IEEE Robotics & Automation Magazine - December 2014 - 60

Average Absolute Human Torque (Nm/kg)

Optimization

Experiment
-17.0%

-12.1%

0.8
0.6
0.4
0.2
0

Baseline/ Support Baseline No Support
No Support
Support
Ankle

Knee

Hip

n=4

Figure 5. The mechanical performance: the average absolute
human joint torque in different conditions. On the left, the
optimization results. The optimization predicts a decrease in
average absolute joint torque of 17%. Note that the effect of the
added weight is not considered in the optimization, and, therefore,
the baseline and no-support condition could not be discriminated.
On the right, the experimental results. The difference between the
support condition and no-support condition is 12.1%.

the optimization were obtained from different subjects in a
different lab, causing intrasubject differences. A video of a
subject walking in the support condition is available in the
supplemental material in IEEE Xplore.
Metabolic Cost
The metabolic power of the various subjects is shown in
Figure 6. For all subjects, the metabolic cost of walking during
the baseline condition was the lowest. The metabolic cost in
the no-support condition was significantly higher (21.2%) than
in the baseline condition (p < 0.05). The added mass explains

Metabolic Power (W/kg)

5
4
3
2
1
0

Baseline

No
No Support 1 Support 2
n=6
Support 1 Support 2

Figure 6. The metabolic cost: the metabolic power for the
different walking conditions. The bars represent the average over
the subjects with the standard deviations.

IEEE ROBOTICS & AUTOMATION MAGAZINE

December 2014

most of this increase since the estimated increase in metabolism due to the added mass was 14%. In contrast with our predictions, the average metabolic cost in the support condition
was 6.1% higher than in the no-support condition ( p = 0.052).
Discussion
We built the XPED2 to test the following hypotheses: 1) the
exotendons do not influence joint angles and total joint
torques and 2) a reduction in the human joint torques results
in a reduction in the metabolic cost of walking. Although our
experiment is based on data from a small number of subjects,
we could identify some clear trends. Our experiments demonstrated that deviations from normal walking occur, and
thereby we disproved our first hypothesis. The changes in
ankle motion were also observed in [9]. Still, the exotendons
did reduce the average absolute human joint torque. The reduction in average absolute human joint torque did not result
in a measurable reduction in the metabolic cost-there was
even an indication that the metabolic cost increased. This
contradicts the second hypothesis. However, the effect could
have been small and unnoticed. The exotendons did only provide a small amount of support (12.1%). When changing the
exotendon parameters or configuration, this number might
be increased. Given the changes observed in the walking pattern, it is unlikely that a 71% reduction in average absolute
joint mentioned in [1] is feasible. The preliminary pilot trials
taught us that subjects were uncomfortable with higher
amounts of support.
Conclusions
Our exoskeleton has shown the contrast between the theory
and the experiment and thereby stressed the importance of
experimental evaluation of exoskeleton designs. We compared our results to experimental results obtained with other
exoskeletons for walking augmentation (Table 1). Although
some exoskeletons have shown a (relative) reduction in metabolic cost, there is no general consensus on how to reduce
metabolic cost most effectively.
The hypothesis that the metabolic cost is reduced when
the absolute joint torque is reduced is valid for isometric contractions, but in walking there are additional effects that
might interfere with this relation. Energy storage and transfer
between joints is already partly covered by the human tendons and biarticular muscles. Adding exotendons might interfere with these energy saving mechanisms. In humans, it
has been shown that the Achilles tendon stiffness is optimal
in the sense that muscle work during walking is minimal [5].
In hopping experiments, it has been shown that adding a
parallel spring reduced muscle force but increased muscle
work [16]. The observed changes in kinematics and kinetics
might be explained by the fact that the support by the exoskeleton enforces a new equilibrium.
The metabolic cost of walking could be partially explained
with the model used by [17]. The model predicts that the metabolic cost of walking emerges from the need to compensate
for energy losses after impact. This is in line with experimental

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2014