ORIGINAL RESEARCH

published: 04 December 2018

doi: 10.3389/fnbot.2018.00080

The Challenges and Achievements of

Experimental Implementation of an
Active Transfemoral Prosthesis
Based on Biological Quasi-Stiffness:
The CYBERLEGs Beta-Prosthesis
Louis Flynn 1*, Joost Geeroms 1 , Rene Jimenez-Fabian 1 , Sophie Heins 2 ,
Bram Vanderborght 1 , Marko Munih 3 , Raffaele Molino Lova 4 , Nicola Vitiello 4,5 and
Dirk Lefeber 1
1
Department of Robotics and Multibody Mechanics, Vrije Universiteit Brussel, and Flanders Make, Brussels, Belgium,
2
Center for Research in Mechatronics, Institute of Mechanics, Materials, and Civil Engineering, Institute of Neuroscience, and
Louvain Bionics, Université Catholique de Louvain, Louvain-la-Neuve, Belgium, 3 Robolab, Faculty of Electrical Engineering,
University of Ljubljana, Ljubljana, Slovenia, 4 Fondazione Don Carlo Gnocchi, Milan, Italy, 5 The BioRobotics Institute, Scuola
Superiore Sant’Anna, Pisa, Italy

The CYBERLEGs Beta-Prosthesis is an active transfemoral prosthesis that can provide

the full torque required for reproducing average level ground walking at both the knee
Edited by: and ankle in the sagittal plane. The prosthesis attempts to produce a natural level
Hyung-Soon Park,
Korea Advanced Institute of Science & ground walking gait that approximates the joint torques and kinematics of a non-amputee
Technology (KAIST), South Korea while maintaining passively compliant joints, the stiffnesses of which were derived from
Reviewed by: biological quasi-stiffness measurements. The ankle of the prosthesis consists of a series
Ming Liu,
elastic actuator with a parallel spring and the knee is composed of three different systems
North Carolina State University,
United States that must compliment each other to generate the correct joint behavior: a series elastic
Sang Hoon Kang, actuator, a lockable parallel spring and an energy transfer mechanism. Bench testing of
Ulsan National Institute of Science and
Technology, South Korea this new prosthesis was completed and demonstrated that the device was able to create
*Correspondence: the expected torque-angle characteristics for a normal walker under ideal conditions. The
Louis Flynn experimental trials with four amputees walking on a treadmill to validate the behavior of
lflynn@vub.ac.be
the prosthesis proved that although the prosthesis could be controlled in a way that
Received: 30 April 2018
allowed all subjects to walk, the accurate timing and kinematic requirements of the
Accepted: 08 November 2018 output of the device limited the efficacy of using springs with quasi-static stiffnesses.
Published: 04 December 2018
Modification of the control and stiffness of the series springs could provide better
Citation:
performance in future work.
Flynn L, Geeroms J,
Jimenez-Fabian R, Heins S, Keywords: prosthesis, transfemoral, active, knee, ankle, powered, quasi-stiffness, compliant
Vanderborght B, Munih M, Molino
Lova R, Vitiello N and Lefeber D (2018)
The Challenges and Achievements of
Experimental Implementation of an
1. INTRODUCTION
Active Transfemoral Prosthesis Based
on Biological Quasi-Stiffness: The
Current transfemoral prostheses are most often passive, modular systems that cannot generate joint
CYBERLEGs Beta-Prosthesis. work. From fully passive ankles such as the SACH foot (Staros, 1957), Energy-Storage-Return (ESR)
Front. Neurorobot. 12:80. (Hafner et al., 2002) carbon types, and fully passive knees, such as the Mauch knee (Mauch, 1968),
doi: 10.3389/fnbot.2018.00080 to current top of the line microcontroller knee prostheses (Otto Bock Genium and C-Leg, Ossur

Frontiers in Neurorobotics | www.frontiersin.org 1 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

