Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021

https://doi.org/10.1007/s40313-021-00863-1

Design and Performance Analysis of a Compact Series Elastic Actuator

for Exoskeletons
F. H. M. de Gaitani1,2 · W. M. dos Santos3 · A. A. G. Siqueira4

Received: 23 April 2021 / Revised: 27 October 2021 / Accepted: 3 November 2021 / Published online: 3 January 2022
© Brazilian Society for Automatics--SBA 2021

Abstract
In this paper, a compact Series Elastic Actuator is designed and evaluated. The proposed actuator will be used to drive the joints
of the modular exoskeleton for lower limbs ExoTAO (Transparent Active Orthosis), available at the Robotic Rehabilitation
Laboratory of EESC/USP. This equipment is being developed to assist in the rehabilitation of people with motor disabilities
resulting from injuries to the central nervous system or to the spinal cord. The differential of the proposed actuator is the use of
a customized and compact torsion spring, providing the necessary levels of torque and velocity to drive the exoskeleton. Such
performance levels were evaluated considering the characteristics of the selected components and the gait patterns available
in the literature. In addition to the theoretical analysis, experimental tests were also conducted with the actuator in order to
evaluate its performance in achieving a satisfactory torque control. Time and frequency responses show the proposed actuator
is able to reproduce the necessary torque to drive the exoskeleton ExoTAO, achieving high bandwidth values.

Keywords Series elastic actuator · Exoskeleton · Robotic · Rehabilitation · Orthosis

1 Introduction has been intensified, as shown by Kwakkel et al. (2008),

Prange et al. (2009) and Laut et al. (2016). The rehabilitation
Over the years, robotics has made great progress in the med- of patients who have suffered a stroke or incomplete spinal
ical field. The use of robots has added improvements in the cord injury are examples of the application of this therapy
work of doctors and other health professionals, in addition, (Mekki et al., 2018). The most common exercises adopted in
to provide a better experience for patients. These machines the robotic therapy are the passive motion, the active-assisted
can help both in performing surgeries and in the rehabilita- movement, and the active-resisted movement (Tong & Hu,
tion of patients. The use of robot-assisted therapy techniques 2008). The combination of these exercises can provide a bet-
ter control of the movements and an objective analysis of the
B F. H. M. de Gaitani evolution of the patient during the treatment.
[email protected] For a good performance of rehabilitation robots, it is nec-
W. M. dos Santos essary that some fundamental characteristics are present in
[email protected] their design. It is essential that force/torque or impedance
A. A. G. Siqueira controllers are implemented in these robots, since physical
[email protected] human–robot interaction is an intrinsical characteristic of the
1 Department of Mechanical Engineering, Engineering School application, resulting in force/torques being applied by both
of São Carlos, University of São Paulo, 400, Trabalhador agents. Hence, controlling such forces/torques ensures the
São-Carlense Avenue, São Carlos, SP 13566-590, Brazil safety and a proper level of physical interaction. In this case,
2 São Paulo State University (UNESP), 505, Profa. Isette Corrêa it is possible to control the interaction between the robot and
Fontão Avenue, São João da Boa Vista, SP 13876-750, Brazil the patient, so that the resistance of the robot to the patient’s
3 Universidade Estadual Paulista Júlio de Mesquita Filho, movement can be controlled, making the movements easier
Câmpus São João da Boa Vista Av. Profa. Isette Correa or harder to be done.
Fontão, 505 – Jardim das Flores, São João da Boa Vista, SP Considering these prerequisites, series elastic actuators
13876-750, Brazil
(SEAs) are the ideal solution for the purposed exoskele-
4 Av. Trabalhador São-Carlense, 400. Parque Arnold ton, because they ensure greater safety in the physical
Schmidt, São Carlos, SP CEP 13566-590, Brazil

