Professional Documents
Culture Documents
Authorized licensed use limited to: Nanyang Technological University. Downloaded on November 02,2020 at 01:40:31 UTC from IEEE Xplore. Restrictions apply.
2 hours. Motors are controlled at 1000 Hz and assistance
and recognition algorithms runs at 100 Hz. The prototype is
designed in two sizes and additional adjustment in width and
length is possible to an extent. In this study, exoskeletons in
all two sizes were used to account for different physiques of
the subjects.
Determining assistance torque pattern in response to the
user gait is a crucial part as well as the hardware design of the
exoskeleton. As was explained in [5], [6], the algorithm uses
particularly-shaped adaptive oscillator (PSAO) to estimate
current gait phase, the normalized stride time, for each
Fig. 2. The evaluation system with GEMSv2 fastened. The system is
joint by reading the hip joint angle. The control algorithm controlled to reproduce hip joint angle trajectory of nominal human walking
then looks up a torque table to determine torque values, while GEMSv2 is running on the assistance algorithm. The evaluation
corresponding to the current gait phases, to command to the system can estimate the external perturbation, which corresponds to the
power delivered by GEMSv2.
actuators.
To compose the torque table, we have adopted torque
patterns from gait analysis data for walking on level ground,
slopes, and stairs introduced in [11], [12], [13]. The torque
patterns were then modified empirically through iterative
trials in order to stabilize the dynamics of the exoskeleton
and to minimize discomfort to users. The overall behavior of
the exoskeleton remained stable if the peak torque timings for
one leg and the other were close to each other. Through trials
we realized that users often perceived negative work from the
exoskeleton as uncomfortable sensation limiting natural gait.
After adjusting the torque profile to remove such discomfort,
the power from the exoskeleton was mostly positive over a
stride.
To further address inter-subject variation in the preference Fig. 3. A subject is walking without exoskeleton and then with exoskeleton
for the timing of assistance, the control algorithm has been on incline of 10% grade
designed to trim the timing in order to align the peaks of
the joint angular velocity with those of the joint power with
specified timing offsets in a step-by-step fashion. From pre- III. T HE E XPERIMENT
liminary experiments, we found that the effect of assistance
A. Subjects and Protocol
is maximal at the timing offset of 0 to 4% gait cycle. In this
study, we kept the timing offset to 0% for consistency. Five subjects, all male adults without known medical
The power generated by the exoskeleton diminishes due issues, participated in the experiment. For each subject, we
to friction and dynamics of the exoskeleton before being applied the following protocol.
delivered to a user. To identify the actual power that is Task 1: Stand still
delivered to the user without using additional torque/force Task 2: Walk (0% grade)
sensors, we performed tests to the prototypes. In the tests, Task 3: Walk (10% grade)
the exoskeleton was mounted on an evaluation system as Task 4: Walk (5% grade)
in Fig. 2, whose two motorized joints are programmed to Task 5: Walk (0% grade with exoskeleton)
regenerate typical human hip joint trajectory of walking. The Task 6: Walk (10% grade with exoskeleton)
exoskeleton then generates the torque from the torque table Task 7: Walk (5% grade with exoskeleton)
with extra torque to compensate the loss. Task 8: Walk (0% grade)
From data collected from the evaluation system, we esti- Task 9: Stand still
mated the power that was delivered to the evaluation system. All the standing lasted for 5 minutes and walking 6 minutes
The efficiency, the ratio between the power transferred to the on a treadmill. To account for leg length, the treadmill√speed
system and the power generated by the exoskeleton, was then was determined from the dimensionless speed 0.42 gl as
computed. For the two sizes ‘L’ and ‘M’, and the left and the in [6], where g is the gravitational acceleration and l is
right motors, the efficiency was found as 0.665 (size L, left), the leg length. The experimental setup is shown in Fig. 3.
0.684 (size L, right), 0.815 (size M, left), and 0.647 (size M, The intensity of assistance, manually set by a programmable
right). Throughout the rest of the paper, the term ‘exoskeleton user interface, was set at the same level for all the subjects:
power’ refers to the mean positive power generated by the peak torque of 7 Nm for extension, 5.4 Nm for flexion. The
exoskeleton after applying the efficiency. information of the subjects are summarized in Table I.
