IEEE Robotics & Automation Magazine - March 2016 - 93

(% TM)

Left Ankle Torques
(Nm/kg)

Right Ankle Torques
(Nm/kg)

Left Knee Torques
(Nm/kg)

Right Knee Torques
(Nm/kg)

Left Hip Torques
(Nm/kg)

Right Hip Torques
(Nm/kg)

torque being delivered. In the
case of the assistive modes, a
0.5
0.5
fraction of these reference hip
joint torques was provided,
0
0
depending on the desired
- 0.5
-0.5
level of assistance.
0
20
40
60
80
100
0
20
40
60
80
100
In general, all computed
Gait Cycle (%)
Gait Cycle (%)
torque reference profiles are
NO
TM
(a)
(b)
similar to the ones found in
LTA
HTA
the literature. This was
LSA
0.5
0.5
expected, since these profiles
HSA
[23]
were precisely generated by
0
0
primitives constructed from
the literature data. We can
- 0.5
-0.5
0
20
40
60
80
100
0
20
40
60
80
100
also observe an overall simiGait
Cycle
(%)
Gait
Cycle
(%)
larity between the torque
(c)
(d)
profiles derived from the
DLMP- and NLMP-based
2
2
controllers. Again, this was
1
1
expected, since healthy
subjects have very similar ki0
0
nematics to the ideal ones
0
20
40
60
80
100
0
20
40
60
80
100
that were used to build the
Gait
Cycle
(%)
Gait
Cycle
(%)
primitives. Nevertheless,
some differences can be
(e)
(f)
found between both approaches. Focusing on the Figure 8. The reference torques of all joints during the 2 min of steady-state walking, normalized
hip, for example, the DLMP- to BW: (a) left hip, (b) right hip, (c) left knee, (d) right knee, (e) left ankle, and (f) right ankle. The
colored solid lines represent the different trials the subjects underwent. The dashed black lines
based approach provided display data from [23] from the literature for comparison. Positive values represent extension, and
larger flexion torque than the negative values represent flexion. This figure shows data averaged across healthy participants.
NLMP-based approach, while
the opposite holds for the extension torque. In addition, we
Oxygen Cost (Walk-Sit)
can observe some divergence in the hip profiles after toe-off
115
(between 65 and 85% of the gait cycle), where the NLMP110
based control output shows a steep decrease in flexion
105
100
followed by a plateau around zero torque, and the DLMP95
based control output shows a smoother transition between
90
flexion and extension torques.
85
Oxygen Cost
From the VO 2 uptake, the normalized oxygen cost of transportation was computed using (3) and the methods reported
in the "Data Processing" section.
A possible fatigue effect within the 6-min WT was discarded because, for all considered subjects, there was no clear
increasing trend of O2 consumption with time during the trial
(see, for example, the records for S2 in Figure 5).
Four of the seven subjects benefited from all the assistive
trials. Two (S2 and S9) benefited from three of the four
trials and displayed a small increment in the last one (less
than 10%). Actually, this trial was the last to be tested for
each of these two subjects, so they could have been impacted by fatigue across trials. One subject (S7) did not display
a reduction in metabolic cost in any of the assistive conditions. For each subject that benefited from assistance, there
was at least one assisted trial with more than

80
75
70
65
60

NO

TM

LTA

HTA

LSA

HSA

Figure 9. The oxygen rate mean values and standard deviations
of the overall group for each of the trials, normalized to the TM.

10% improvement with respect to TM. S1, S3, and S4 even
presented trials displaying a decrease in oxygen cost with
respect to the NO trial. In general, the assistive trials
managed to bring the subjects' metabolic rates to a condition in between NO and TM, showing a decrease with
respect to TM but still not enough to compensate for the
payload of wearing the device.
The average cost data are shown in Figure 9, which displays a general increase in oxygen cost when moving from the
march 2016

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

93
Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2016