123
Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021 1013

human–robot interaction and allow compliant control, due

to the spring system connected in series between the motor
reduction system and the load. In these actuators, firstly
described by Pratt and Williamson (1995), the springs decou-
ple the inertia of the motor and other nonlinearities from the
system output and isolating it from shocks.
In Kong et al. (2009), a rotary SEA (rSEA) was developed
to assist the movement of lower limbs. The main difference
between SEA and rSEA is the presence of a torsional spring.
Besides the spring, this actuator presents a direct current
motor, a planetary gearbox, and two potentiometers to detect
the deformation in the spring and the position of the output
shaft. For this arrangement (a spring between the motor and
the load), it is necessary that the spring supports the large
torques that will be generated, and at the same time has a low
stiffness.
The use of springs with very high stiffness compromises
the system because it diminishes the performance of force or
toque control (Pratt & Williamson, 1995). Besides that, the
nonlinearities of a spring with high stiffness can not simply
be overlooked. An alternative method to overcome these dif-
ficulties and disadvantages is to elaborate customized elastic
elements (Irmscher et al., 2018). These elements can present
a great mechanical resistance without increasing their stiff-
ness value. Fig. 1 Exoskeleton ExoTAO
Several SEAs have already been developed in literature,
considering the alternative mechanical designs of the cus-
tomized springs (Carpino et al., 2012; Chen et al., 2017; dos EESC/USP, focusing on making it more compact without
Santos et al., 2017; Lagoda et al., 2010; Paine et al., 2015). In reducing its mechanical resistance. In this way, the exercises
the designed actuator this element was properly dimensioned and the rehabilitation will be more efficient and the results
to resist the necessary torques of the walking movement. tend to be more satisfactory.
As shown by del Carmen Sanchez-Villamañan et al. Compared with the rSEA designed and evaluated in dos
(2019), most of the lower limb exoskeletons use Series Elas- Santos et al. (2017), the proposed rSEA has a smaller volume
tic Actuators. Comparing this actuator with others developed and mass, a higher value of combined stiffness, less back-
in the literature with a customized torsional spring can be lash, and a customized springer modeled to have symmetrical
highlighted its weight (1.6 kg), which is very low com- characteristics and higher mechanical resistance. Regarding
pared with many actuators already developed and its relation the torque control performance, the proposed rSEA achieved
between the mass and the peak torque, that for this actuator higher values of bandwidth.
is 0.032. A smaller value compared with the SEA developed The paper is organized as follows: Sect. 2 shows the the-
by dos Santos et al. (2017). oretical modeling of the SEA, the Sect. 3 explains how the
This work presents the design and performance analysis series elastic actuator was designed. Section 4 deals with the
of a compact Series Elastic Actuator, which can be used to experimental tests made in the actuator and finally, in Sect. 5
drive the robotic exoskeleton of lower limbs ExoTAO, avail- are made some conclusions about the work.
able at the Robotic Rehabilitation Laboratory of EESC/ USP,
Fig. 1. The actuator design consists of structural calculations
and dimensioning of components based on the requirements
of necessary torque and velocity of a standard gait pattern. 2 Theoretical Modeling
The performance analysis includes the structural analysis of
the customized elastic element, the analysis of the actua- In view of the observations and disadvantages of some meth-
tor’s torque and velocity characteristics, and the application ods described in the Sect. 1, a new rotary Series Elastic
of a torque control to the actuator. The intention is to make Actuator was developed. In this section, it will be described
this actuator an alternative to the existing ones, including how the theoretical analysis was carried out, in order to later
those developed at the Robotic Rehabilitation Laboratory of assemble the SEA.

123
1014 Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021

Fig. 3 Static analysis of Von Mises tension in the customized torsional

spring

Fig. 2 Customized torsional spring developed in the project

2.1 Customized Torsional Spring

The elastic component of the actuator has to be resistant

to support high torques; however, it needs to present low
stiffness for an effective force control. Thus, a new torsional
spring was developed, Fig. 2. For this spring, the transfer of
torque occurs between two rings with different diameters,
inserting flexing elements between these two rings, which
will characterize the spring stiffness.
The external ring will be fixed to the wheels of the anti-
Fig. 4 Static analysis of displacement in the customized torsional spring
backlash worm-wheel gearbox, while the inner ring will be
fixed with a key to the output shaft, which will be connected
to the exoskeleton ExoTAO. In the static simulation, first of all, the torque applied in
This design was made taking into account the necessity the spring was calculated using:
of reducing the weight of the spring, without affecting its
mechanical characteristics, such as strength and stiffness.
The model was based in the work of dos Santos et al. (2017), T r×F (1)
trying to eliminate the asymmetry and modeling the stiffness
varying the internal ratios and the thickness of the spring. where T is the torque applied in the spring, r is the ratio of
Static analyzes were performed in the elastic elements the spring, which is 36 mm, and F is the force programmed
to estimate the maximum strength and the stiffness of the to act in the analyzes.
spring. The tests are shown in Figs. 3 and 4. To perform the Thus, by making tests with different forces and comparing
analysis, a force was modeled to act in the external ring of with the maximum tension of the material (7 × 108 Pa), the
the spring, while the center, where the shaft will be assembly, result was that the spring supports a torque of up to 25 N.m.
was modeled as a fixed support. All analyzes were performed In Fig. 3, there is the test with a 700 N force, which equals
using Solidworks software. to 25.2 N.m torque. The results have shown a 5.982 × 108 Pa