247
Authorized licensed use limited to: Nanyang Technological University. Downloaded on November 02,2020 at 01:40:31 UTC from IEEE Xplore. Restrictions apply.
TABLE I
T READMILL S PEEDS AND S UBJECT I NFORMATION
100
B. Metabolic Cost of Walking 0
To assess the metabolic cost of walking we used K5 0 5 10
(Cosmed, Italy), a pulmonary gas measurement system. The Surface gradient (%)
equipment computed energy expenditure rate in kcal/min Fig. 4. Gross metabolic rate averaged for 3 subjects at different conditions:
based on the measured respiratory gas volume rates. We the bars indicate 1 standard deviation.
computed the mean of the metabolic power for the last 3
minutes from walking tasks and the last 2 minutes from
standing tasks to be safe from transient effect. In the table, the 2nd task of walking on level ground
TABLE II
without exoskeleton was repeated at the 8th task to verify
G ROSS M ETABOLIC R ATE (W)
that the measurement was not seriously flawed. The changes
of GMR between the two tasks ranged from -21 to 14 W
with the mean of -1.9 W, within a permissible error range.
Subject Number
Task
1 2 3 4 5 Measurement in the 8th task was not used when calculating
1. Stand Still 109.0 142.8 98.8 116.7 135.1 the change in NMR.
2. 0% grade, No-exo 292.2 378.5 318.7 344.7 350.5
3. 10% grade, No-exo 502.9 670.8 544.9 622.2 624.8
The mean and standard deviation (s.d.) of the GMR
4. 5% grade, No-exo 383.4 497.4 437.9 455.8 505.0 measured for walking with and without exoskeleton are
5. 0% grade, Exo 274.9 335.5 278.6 295.5 349.0 illustrated in Fig. 4 along with those of BMR for comparison.
6. 10% grade, Exo 443.7 611.3 513.1 552.0 625.8
7. 5% grade, Exo 340.3 455.1 369.2 422.1 443.1
The mean increased linearly with the grade in both cases
8. 0% grade, No-exo 298.7 364.2 325.2 334.6 371.2 of walking with and without exoskeleton. For each incline,
9. Stand Still 110.4 129.5 107.9 125.7 151.6 the mean GMR for walking with exoskeleton was always
No-exo=walking without exoskeleton; Exo=walking with exoskeleton found lower than walking without exoskeleton, indicating
that the exoskeleton reduced metabolic energy expenditure
while walking uphill as well as on level ground.
TABLE III
BASAL M ETABOLIC R ATE (W) TABLE IV
R EDUCED NMR BY E XOSKELETON
Subject Number
Task
1 2 3 4 5 Reduced NMR (W) Reduced NMR (%)
1. Stand Still 109.0 142.8 98.8 116.7 135.1 Subject Surface grade (%) Surface grade (%)
2. 0% grade, No-exo 109.2 141.2 100.0 117.8 137.1 0 5 10 0 5 10
3. 10% grade, No-exo 109.3 139.5 101.1 119.0 139.2 1 17.8 43.6 59.8 9.7 15.9 15.2
4. 5% grade, No-exo 109.5 137.8 102.2 120.1 141.3 2 38.0 37.3 54.4 16.0 10.4 10.2
5. 0% grade, Exo 109.7 136.2 103.4 121.2 143.3 3 43.5 72.1 35.2 19.9 21.5 7.9
6. 10% grade, Exo 109.9 134.5 104.5 122.3 145.4 4 52.6 37.0 73.5 23.2 11.0 14.6
7. 5% grade, Exo 110.0 132.8 105.6 123.5 147.4 5 7.7 68.1 5.2 3.6 18.7 1.1
8. 0% grade, No-exo 110.2 131.1 106.8 124.6 149.5 (Mean) 31.9 51.6 45.6 14.5 15.5 9.8
9. Stand Still 110.4 129.5 107.9 125.7 151.6
No-exo=walking without exoskeleton; Exo=walking with exoskeleton
248
Authorized licensed use limited to: Nanyang Technological University. Downloaded on November 02,2020 at 01:40:31 UTC from IEEE Xplore. Restrictions apply.