Rheo Knee, Blatchford/Endolite Pleo, Freedom Innovations Plié, have a motor that can be engaged and disengaged using a clutch
among others), there are no systems that provide additional (Lenzi et al., 2015) or a transmission (Lenzi et al., 2018) when the
output energy; the only energy used for prosthesis propulsion is actuator is required, such as for sit-to-stand and stair climbing
energy that has been captured from the gait cycle. operations.
Providing positive work is an important aspect of the Here we have designed an active prosthesis, the CYBERLEGs
biological joint and there are new robotic designs that are Beta-Prosthesis (Figure 1), that uses compliant springs whose
capable of delivering it. Devices such as the OttoBock emPOWER stiffnesses correspond to the quasi-stiffness of the knee and ankle,
(previously iWalk, BionX, BiOM Au and Herr, 2008), or the which results in relatively soft series spring values. These soft
Össur/Springactive Odyssey (Hitt et al., 2008) ankles or the Össur springs theoretically reduce the energy consumption for the
Power Knee are available, or will soon be available, as commercial motors during normal walking by minimizing motor output
devices. While not widely prescribed at the moment, they are work (Geeroms et al., 2017, 2018), but can cause issues for
beginning to find use in the market. There are many reasons motor controller techniques due to the reduction of actuator
for believing that active ankle and knee propulsion provides bandwidth associated with low output stiffness. The intention is
benefit such as tests which have shown a reduction in loading to determine if passively following these quasi-stiffness behaviors
of the unaffected leg using a powered ankle (Grabowski and with a position based state machine controller provides normal
D’Andrea, 2013), reduction of the metabolic energy consumption level ground walking capability with low actuator effort while
of a transtibial amputee to the level of a non-amputee while using remaining controllable enough for individuals to ambulate and
robotic ankles (Herr and Grabowski, 2012; Caputo and Collins, perform other tasks.
2014), and simulations showing reductions in metabolic cost The knee torque angle characteristics were divided into three
below normal human walking (Handford and Srinivasan, 2016). gait regions, and using separate systems that were optimized for
A major drawback of all of these active propulsion systems particular portions of the gait cycle. The main knee actuator is
is the high electrical demand and increased actuator complexity a highly compliant series elastic actuator, which has the ability
of providing such work. This has led to a large variety in to change knee energy during all periods of the gait cycle. There
the designs of robotic prostheses, mainly differing from how is a Weight Acceptance (WA) mechanism to efficiently handle
passive compliance elements are used within the system in order stance flex in the knee directly after heel strike. Connecting the
to reduce total energy consumption and motor requirements. knee and ankle is a special Energy Transfer (ET) mechanism, a
Representing the fully active design principle, the Vanderbilt system built with the intention of utilizing captured work from
prosthesis only contains one spring, a 6 Nm/deg parallel spring the knee to assist the ankle motor in driving the ankle. This has
(Lawson et al., 2014) in the ankle that works during the been explored before in passive devices (Matthys et al., 2012; Unal
plantarflexion stage of the gait cycle. The operation of this device et al., 2013), but never in an active prosthesis. The combination
is fully driven through the use of motors and harmonic drives, of systems of the CYBERLEGs Beta-Prosthesis creates a highly
requiring the use of impedance controllers to create compliant passive system for normal level ground walking while remaining
prosthesis behaviors. The CSEA (Rouse et al., 2014) knee uses capable of providing the high torques and power output for high
a simple friction clutch to lock a series-elastic spring element energy output tasks. The result is a device that has a unique mix
which passively replicates the torque-angle characteristics during of passive and active capabilities, allowing efficient locomotion
the portion of the gait cycle directly after heel strike. Outside through passive behaviors, but is capable of actively driving the
of locking during this short period, the knee is driven with a joints when necessary. For a summary of other tasks such as sit-
stiff SEA. This is similar to the clutch and SEA arrangement in to-stand, obstacle avoidance, and stair climbing capabilities that
the Össur Power knee, which utilizes a dog clutch behind the were attempted with this device, please refer to Flynn et al. (2018).
harmonic drive which is capable of performing the same type
of function (Gilbert and Lambry, 2013), although it is unclear 1.1. Article Contribution
if the clutch is used in this manner. Both of these devices have Here we discuss the development of the CYBERLEGs Beta-
spring stiffnesses chosen to approximate the biological knee Prosthesis, the design of which requires four major systems to
quasi-stiffness during the early stance phase of the gait cycle. work together to produce the desired joint torque/angle output.
The ETH/Delft knee (ANGELAA) spent fine attention to the This device was first tested on the bench and found to reduce
arrangement of parallel and series elastic elements to passively motor energy consumption while generating expected torque-
match the actuator stiffness to the desired actual joint stiffness angle characteristics for a normal walker under ideal conditions.
based on simulation and gait studies (Pfeifer et al., 2014). This In amputee trials, four individuals were able to walk over
allows a passive minimization of actuator work necessary to level ground with the prosthesis using a simple state machine
provide desired output impedance. based controller. Because of the requirements of the complex
These devices all have added mechanical complexity and relationships of the four major systems, the output kinematics,
require additional control techniques to accurately detect gait and the behavior of the person using the device, the position
and activate control when compared to their passive equivalents. based control technique was not capable of producing the
The RIC Hybrid Knee Prostheses avoid some of this actuator desired output kinematics rendering the ET system ineffective.
complexity by removing the influence of the actuator on the In general, the use of the quasi-stiffness to determine series
normal gait cycle. The devices consists of a passive mechanical actuator spring stiffness with a controller using position setpoints
knee that is used for most normal walking conditions, but also can reduce the motor electrical energy consumption as long