123
Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021 1015

tension, which is supported by the spring with a 1,17 safety

factor.
The displacement result shown in Fig. 4 was used for the
computation of the spring stiffness. To this analysis, a force
of 500 N was applied, and then, to reach the required value,
several equations were necessary. First, Eq. (2) was used to
convert the linear deformation into an angle, in radians:

θ  d/r (2)

where θ is the deformation angle, in radians, d is the maxi-

mum deformation calculated by static analyzes, in mm, and
r is the same as in (1). After having the angle of deformation,
the elastic constant of the spring was estimated as:

K  T /θ (3) Fig. 5 Torque x angular velocity for the hip joint

where K is the elastic constant of the spring, in N.m/rad,

while T and θ were defined before. meet the characteristics of the walking movement was also
With the calculations properly performed, it was possible computed.
to determine that the custom spring has an elastic constant For this, data of a standard gait provided by Kirtley (2006)
of approximately 130 N.m/rad. It is noteworthy that several were used. They include the position profiles for the ankle,
analyzes with different forces were made, resulting in sev- knee, and hip joints. After performing an inverse dynamic
eral elastic constants. The mean value of these results was calculation to obtain the necessary torques at the motor side,
considered as the estimated stiffness. we compared them to the torques provided by the motor used
Finally, the stiffness and resistance of the set of the springs in the designed actuator.
connected in parallel were calculated. For this, were used (4) For the calculations, some values were defined, such as
and (5): efficiency (ε  0.7) and reduction ratio (r  150) of the
gearbox. The angular velocity of the motor (ωmotor ), in rpm,
Tmax,assembly  2 × Tmax,spring (4) to each joint, was computed as:

where Tmax,assembly refers to the maximum torque supported ωmotor  ω × r (6)

by the set of springs in parallel and Tmax,spring refers to the
maximum torque supported by one spring, and as defined where ω is the angular velocity, in rpm, for each joint, as
before worth 25 N.m. Thus, it was calculated that the maxi- described in Kirtley (2006). Besides the angular velocity, the
mum torque which the set can support is 50 N.m. It was used torque of the motor for each joint (Tmotor ), in mN.m, was
the springs connected in parallel with each other to increase calculated as:
the stiffness of the set of springs.
Finally, the elastic constant of the set of springs is given Tmotor  T /(r × ε) (7)
by:
where T is the torque made by each joint, in mN sm, accord-
K assembly  2 × K (5) ing to Kirtley (2006).
Thereby, it was possible to generate graphs for each joint
where K assembly is equal to 260 N.m/rad. of the motor required torque, in mN.m, versus angular veloc-
Therefore, the designed set of springs has suitable charac- ity, in rpm. In all graphs, the ideal curve of the motor (green
teristics for its use in the rSEA, as it has high resistance and line) and the maximum continuous torque (blue dashed line)
a relatively low stiffness, providing a better precision in the were drawn. In Fig. 5, the result for the hip joint can be seen,
measurement of the applied force. in Fig. 6 for the knee and in Fig. 7 for the ankle.
As it can be seen, for the hip and knee joints, the torque
2.2 Motor Torque Analyzes remains mostly within the ideal range of continuous torque
(between the dashed lines), with peaks at certain velocities.
In addition to the calculation of the customized torsional Therefore, it is possible to use the motor to control these two
spring’s characteristics, the torque required by the motor to joints.