25 120
p = 0.015
p = 0.002 Without Exo
With Exo
20
Reduced NMR (%)
Cadence (step/min)
115
p = 0.019
15
110
10
105
5
0 100
0 5 10 0 5 10
Surface gradient (%) Surface gradient (%)
Fig. 5. The distribution of the reduction rate for NMR due to the hip Fig. 7. The mean cadence while walking on each incline is plotted. The
exoskeleton for 0, 5, and 10% grade is shown. The p-values from one- upward or downward bars indicate 1 s.d. The mean cadence decreased
sample t-test are at the top of the boxes, all the cases indicating statistical slightly as the incline increased. The exoskeleton increased the mean
significance. cadence slightly.
TABLE V
80 p = 0.018 M EAN P OSITIVE E XOSKELETON P OWER D ELIVERED (W)
p = 0.003
Reduced NMR (W)
249
Authorized licensed use limited to: Nanyang Technological University. Downloaded on November 02,2020 at 01:40:31 UTC from IEEE Xplore. Restrictions apply.
Level 5% grade 10% grade
1 1
12
0 0
-0.5 -0.5
11
0 50 100 0 50 100
10
Right Joint Torque (Nm)
10 10
Left Joint Torque (Nm)
5 5 0 5 10
Surface gradient (%)
0 0
Fig. 9. Assistance Power by Exoskeleton
-5 -5
-10 -10
0 50 100 0 50 100
0.8
30 30
Right Joint Power (W)
Apparent Efficiency
Left Joint Power (W)
20 20 0.6
10 10
0.4
0 0 0.26
0.20 0.20
0.2
-10 -10
0 50 100 0 50 100
0
Fig. 8. Exoskeleton data recorded from a subject is shown regarding
joint flexion angle, extension torque and generated power. The joint torque 0 5 10
represents the desired assistance torque before adding compensating torque. Surface gradient (%)
The joint power is computed from sensed joint angular velocity (not
illustrated) and motor torque (calculated from motor current), and then Fig. 10. The median of apparent efficiency is higher for walking on level
discounted by the efficiency. ground than walking uphill. However, no significant difference was noticed.
First, the apparent efficiency ηhip for hip is computed using IV. D ISCUSSION
an equation for augmentation factor (AF), introduced in [16]
A. Efficiency for Walking on Different Gradients
p+ + pdis Both the metabolic power saved by the exoskeleton and
ηhip = , (1)
AF + βmexo the exoskeleton power delivered to the subjects increased as
where p+ is the mean positive power added by the exoskele- the gradient increased from 0 to 5 and 10%. The rate of
ton; pdis is the net dissipated power; mexo is the mass of the the increase was much higher in the saved metabolic power
exoskeleton (2.4 kg); β is the location factor (3.3 W/kg for than in the delivered exoskeleton power, resulting in decrease
the waist) for the added mass mexo ; AF corresponds to the in the apparent efficiency and increase in the exoskeleton
saved metabolic power. In our case, we can safely assume performance index.
pdis = 0.2 The performance index Π is then computed using The small increase in the exoskeleton power can be
an equation from [15] as explained by the change in the joint kinematics as the surface
ηmuscle gradient increases. The exoskeleton output torque was fixed
Π= , (2) to the same profile regardless of subject weight or walking
ηhip
speed and grades. From Fig. 8, we could verify that hip joint
where ηmuscle = 0.25 is the efficiency of separated muscle. kinematics had been affected by the surface inclines while
Computed values for the apparent efficiency and the the assistance torque had not changed much.
performance index are shown in Table VI along with the From the apparent efficiency and the exoskeleton perfor-
distribution for the apparent efficiency in Fig. 10. mance index listed in Table VI, we can predict, between
2 One should be careful with the meaning of pdis , different from mean level ground and inclined surfaces, on which surface human
negative power p− . pdis = p+ + p− when p+ + p− < 0 and pdis = 0 hip joint works more efficiently as well as on which one the
otherwise. Hence, it is possible that a passive ankle exoskeleton has positive exoskeleton is more helpful when it does the same amount
AF if it absorbs p− < 0 to its spring for one period and outputs p+ > 0
from the spring for other period and if net power loss pdis is much less of work. The apparent efficiency 0.26 for level walking is in
than p+ so that p+ + pdis > 0. good agreement with the value 0.29 in our previous work [6]
250
Authorized licensed use limited to: Nanyang Technological University. Downloaded on November 02,2020 at 01:40:31 UTC from IEEE Xplore. Restrictions apply.