Frontiers in Neurorobotics | www.frontiersin.org 2 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 1 | The CYBERLEGs Beta-Prosthesis. Left is a CAD model with all of the relevant components labeled. The front of the knee shows the components of the
knee drive including the carriage and series elastic springs. The WA and ET mechanisms can be found at the back of the knee. Right is the realized prosthesis with the
electronics and Energy Transfer module connected in the locked position.

as the output kinematics, and therefore the joint torques, are 2. MATERIALS AND METHODS
near normal. In actual use, people do not find a way to use
the device in a way that provides these natural kinematics, and The Beta-Prosthesis is a transfemoral prosthesis that contains an
therefore the position targets must be changed allowing people active drive in both the knee and the ankle that are both capable
to walk, but reducing the efficacy of the series elastic actuators of net power output on each of the joints in the sagittal plane. The
due to the compensation for deviations in normal torque/angle design began as a passive knee/active ankle system in the Alpha-
characteristics. Prosthesis (Flynn et al., 2015) and had many new concepts added,
This paper defines each of the different systems that are particularly an entirely new knee system that allowed for net
contained within the CYBERLEGs prosthesis, first describing the positive work actuation at torques higher than normal walking
design rationale, desired behavior, and solutions (section 2). The as well as keeping the passive elements that were demonstrated
experiments run on the bench and in subject trials are described to work in the Alpha-Prosthesis.
in section 3. Results of the behavior of the prosthesis during
bench (section 4.2) and amputee validation testing (section 4.3) 2.1. Development of the Beta-Prosthesis
are then presented. We then discuss the results (section 5) and The CYBERLEGs Prosthesis was created as a part of the
conclude with future work planned for the prosthesis system CYBERLEGs FP7-ICT Project, which combines a prosthesis
(section 6). system to replace a lost limb in parallel with an exoskeleton to