123
1016 Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021

Fig. 6 Torque x angular velocity for the knee joint

Fig. 8 rSEA developed in the project

As can be seen in Fig. 8, the actuator has three shafts (c, h

Fig. 7 Torque x angular velocity for the ankle joint and m) that were properly dimensioned to withstand torsional
efforts. In addition to the shafts, the actuator has two bearings
(b) to reduce the friction and allow the rotation of the shaft.
However, for the ankle joint, the torque is mostly above The fixing of the springs and the wheels of the anti-backlash
the maximum continuous torque supported by the motor. It worm-wheel gearbox to their respective shafts were made by
means that the actuator can be used in the ankle joint only keys.
at certain periods of the walk. As the exoskeleton will be The shaft (m) in Fig. 8, is connected to the motor by an
used for rehabilitation, this limitation can be adjusted as the elastic coupling, holding the worm in the ideal position for the
patient walks as needed. correct functioning of the anti-backlash worm-wheel gear-
box. The gearbox was selected to have a 150:1 ratio. The
motor is fixed to the actuator housing by an element designed
3 Mechanical Design specifically for this purpose (e).
The spring system was configured to work in parallel,
With these theoretical analyses performed in the Sect. 2, it i.e., the springs are connected at the same input and output
is possible to conclude that the actuator has suitable charac- shaft, doubling the torque resistance and the stiffness when
teristics to work in the exoskeleton joint. Therefore, in this compared to only one spring.
section, it will be described the operation and its components. Finally, an incremental encoder (i) was used to effectively
Figure 8 shows the design of the actuator and its components measure the position of the output shaft.
and Table 1 relates the letters and the name of each compo- According to the design requirements, the actuator should
nent and gives the specification of the purchased elements. be compact and in Fig. 9 it is possible to see the dimensions
of this new actuator.

123
Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021 1017

Table 1 Components and specifications

Components Letter Specification

Lid (a)
Bearing (b) SKF – 3200 ATN9
Spring shaft (c)
Bearing house (d)
Motor support (e)
Motor (f) RE40 Maxon
Elastic coupling (g)
Wheel shaft (h)
Encoder (i) THI40N-1CAK2R6TN-00500
Encoder support (j)
Housing (l)
Motor shaft (m)
Motor bearing (n)
Wheel (o) HPC ABM0.5–150
Springs (p)
Worm HPC W0.5–1

Fig. 10 Dimensions of the customized torsional spring (dimensions in

mm)

Fig. 9 Dimensions of the compact SEA (dimensions in mm)

3.1 Customized Torsional Spring

Fig. 11 Final assembly of the SEA
The elastic component of the actuator, as cited in Sect. 2, has
a stiffness of 130 N.m/rad and resists a torque of up to 25 N.m.
This spring has a weight of 0.086 kg and its dimensions are 4 Experimental Results
represented in Fig. 10.
4.1 Experimental Evaluation of the Spring Stiffness
3.2 Final Assembly
With the actuator already machined, an experiment was car-
Considering the developed design, the proposed actuator was ried out in order to measure the actual stiffness value of the
machined in a tooling shop dedicated to its manufacture. spring. For the experiment, it was used the industrial robot
The external elements, (lid (a), motor support (e), encoder KUKA KR16. This robot has six joints that can be moved
support (j), and housing (l)) are made of Aluminum, due to independently. To perform the experiment, the actuator was
its low density and great resistance. The actuator’s shafts, fixed at a table in the range of the robot KUKA and its out-
are of Steel 4340, due to its high resistance. The customized put shaft was coupled to the force sensor attached to the end
springs are made of Steel 1.8159, an alloy suitable for use in effector of the robot, as can be seen in Fig. 12.
springs. With this configuration, a program was implemented using
Figure 11 shows the internal part of the actuator with the system language of the robot, called KRL (KUKA Robot
the customized elastic elements that were developed and the Language). The program was made to active only the A6
actuator assembled together with the exoskeleton joint. joint of the robot (Fig. 13) and rotates this joint 2.5 degrees

123
1018 Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021

Fig. 12 Experimental setup for spring’s stiffness estimation

Fig. 14 Torque x deformation of the spring

Fig. 13 A6 joint of the industrial robot KUKA

Fig. 15 Linear regression in the counterclockwise

in clockwise and 2.5 degrees in counterclockwise. While the

robot turns its joint, its force sensor measures the applied of the setup of the experiment, to avoid initial backlashes of
force and generates a graph of torque by position. This graph the actuator.
can be seen in Fig. 14. Comparing the results obtained with the simulations and
To calculate the stiffness of the spring, it was made a lin- with the experimental test it is possible to conclude that
ear regression in this graph and the straight slope will be the simulations were very faithful to the real model and
the spring stiffness. As can be seen in Figs. 15 and 16, two approached the actual result.
different values of stiffness were found, one for the counter-
clockwise (Fig. 15) and another for the clockwise (Fig. 16). 4.2 Torque Control
The experimental stiffness of the spring is 214.4 N.m/rad in
counterclockwise and 331.3 N.m/rad in clockwise. A mean To perform the tests of torque control, the actuator was fixed
value can be computed as 272.85 N.m/rad. The difference at a table, as well as the rotative shaft of the exoskeleton joint.
between the stiffness in counterclockwise and in clockwise In this way, the torque can be measured by the deformation
can be justified due to a misalignment between the shafts of of the spring, using two encoders. One of these encoders was
the force sensor and of the actuator. Another relevant factor is already cited in Sect. 3 and has a resolution of 2000 pulses per
the separation of the test into two parts, one for the CW direc- revolution in quadrature decoding mode. The other encoder
tion and the other for the CCW direction, due to limitations was connected to the motor to measure its rotation and esti-

