IEEE Robotics & Automation Magazine - March 2016 - 89

Many of the stimulations actually received
Table 2. The participants' features and trial ordering.
a negligible contribution for some of the
primitives (i.e., several weights were very
Subject Gender Age Weight (kg) Order of Assistive Trials
close to zero). Any coefficients that were at
30
85
TM - LTA - HSA - LSA - HTA - NO
S1
Male
least ten times smaller than the most influ30
76
TM - LSA - LTA - HSA - HTA - NO
S2
Male
ential one were rounded to zero. In addi27
87
TM - LSA - LTA - HSA - HTA - NO
S3
Male
tion, the weight evolutions across the five
35
68
TM - LTA - HSA - HTA - LSA - NO
walking cadences were interpolated using
S4
Female
second-order polynomials.
66
NO - LSA - TM - LTA - HSA - NO - HTA
S5*
Female 33
Interestingly, the number of significant
35
82
NO - LTA - HTA - LSA - HSA - TM
S7
Male
primitives found is similar to the literature
S9
Male
22
73
NO - LTA - HTA - TM - LSA - HSA
reports that extracted primitives from real
*For this subject, a technical issue caused the recording to stop before the last trial. An extra
EMG data [18], [19]. In these studies, the
session was thus organized, with HTA as a unique assisted trial, along with an extra NO trial
authors took a larger number of muscles
for normalization of oxygen uptake.
into account. However, they analyzed
mainly walking and running, while we introduce very different locomotion modes, that is, ascending motor intention. This was achieved by using a wearable senand descending stairs.
sory apparatus (WSA) and the methods reported in [30].
To synchronize the control primitives with the actual gait
Experimental Validation
phase of the subject, AOs were used [22], [31]. In this experiA preliminary validation experiment of the primitive-based ment, the hip angles were used as the oscillators' inputs, procontroller was conducted, focusing on the walking maneu- viding the walking frequency (i.e., the gait cadence in hertz)
ver. To deliver the desired assistive torques, the active pelvis and the instantaneous
orthosis (APO) developed within the CYBERLEGs project phase of each leg (see FigThe NLMP approach should
was used [28]. This device is an advanced version of a previ- ure 1). The same primious laboratory prototype presented in [29]. It has two actua- tives were used for both
be more convenient for
tion units that employ a series elastic actuator to actively legs, while a separate AO
drive the hip flexion/extension movement of each leg. (The was used for each side.
patients with pathologies
design of this actuation system is under patent application: Consequently, the move"Sistema di Attuazione per Ortesi di Anca," Italian patent ap- ment phase and frequency
causing the natural way
plication FI/2015/A/000025.) Consequently, assistance was were independently deprovided to both hips for flexion/extension, while the com- tected for each leg and
of walking to differ
puted assistive torques for the knee and ankle were not deliv- synchronized to the correered to the participants. The APO weighs about 8.5 kg. It was sponding heel-strike infrom the ideal one of
used as a stand-alone device in this experiment, although it stants. This approach
was originally designed to be coupled to a leg exoskeleton allowed the delivery of
healthy subjects.
and/or a prosthesis, which could directly transfer part of its nonsymmetrical torque
weight to the ground. Consequently, it embeds the electron- profiles, since the leg freics and battery for actuating more joints than the hip only, quencies were not conexplaining why it is heavier than concurrent solutions.
strained to be equal, and the heel-strike dephasing was not
Tests were conducted with both proposed bioinspired meth- constrained to be 180˚.
ods: the DLMP-based control with direct torque outputs [Figure
1(a)] and the NLMP-based control coupled to the muscluloskel- Participants
etal layer [Figure 1(b)]. A fraction of the desired assistive torques Nine healthy participants (S1-S9) took part in the experigenerated by either method was delivered to the joints as a func- ment: two females and seven males. The participants were
tion of the desired level of assistance.
between 22 and 35 years old and weighed between 66 and 85
In the case of the NLMP, the musculoskeletal model was kg. The participants signed a written consent form before
adapted to each subject. To do so, numerical parameters of the starting the experiments. Two subjects were discarded from
muscle units and of the skeletal model were scaled as a func- the analyses due to problems in oxygen consumption recordtion of the subject's anthropometry. The muscles attachment ings: S8 because of technical issues and S6 because he adoptlever, optimal, and slack lengths were modulated proportional- ed a highly inconsistent behavior across the assisted trials. In
ly to the subject's height, while the maximal isometric force particular, he increased and decreased his walking speed in
was modulated proportionally to the square of the height.
some of the assisted trials by 10-20%. This led to a high imFurthermore, to assist the participants with the appropri- pact on his metabolic consumption, which significantly inate torque profiles (i.e., the appropriate combination of the creased. All the other subjects (including S8) walked
primitives and set of weights), it was necessary to infer their consistently across trials.
march 2016

IEEE ROBOTICS & AUTOMATION MAGAZINE

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