Frontiers in Neurorobotics | www.frontiersin.org 3 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

assist the sound leg (Giovacchini et al., 2015), and sensory array
to control both systems (Goršic et al., 2014). The end goal of the
CYBERLEGs system was to assist those who have both a loss of
a limb and weakness in the remaining limb to regain walking
function and improve walking behavior. Integration within this
complete system had influence on the design of the device,
particularly in control and electronics architecture.
The CYBERLEGs Beta-Prosthesis consists of four major
systems (Figure 1) that when combined can reproduce the knee
and ankle torque and kinematics for the knee and ankle for
normal walking as determined by biological data. The first system
is a powered ankle based on a MACCEPA architecture which
gives the ankle a very low stiffness around the neutral position FIGURE 2 | Beta-Prosthesis ankle actuator schematic. Configuration of the
and quickly stiffens as the ankle is displaced from the neutral selected MACCEPA using rigid linkages. Note the Beta-Prosthesis includes a
position. The ankle is driven by a 200 W motor capable of high parallel spring system with a predetermined rest position as well as a manual
screw to change the MACCEPA pretension (P).
net power output. The ankle has an added parallel spring to
change the passive stiffness of the ankle, to assist the drive and
reduce the peak torque required of it. The second system is the
Weight Acceptance (WA) system. This is a simple spring that fully described in section 2.1.2.3.
is inserted at the knee during early stance phase to provide the
natural characteristics of stance flex without powered actuation. TA = TMACCEPA (α, P) + TParallel (θA ) + TET (θA , θK , L) (1)
(
The WA system is capable of producing large reaction torques TET if L = Locked
as external torques are applied, removing the need for the main where TET =
0 if L = Unlocked
actuator to operate during early stance and greatly reducing
energy consumption. The third system is the Knee Drive Baseline 2.1.1.1. Ankle actuator
Actuator (KD). This actuator is the main positive energy source The ankle of the device is a Mechanically Adjustable Compliance
of the knee joint, but under nominal use is primarily used to and Controllable Equilibrium Position Actuator (MACCEPA)
hold the Baseline Spring (BL) in place. During the gait cycle it is series elastic architecture (Van Ham et al., 2007; Jimenez-Fabian
possible to fully drive the knee using this system, and it provides et al., 2017) with a parallel spring to reduce required peak torques
all of the power for sit-to-stand and stair climbing operations. and can allow for smaller motor size. The ankle actuator is
The fourth system is the ET system. This system provides the composed of a main motive actuator, a series elastic linkage, and
late stance extension torque of the knee as the knee flexes, a fixed parallel spring, as seen in Figure 2. The actuator torque is
delivering this negative work from the knee joint as positive created by relative displacement of the moment arm ac ¯ around
work at the ankle to reduce the ankle torque, known as the the ankle axis a from the axis ab, ¯ a displacement called α. This
energy transfer period. This system does not provide net output displacement is caused by a motor that is mounted in the shank of
energy, but rather uses a binary locked/unlocked condition to the ankle, which in this schematic is represented by the immobile
physically connect the knee and ankle. The combination of these link ag¯ to the left. When the moment arm is aligned with the axis
four systems provides energy efficient and natural gait kinematics ¯ the actuator is in its neutral position and there is no actuator
ab,
through the level ground gait cycle with minimal actuation, joint torque. In this configuration, the actuator has low stiffness,
yet provides opportunity to modify the behavior delivering or but as the output is deflected, the natural stiffness quickly rises,
removing external energy during the gait cycle and provide much like in a normal ankle. This behavior is fully outlined in
different characteristics while attempting unlevel surfaces, sit-to- Flynn et al. (2015) and Jimenez-Fabian et al. (2015). Notably in
stand, and stair climbing. The prosthesis was developed using the Beta-Prosthesis the main MACCEPA spring pretension (P) is
torque and kinematics targets from Winter (2009), using these not motor controlled but is simply a manual screw mechanism.
data to gauge the behavior and requirements of the prosthesis.
2.1.1.2. Ankle parallel spring
2.1.1. Ankle A parallel spring system was added to the ankle to reduce
The ankle can be fully represented by the schematic shown the energy consumption by reducing the necessary holding
in Figure 2. The total ankle joint (TA , Equation 1) is the torque required by the motor and increase the velocity of
torque around joint a and is the summation of three torques, ankle actuation, as shown in Figure 2. Here the parallel spring
the first from the MACCEPA actuator (TMACCEPA ), which is engagement depends only on the ankle angle θ , which can be
dependent on the relative ankle moment arm displacement α changed by changing the rest position of the parallel spring with
and the actuator pretension (P), the second from the parallel shims. This has been done in previous designs, most notably
spring (TParallel ), which is only dependent on the ankle output the powered prosthetic ankles from Au and Herr (2008) and
displacement θA , and the third from the Energy Transfer system Vanderbilt (Lawson et al., 2014).
(TET ), which is dependent on the angles of the knee, θK , and ankle An example of how the torque output of the actuator is
and the locking condition L (if unlocked TET is equal to zero), affected by two different parallel spring configurations is shown

