see Fig. 3C) were applied over a limited portion of the foot plant, close to the first metatarsal head. Figure 13A shows foot kinematics in one typical experiment. In both stepping with the taped foam (middle), and stepping on the pillow (right), foot trajectory during stance was gently curved as in air-stepping (left), but minimal reaction forces occurred at the point of contact. Step-by-step variability decreased substantially in the presence of these surrogate contact forces as compared with air-stepping. This can be seen in the spatial density plots of the lower row of Fig. 13A for the same experiment depicted in the top row of the same figure, and by considering the mean (± SD, 4 subjects) values of the VM tolerance area over swing in Fig. 13B. On average, mean density was 0.8 ± 0.1, 1.7 ± 0.3, and 1.6 ± 0.2 in air-stepping, foam and pillow conditions, respectively. VM tolerance area was 12.5 ± 3.1 cm²/cm, 3.9 ± 1.3 and 4.2 ± 0.6, respectively. The mean values in foam and pillow conditions are significantly (P < 0.05, paired t-test) different from those in air-stepping.

View larger version (20K): [in this window] [in a new window]   Fig. 13. A: air-stepping at 100% BWS with surrogate contact forces in 1 subject and corresponding values of VM spatial density. Left: standard air-stepping. Middle: a foam-rubber was taped under the subjects' feet and lightly touched the belt of the treadmill during "stance". Right: the subject stepped on a pillow. Note that for both foam- and pillow-stepping, foot trajectory during "stance" was gently curved as in air-stepping but minimal reaction forces occurred at the point of contact (see Fig. 3). B: Mean ± SD values (over all subjects) of total VM tolerance area. Variability decreased substantially in the presence of surrogate contact forces relative to standard air-stepping.

Effect of instruction

It could be argued that the kinematic results reported so far depend strictly on the experimental arrangement and the specific instruction given to the subject. To address this issue, we carried out the following protocols. In one condition the subject, unloaded at 0, 95, or 100% BWS, stepped at 3 km/h while he performed mental arithmetics. This task is known to interfere with other concomitant high-level cognitive tasks and therefore one would expect that high-level components of walking should be disrupted or at least modified by the concurrent mental arithmetics. In other words, if the results were mainly attributable to the instructions given to the subjects and their subsequent adaptive high-level interventions, one should see a behavior different from that seen in the standard condition performed without mental arithmetics. Instead, we found that both the shape (mean area of VM trajectories) and step-by-step variability (VM tolerance area over swing or stance) of foot path did not differ significantly from the values obtained in the control condition (without mental arithmetics).

In a different set of conditions, the subject was not suspended by the WARD system, but he raised himself above the treadmill by pushing with the extended arms against the lateral rollbars of the treadmill. In one task, the instruction to the subject was to step in the air, as if he walked on ground (as in the standard air-stepping). In another task, the instruction was to step in the air while trying to touch the ground with the heels (without actual contact). This was obtained by dorsi-flexing the feet at the ankle. In still another task, the subject was asked to reproduce in the air the cyclical movement involved in bicycle pedaling. As one would expect, these different instructions resulted in foot trajectories with a different path: leaf-like trajectories in standard air-stepping, roughly elliptic trajectories in air-stepping with ankle dorsi-flexion, and roughly circular trajectories in air-pedaling. However, the step-by-step variability (VM tolerance area over swing or stance) of foot path did not differ significantly between these conditions, nor did it differ from that measured when using the WARD suspension system to obtain 100% BWS.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

There are three main points in this work: 1) lower limb kinematics is accurately controlled over a wide range of walking speeds and loads, in spite of the drastic changes of limb kinetics (ground contact forces and muscle activity); 2) minimal vertical contact forces sensed during stance are sufficient for accurate control of ipsilateral foot trajectory; and 3) air-stepping shows kinematic characteristics different from those of ground stepping, and therefore will be discussed separately. We shall address the methodological issues first, then each of the preceding points.

Methodological issues

In most conditions, subjects walked on a treadmill with constant belt speed. To verify the effect of changing the nature of the mechanical interaction between the subject and the treadmill belt, we also used a protocol in which the subjects could vary their walking speed at will and still remain within the central area of the treadmill because treadmill velocity was automatically adjusted to the instantaneous position of the subject. The results were comparable to those obtained under the standard conditions.

The BWS mechanism we used applies a controlled vertical force about the center of body mass. As a result, each supporting limb experiences a simulated reduction of gravity proportional to the applied force, while the swinging limb experiences 1 g. The overall error in the applied force has previously been estimated to be <5% of body weight (Gazzani et al. 2000). In the present study, the peak resultant vertical contact forces deviated by <10% from the values predicted by assuming that they scale by the same amount as the nominal BWS.

With our setup, we could only measure the vertical component of contact forces, but Griffin et al. (1999) measured horizontal fore-aft forces along with vertical forces using a similar unloading mechanism. They found that peak horizontal forces were reduced in near proportion to peak vertical forces with unloading between 0 and 75% BWS.

The present results can be related to those obtained using similar simulators of reduced gravity (Barbeau et al. 1987; Donelan and Kram 1997; Farley and MacMahon 1992; Finch et al. 1991; Griffin et al. 1999; Harkema et al. 1997). Thus global gait parameters such as stride length and stance relative duration were comparable to those previously reported for matched speeds and BWS (Donelan and Kram 1997