123
Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021 1019

Fig. 16 Linear regression in the clockwise

Fig. 17 Torque x time for positive torque

mate the position of the worm-wheel gearbox. This encoder

has a resolution of 4096 pulses per revolution in quadra-
ture decoding mode. The spring deflection can be estimated
by measuring the difference between these two positions.
Besides that, the torque resolution can be calculated using:

r T  K × (2 × π /2000) (8)

where r T is the torque resolution and K is the spring stiffness.

So, the resolution is 0.86 N.m.
To implement the torque control in the compact SEA, a C +
+ language program was used through the Microsoft Visual
Studio program present on the laboratory’s computer and
which was already used in the exoskeleton ExoTAO (Jardim,
2009). Thus, a PI (Proportional Integrative) torque control
was implemented.
With the program, it was possible to define proportional Fig. 18 Torque x time for negative torque
and integrative gain values. For the correct realization of the
control method, the output shaft of the actuator was fixed, pre-
venting its movement. The stiffness of the spring to perform ure 21 shows a zoom-in one of the peaks of the graph. These
the tests was used as the mean value, that worth 272.85 N.m, discontinuities can be due to small backlashes in the system
as calculated in Sect. 3. since with higher spring stiffness, a small gap can represent
First of all, the gains values were adjusted, these values a large variation in torque.
were 200 for the proportional gain and 4 for the integrative To finish the tests, a frequency response function (FRF)
gain. In the initial tests, a step input for the reference torque of was also performed, using a chirp signal of 0 to 20 Hz and
5 N.m and 10 N.m were defined. Figure 17 shows the results. an amplitude of 5 and 10 N.m. In Fig. 22, the results of
In the second test, the same input was carried out, but this the frequency response can be seen. It is important to note
time with -5 N.m and -10 N.m, the results can be seen in that the higher the torque the lower the bandwidth of the
Fig. 18. In the test 3 and 4, sinusoidal inputs with amplitudes actuator. For the amplitude of 5 N.m, the bandwidth value
of 5 N.m and 10 N.m were carried out, and respectively, was approximately 14.5 Hz, and for the amplitude of 10 N.m,
shown in Figs. 19 and 20. the bandwidth value was approximately 12.4 Hz.
As it can be seen in the graphs, the compact SEA was Compared with the rSEA designed and evaluated in dos
able to follow the desired torque curve with certain accuracy. Santos et al. (2017), the proposed rSEA achieved higher val-
Small discontinuities in the actual torque curve are also noted ues of bandwidth. For the amplitude of 5 N.m, the bandwidth
in the results of the sinusoidal torque (Figs. 19 and 20). Fig- value for the rSEA described in dos Santos et al. (2017) was

123
1020 Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021

Fig. 21 Zoom on sinusoidal torque (10 N.m)

Fig. 19 Torque x time for sinusoidal torque (5 N.m)

Fig. 20 Torque x time for sinusoidal torque (10 N.m)

Fig. 22 Frequency response of the compact SEA

at most 9.6 Hz. A higher bandwidth is achieved since the

stiffness of the new actuator is higher.
control system, are able to perform a good torque control for
the use of the actuator in the exoskeleton.
5 Conclusions Acknowledgments An early version of paper was presented at XXIII
Congresso Brasileiro de Automática (CBA 2020). This work is sup-
In this paper, was designed and performed a rotational Series ported by São Paulo Research Foundation—FAPESP under grants num-
ber 2019/23413-5 and 2019/05937-7 and by the Coordenação de Aper-
Elastic Actuator that will be used to drive the joints of the feiçoamento de Pessoal de Nível Superior—Brasil (CAPES)—Finance
exoskeleton ExoTAO. There were made several theoretical Code 001, as part of the Programa de Apoio à Pós-Graduação e à
analysss and experiments were to evaluate its performance. Pesquisa Científica e Tecnológica em Tecnologia Assistiva no Brasil
Therefore, from the design and performance evaluation, (PGPTA), Grant Number 3457/2014.
it can be concluded that the rotational Series Elastic Actua-
tor that was developed can be used to compose the ExoTAO Declarations
exoskeleton, fulfilling its main functions. The motor is con-
sistent with the actuation for whatever the exoskeleton joint, Conflict of interest The authors declare that they have no conflict of
while the elastic element, together with the encoders and the interest.