Frontiers in Neurorobotics | www.frontiersin.org 4 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 3 | Two examples of adding a parallel spring to modify the Torque/Angle Characteristics of the ankle. By subtracting the torque from the parallel spring (red)
from the required ankle torque (black), the required motor torque is determined. In the left example, which was chosen to minimize the peak torque using two linear
springs, the peak torque of the actuator is reduced from 130 to 50 Nm. In the right example, the spring was chosen to assist as much as possible without the motor
needing to work against the parallel spring during the gait cycle. The peak torque is reduced to 80 Nm.

in Figure 3. In the left of the Figure, a two stage spring, which can the center of the ankle, allowing the motor to be housed within
be seen as a change in stiffness at –7◦ , was chosen to minimize the structure of the shank. The knee system clamps onto this
the maximum torque while remaining easy to implement with shaft allowing adjustment of the distance and transverse rotation
a nested two spring system. The right of the Figure implements between the knee and ankle axes.
a single spring to simply capture the vault over energy of the
stance phase, while avoiding loading during the other parts of 2.1.2. Knee Architecture
the gait cycle. This second configuration is a bit more realistic The knee is comprised of three major systems that, when used in
to use, as the motor should never need to load the spring during combination, can approximate the total knee torque of normal
normal walking, but the reduction of the peak torque is smaller. walking with low electrical cost. These systems are the KD, the
In addition it only uses one spring, and for tasks where the ankle WA, and the ET system. The roles of each of these systems are
is passive, such as sitting in a chair, the parallel spring doesn’t outlined in Figure 4. The main KD system consists of a tuned
hinder motion as much. Adding this passive element to the SEA which provides the nominal torque required for normal
ankle joint does not change the net amount of output work the walking without needing to actuate. When the drive is held
motor should provide; the integrals of the absolute values of the at its nominal neutral position (zero torque at 60 degree knee
curves for normal walking with and without the parallel spring flexion), this drive provides the baseline torque shown in blue in
remain the same. However, the required peak torque, which is Figure 4. The second system provides a stance flex torque during
directly related to the current of the motor, in the left example in the weight acceptance phase to reduce collisional costs associated
Figure 3 is greatly reduced from 120 Nm to about 50 Nm for a with heel strike, shown in green in the Figure. The third provides
healthy person of 80 kg, and the right example the peak torque torque during the flexion phase of the gait cycle through delivery
is reduced to about 80 Nm. This reduces the holding torque the of negative work from the knee joint as positive work at the ankle,
motor needs to provide, which is energy lost without providing known as the energy transfer period. The physical relationships
any output work, but does have an effect on the required power of these systems can be found in Figure 5 and a general weight
output profile. Overall it allows a reduction in gear ratio of the and dimension table can be found in section 4.1. This schematic
drive leading to increased actuator velocities and reduction in the shows the knee motor (MK ), the knee Baseline Spring (KBL )
electrical consumption of the system. and Extension Spring (KEX ), and the Weight Acceptance section,
which contains the Weight Acceptance motor (MWA ) and spring
(KWA ). The knee joint torque is shown as τK and the force
2.1.1.3. Ankle realization transmitted to the ankle through the energy transfer mechanism
The left side of Figure 1 shows a CAD model of the Beta- is represented by FET .
Prosthesis where important features are labeled and can be
compared to Figure 2. The parallel spring system can be found 2.1.2.1. Knee drive (KD) actuator
in the heel of the device which provides approximately 4 Nm/deg The front of the knee houses the KD actuator, as in Figure 1.
plantarflexion torque. In this design the motor has been placed in This actuator consists of a small 50 W motor (Maxon ECi-40)