where GEMSv1 was used. The hip apparent efficiency ηhip R EFERENCES
for walking uphill is found as low as 0.20 for both 5 and [1] H. Shimada, T. Hirata, Y. Kimura, T. Naka, K. Kikuchi, K. Oda,
10% grade. It suggests that the muscle-tendon complex in K. Ishii, K. Ishiwata, and T. Suzuki, “Effects of a robotic walking ex-
the hip joint works less efficiently when walking uphill than ercise on walking performance in community-dwelling elderly adults.”
Geriatrics & gerontology international, vol. 9, no. 4, pp. 372–81, dec
on level ground. This aligns with [17], where Sawicki et al. 2009.
predicted that proximal joints such as hip and knee should [2] C. Buesing, G. Fisch, M. O’Donnell, I. Shahidi, L. Thomas, C. K.
have lower apparent efficiency than ankle and that it will be Mummidisetty, K. J. Williams, H. Takahashi, W. Z. Rymer, and
A. Jayaraman, “Effects of a wearable exoskeleton stride manage-
even lower when walking uphill. ment assist system (SMA
) R on spatiotemporal gait characteristics
As for the exoskeleton, the performance index Π was 0.97 in individuals after stroke: a randomized controlled trial.” Journal of
in level walking and 1.24 in 5 and 10% grade, implying that neuroengineering and rehabilitation, vol. 12, no. 1, p. 69, aug 2015.
[3] F. Giovacchini, F. Vannetti, M. Fantozzi, M. Cempini, M. Cortese,
assisting in the hip joint is more economical on inclined A. Parri, T. Yan, D. Lefeber, and N. Vitiello, “A light-weight active
surfaces than on level ground. In addition to our previous orthosis for hip movement assistance,” in Robotics and Autonomous
suggestion in [6] that hip exoskeletons should be more Systems, vol. 73, sep 2015, pp. 123–134.
[4] V. Ruiz Garate, A. Parri, T. Yan, M. Munih, R. Molino Lova,
economic than ankle exoskeletons in saving metabolic cost N. Vitiello, and R. Ronsse, “Walking Assistance Using Artificial
of walking for the same amount of exoskeleton power, this Primitives: A Novel Bioinspired Framework Using Motor Primitives
study implies that assisting the hip joint when walking on for Locomotion Assistance Through a Wearable Cooperative Exoskele-
ton,” IEEE Robotics and Automation Magazine, vol. 23, no. 1, pp.
inclined surface is even more economic than on level ground. 83–95, 2016.
We should remark that the apparent efficiency ηhip could [5] K. Seo, S. Hyung, B. K. Choi, Y. Lee, and Y. Shim, “A new adaptive
possibly represent the combined apparent efficiency of the frequency oscillator for gait assistance,” in 2015 IEEE International
Conference on Robotics and Automation (ICRA). IEEE, may 2015,
whole lower limb rather than the separate hip joint. In pp. 5565–5571.
fact, there exist biarticular muscles mechanically linking the [6] K. Seo, J. Lee, Y. Lee, T. Ha, and Y. Shim, “Fully autonomous
hip joint and the knee joint. There was also a report that hip exoskeleton saves metabolic cost of walking,” in 2016 IEEE
International Conference on Robotics and Automation (ICRA). IEEE,
unilateral assistance in hip joint reduced ankle muscle activ- may 2016, pp. 4628–4635.
ities [18]. To clarify this point, individual muscle activation [7] Y. Lee, B. Choi, J. Lee, M. Lee, S.-g. Roh, J. Kim, H. Choi, and Y.-J.
should be examined using the electromyography. Kim, “Flexible sliding frame for gait enhancing mechatronic system
(GEMS),” in 2016 38th Annual International Conference of the IEEE
More precise estimate of apparent efficiency and perfor- Engineering in Medicine and Biology Society (EMBC). IEEE, aug
mance index can be obtained by equipping force sensors 2016, pp. 598–602.