123
Journal of Control, Automation and Electrical Systems (2022) 33:1012–1021 1021

References Mekki, M., Delgado, A. D., Fry, A., Putrino, D., & Huang, V. (2018).
Robotic rehabilitation and spinal cord injury: A narrative review.
Carpino, G., Accoto, D., Sergi, F., Luigi Tagliamonte, N., Guglielmelli, Neurotherapeutics, 15(3), 604–617.
E. (2012). A novel compact torsional spring for series elastic actu- Paine, N., Mehling, J. S., Holley, J., Radford, N. A., Johnson, G.,
ators for assistive wearable robots. Journal of Mechanical Design Fok, C. L., & Sentis, L. (2015). Actuator control for the nasa-
134(12) jsc valkyrie humanoid robot: A decoupled dynamics approach for
Chen, B., Zhao, X., Ma, H., Qin, L., & Liao, W. H. (2017). Design and torque control of series elastic robots. Journal of Field Robotics,
characterization of a magneto-rheological series elastic actuator 32(3), 378–396.
for a lower extremity exoskeleton. Smart Materials and Structures, Prange, G., Jannink, M., Groothuis-Oudshoorn, C., Hermens, H., &
26(10), 105008. Ijzerman, M. (2009). Systematic review of the effect of robot-aided
del Carmen Sanchez-Villamañan, M., Gonzalez-Vargas, J., Torricelli, therapy on recovery of the hemiparetic arm after stroke. Journal
D., Moreno, J. C., & Pons, J. L. (2019). Compliant lower limb of Rehabilitation Research and Development, 43(2), 171–184.
exoskeletons: A comprehensive review on mechanical design prin- Pratt, G. A., & Williamson, M. M. (1995). Series elastic actuators. In:
ciples. Journal of Neuroengineering and Rehabilitation, 16(1), 55. Proceedings 1995 IEEE/RSJ international conference on intelli-
dos Santos, W. M., Caurin, G. A., & Siqueira, A. A. (2017). Design and gent robots and systems. human robot interaction and cooperative
control of an active knee orthosis driven by a rotary series elastic robots, IEEE, Vol. 1, pp. 399–406
actuator. Control Engineering Practice, 58, 307–318. Tong, R. K. Y., & Hu, X. (2008). Service robotics: Robot-assisted train-
Irmscher, C., Woschke, E., May, E., & Daniel, C. (2018). Design, opti- ing for stroke rehabilitation. Service Robot Applications, 107
misation and testing of a compact, inexpensive elastic element
for series elastic actuators. Medical Engineering & Physics, 52, Publisher’s Note Springer Nature remains neutral with regard to juris-
84–89. dictional claims in published maps and institutional affiliations.
Jardim, B. (2009). Atuadores elásticos em série aplicados no desenvolvi-
mento de um exoesqueleto para membros inferiores. PhD thesis,
Universidade de Sao Paulo
Kirtley, C. (2006). Clinical gait analysis: theory and practice. Elsevier
Health Sciences
Kong, K., Bae, J., Tomizuka, M. (2010). A compact rotary series
elastic actuator for knee joint assistive system. In: 2010 IEEE
international conference on robotics and automation, IEEE,
pp. 2940–2945
Kong, K., Bae, J., & Tomizuka, M. (2009). Control of rotary series
elastic actuator for ideal force-mode actuation in human–robot
interaction applications. IEEE/ASME Transactions on Mechatron-
ics, 14(1), 105–118.
Kwakkel, G., Kollen, B. J., & Krebs, H. I. (2008). Effects of robot-
assisted therapy on upper limb recovery after stroke: A systematic
review. Neurorehabilitation and Neural Repair, 22(2), 111–121.
Lagoda, C., Schouten, A. C., Stienen, A. H., Hekman, E. E., & van
der Kooij, H. (2010). Design of an electric series elastic actuated
joint for robotic gait rehabilitation training. In: 2010 3rd IEEE
RAS & EMBS international conference on biomedical robotics
and biomechatronics, IEEE, pp. 21–26
Laut, J., Porfiri, M., & Raghavan, P. (2016). The present and future of
robotic technology in rehabilitation. Current Physical Medicine
and Rehabilitation Reports, 4(4), 312–319.

123