Frontiers in Neurorobotics | www.frontiersin.org 5 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

connected through a 5.8:1 gearbox to a 2 mm lead ball screw in 0.5s. This drive is connected to a carriage that houses the series
drive. This actuator can run at a linear velocity of 80 mm/s using elastic springs, similar in function to the designs of Pratt et al.
a 36 V supply, meaning running from full flexion to full extension (2002). The springs in turn actuate on a push/pull rod which
drives the knee joint. The knee joint is connected to a standard
socket pyramid for interfacing to the subject.
There are two series elastic springs held within the carriage.
The Baseline Spring (BL) provides the flexion torque of the knee
that is shown in Blue in the Figure when the knee carriage is
held at a constant position, corresponding the neutral position
at approximately 60 deg. The torque created by this spring,
approximately 0.3 Nm/deg flexion, can be modified while under
load during all phases of the gait cycle. It is of note that this
is much softer than the estimated physiological stiffness seen in
Pfeifer et al. (2014) which ranges from 5 to 17 Nm/deg. It is
also important to note that energy from the knee motor can be
used directly or stored in the knee SEA, even when the WA
mechanism is engaged. Opposite to the BL spring is the Knee
Extension spring (EX) which provides compliant actuation when
stair walking or going from sit to stand. Because the extension
moment is theoretically not used during normal walking and only
used during high power, non-repetitive motions such as sit to
stand, a shorter and stiffer (approximately 6 Nm/deg extension)
spring is used, which is better suited to these tasks, used to
FIGURE 4 | Torque/Angle Characteristics of a 80 kg individual showing the
insulate against shocks, and provide higher forces before full
behavior of the Baseline Spring (blue) and the torque during WA (green). The
gait cycle begins and ends at the heel strike, progressing to the Weight compression.
Acceptance phase where the WA system provides the majority of the torque. The right half of Figure 5 shows the kinematic relationships
After the WA phase, the ET system provides the necessary extension torque to for determining knee torque around the knee joint a. Link Ak
keep the knee from collapsing during pushoff by pulling on the ankle. After full is directly tied to the thigh while Bk is the red pushrod in
flexion, the ET system is disengaged and the BL spring provides flexion torque ¯ is anchored to the prosthesis, and the carriage,
Figure 1. Link bc
to arrest the end of swing phase and is adjusted by the KD system. In this
example there is no pretension on the carriage and the carriage is placed in which has a length D, is allowed to slide along the shaft. The
the nominal position (zero torque at 60 degree knee flexion). equations governing the knee torque due to the main knee drive

FIGURE 5 | The knee architecture schematic. The left side of the diagram shows the main knee carriage, as well as the baseline (BL) and extension (EX) springs. The
BL spring provides the breaking torque during knee extension during normal walking while the EX spring provides the torque during high power extension operations.
The carriage moves the rest position of the two springs. This figure also shows the relationship of the WA with the main knee drive. The right figure shows the
kinematic definitions used in determining knee actuator torque, as defined in section Knee Actuator Kinematics (see Supplementary Material).