that can measure the interaction between the user and the [8] J. Jang, K. Kim, J. Lee, B. Lim, and Y. Shim, “Online gait task recog-
nition algorithm for hip exoskeleton,” in 2015 IEEE/RSJ International
exoskeleton. The efficiency of GEMSv2 measured above Conference on Intelligent Robots and Systems (IROS). IEEE, sep
using the test jig does not reflect the power lost due to 2015, pp. 5327–5332.
the interaction with soft human tissue and insecure fastening [9] B. Lim, K. Kim, J. Lee, J. Jang, and Y. Shim, “An event-driven
control to achieve adaptive walking assist with gait primitives,” in
of the device. The general observation that the performance 2015 IEEE/RSJ International Conference on Intelligent Robots and
index was greater on inclines than level ground, however, Systems (IROS). IEEE, sep 2015, pp. 5870–5875.
should not be affected by the precision of the exoskeleton [10] H.-J. Lee, S. Lee, W. H. Chang, K. Seo, Y. Shim, B.-O.
Choi, G.-H. Ryu, and Y.-H. Kim, “A Wearable Hip Assist
power because error in estimating the exoskeleton power Robot Can Improve Gait Function and Cardiopulmonary Metabolic
should be small when compared to changes in the metabolic Efficiency in Elderly Adults,” IEEE Transactions on Neural Systems
cost. and Rehabilitation Engineering, accepted for publication. [Online].
Available: http://ieeexplore.ieee.org/document/7843664/
[11] J. Perry, Gait Analysis: Normal and Pathological Function, 2010,
B. Limitation vol. 12.
Because we did not randomize the trials in the protocol, [12] R. Riener, M. Rabuffetti, and C. Frigo, “Stair ascent and descent at
different inclinations,” Gait & Posture, vol. 15, no. 1, pp. 32–44, feb
we can not exclude the possibility that the result had been 2002.
affected by certain types of time effect such as fatigue or [13] M. Q. Liu, F. C. Anderson, M. H. Schwartz, and S. L. Delp, “Muscle
customization to the device and the treadmill. contributions to support and progression over a range of walking
speeds,” Journal of Biomechanics, vol. 41, no. 15, pp. 3243–3252,
2008.
C. Conclusion [14] E. Asmussen and F. Bonde-Petersen, “Apparent efficiency and storage
Our result for healthy adults indicates that GEMSv2 can of elastic energy in human muscles during exercise.” Acta physiologica
Scandinavica, vol. 92, no. 4, pp. 537–45, dec 1974.
assist walking to reduce metabolic energy expenditure on [15] G. S. Sawicki and D. P. Ferris, “Mechanics and energetics of level
both level ground and slopes. Analysis on the performance walking with powered ankle exoskeletons,” Journal of Experimental
index and the apparent efficiency shows that walking as- Biology, vol. 211, no. 9, pp. 1402–1413, may 2008.
[16] L. M. Mooney, E. J. Rouse, and H. M. Herr, “Autonomous exoskeleton
sistance in the hip joint is more efficient for uphill than reduces metabolic cost of human walking.” Journal of neuroengineer-
level ground, which in turn implies that human leg is no ing and rehabilitation, vol. 11, no. 1, p. 151, jan 2014.
more efficient for walking on inclined surface than on [17] G. S. Sawicki and D. P. Ferris, “Mechanics and energetics of incline
walking with robotic ankle exoskeletons.” The Journal of experimental
level ground. Further research should be directed toward biology, vol. 212, no. Pt 1, pp. 32–41, jan 2009.
the possibility that the elderly with lower muscle-tendon [18] T. Lenzi, M. C. Carrozza, and S. K. Agrawal, “Powered Hip Exoskele-
efficiency can benefit even more than young adults can from tons Can Reduce the User’s Hip and Ankle Muscle Activations During
Walking,” IEEE Transactions on Neural Systems and Rehabilitation
the exoskeleton when walking in their neighborhood with Engineering, vol. 21, no. 6, pp. 938–948, nov 2013.
hills.
251
Authorized licensed use limited to: Nanyang Technological University. Downloaded on November 02,2020 at 01:40:31 UTC from IEEE Xplore. Restrictions apply.