IEEE Robotics & Automation Magazine - March 2016 - 94

NO trial to the TM trial of about 22%. This increase was
mainly due to the burden of wearing the 8.5-kg device, which
was originally intended to be coupled to a leg orthosis and/or
prosthesis, which could transfer a part of its weight directly to
the ground. Nevertheless, a decrease in cost with respect to
TM can be observed in any of the assistive modes, although
standard deviations reveal large variability in the subject's adaptation during the different trials. t-tests of LTA, LSA, and
HSA trials with respect to TM revealed significant reductions
^ p = 0.0146, p = 0.0409, and p = 0.0322, respectively).
Only HTA did not reach significance ( p = 0.2704 ). The oxygen cost decrease was the largest for the LSA and HSA trials,
where it reached about 50% of the extra cost caused by wearing the device.
Conclusion and Perspectives
In this article, a novel bioinspired controller based on motor
primitives for locomotion assistance was developed. Based on
data from the literature, a basic set of primitives was extracted
that could reconstruct joint torques (by three DLMPs) or muscle stimulations (by six NLMPs) for walking cadences between
0.6 and 1 Hz and for ascending and descending
Our method combined
stairs. The controller used
this reduced set of primidifferent mechanisms to
tive signals together with
information from a WSA,
continuously adapt the
i.e., the kinematics (joint
angles) and ground reacdelivered torques to the
tion forces (swing/stance),
to generate reference joint
subject's actual kinematics. torques. This reduced set
of sensors was directly attached to the user and,
therefore, independent of the robot being used. This, together
with the direct torque control output, makes the proposed
control approach easily transferable to any torque-controlled
assistive robot. This is especially true if we consider the
NLMP-based control, which implements a virtual-musculoskeletal model. This controller naturally induces some adaptation to the subject's particular kinematics through artificial
muscles but avoids the complex sensor deployment and signal
processing required by EMG.
To assess the assistive capability of both control approaches,
several experiments were conducted. This included two analyses: the capability of the assistance to decrease the oxygen cost
and the difference between using DLMP directly outputting
torques or using NLMP and a musculoskeletal model.
Results on healthy walkers displayed between a 20 and
30% increase in metabolic cost when switching from the NO
trial to TM, i.e., when the wearer is able to walk without encountering resistive actions from the orthosis. This reveals
that the device weight had a significant negative impact on
the subject cost of transport, despite all the efforts made to
maximize its transparency. Nonetheless, the tested assistive
methods proved to assist the subjects by reducing the oxygen
94

IEEE ROBOTICS & AUTOMATION MAGAZINE

march 2016

cost with respect to TM, reaching a significant level for three
of the four assistive trials. Therefore, it may be forecast that
using a lighter exoskeleton would bring this reduction down
to lower values than NO.
Our method combined different mechanisms to continuously adapt the delivered torques to the subject's actual kinematics. First, DLMPs were time-scaled as a function of the
gait phase and frequency, being continuously and independently monitored for each leg by AOs. This opens the possibility to deliver nonsymmetrical patterns, a desirable feature
for patients with hemideficits. Then, the primitives were
multiplied by frequency-dependent weights, guaranteeing
the adaptation of the torque profile to the current walking
cadence. On top of that, NLMP generated stimulations that
entered into a musculoskeletal model. This model captured a
bioinspired muscle impedance, so that torques were generated not only as a function of the gait phase and walking frequency but also of the particular subject's kinematics. In
brief, the generated torque adapts to the natural walking pattern of each different subject. Therefore, the NLMP approach
should be more convenient for patients with pathologies
causing the natural way of walking to differ from the ideal
one of healthy subjects. Last but not least, this approach is
more prone to integrate other bioinspired stimulations, such
as reflexes or postural control [14].
Future work will focus on similar experiments with patients
displaying a pathological gait to better assess the relevance of
the NLMP-based method to adapt to each particular gait.
Finally, the system can easily be adjusted to convey the assistive strategy of a full orthosis or an active prosthesis. This
can be done by following the same principles as the APO assistive strategy and adapting the outputs to the necessary control inputs of the device, if not directly torque driven.
Acknowledgments
The authors would like to thank the Fondazione Don Carlo
Gnocchi and especially Guido Pasquini and Federica Vannetti
for their support with the recruitment of subjects, installations, and advice during the experiments.
This work was supported in part by the European Union
within the CYBERLEGs project (FP7-ICT-2011-2.1 Grant
287894), by Fondazione Pisa within IUVO project 154/11,
and the Belgian F.R.S.-FNRS (crédits aux chercheurs 6809010
to R. Ronsse, and travel grant to V. Ruiz Garate).
The two senior authors, listed last, contributed equally to
cosupervise this work.
References
[1] World Population Ageing, Department of Economic and Social Affairs,
Population Division, United Nations, 2013.
[2] J. Verghese, A. LeValley, C. B. Hall, M. J. Katz, A. F. Ambrose, and R. B.
Lipton, "Epidemiology of gait disorders in community-residing older adults,"
J. American Geriatr. Soc., vol. 54, no. 2, pp. 255-261, 2006.
[3] T. Lenzi, M. C. Carrozza, and S. K. Agrawal, "Powered hip exoskeletons
can reduce the users hip and ankle muscle activations during walking," IEEE
Trans. Neural Syst. Rehab. Eng., vol. 21, no. 6, pp. 938-948, 2013.

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