Frontiers in Neurorobotics | www.frontiersin.org 6 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 6 | The WA system (Left) and a schematic of the system as it is in the prosthesis. The screw drives the spring up and down so the knee interacts with it at a
desired angle. The small motor only needs to overcome the friction in the gear drive and nut to move the spring. Initialization is handled by a small optical switch. The
WA system schematic (Right) shows the relevant relationships needed to calculate the resulting torque from the WA system. The governing equations are presented
in section Weight Acceptance Kinematics (see Supplementary Material).

can be found in Heins et al. (2018) and are reproduced in trapazoidal screw with a 3 mm lead, the spring is locked in place
Supplementary Material. while under load without the need for motor input. The stiffness
of the spring is chosen, when combined with KBL , to provide
2.1.2.2. Weight acceptance (WA) mechanism the torque characteristic shown in Figure 4, represented by the
Directly after heel strike is the weight acceptance stage of the gait green line. The WA provides a relatively consistent joint stiffness
cycle. This is characterized by a spring-like loading and unloading during the stance phase, which corresponds to biomechanical
period of the knee joint, as seen in Figure 4 in green. Creating this data (Shamaei et al., 2013). Positioning of the WA spring requires
torque-angle characteristic using the main motor is electrically a small motor to overcome the friction and inertia of the spring
costly, and would require the actuator to unload and reload the system. The small motor allows the WA to actuate with a linear
EX spring to create this torque. Also the loading and unloading velocity of 37 mm/s when using a 36 V supply, meaning the
characteristics are well suited to be replaced by a static spring. device can move from fully unlocked to fully locked in about 0.5s.
The problem with this is the spring needs to be inserted during The WA system has been designed with a unilateral constraint so
the stance flexion stage and removed for the swing phase so it that during swing phase the leg swing can lead the position of
does not interfere with the swing flexion, requiring some sort of the spring. This means if the motor is too slow to track the swing
locking mechanism. phase, the passive swing extension is not hindered by the WA
The WA requires a lock that is capable of providing high motor.
forces when compressed, but requires little energy to position This mechanism is attached behind the knee by a moment arm
the spring under low to zero load. The WA does not need to of length Ck , as seen in the right schematic of Figure 6. In this
actively provide joint work above the passive storage and recoil schematic the spring is positioned by changing ZWA by driving
of the spring. Also the WA should not unlock when there is a the screw. Y is the distance between points a, the knee joint
high force on the spring, because if the spring is loaded, it usually center, and b, the point the WA is anchored to the prosthesis.
means the user of the prosthesis is loading the knee for stability. The slider at point c is allowed to compress the spring (KWA ) as
One way to satisfy these requirements including infinite locking the knee angle, θk , changes. This changes the effective length XWA
positions, high locking torque, and low continuous locking and creates the knee torque. Note that if ZWA is small, the knee
power is a friction based, non-backdrivable gearing (Plooij et al., will never compress the spring at any angle θk . The equations
2015). governing this system can be found in Supplementary Material.
A non-backdrivable screw was chosen as the main drive
mechanism, where the rest position of the spring is modified by a 2.1.2.3. Energy transfer mechanism
small 30 W motor (Maxon EC16) through a gear drive so that During normal walking the knee primarily acts as a dissipative
as the knee flexes it compresses the spring at a desired angle, system, which is one reason why active damping systems can
as shown in Figure 6. Because the screw is a non-backdrivable provide relatively good behavior at the knee joint. This necessity

Frontiers in Neurorobotics | www.frontiersin.org 7 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

moment arm E. Pulley e is used as a locking mechanism which is

accomplished through the use of a non-backdrivable trapezoidal
screw of 3 mm lead and 30 W (Maxon EC16) motor, that was
placed on the back of the shank. The screw was used to position
pulley e, increasing and decreasing the length the ET cable needed
to be routed. A stiff spring, KET was placed in the cable so the
cable force would be limited by the spring displacement. The
routing provides a ratio of approximately 2:1 for knee motion
to ankle motion. Because the joints are directly connected in
tension, a flexion torque of the knee creates a plantarflexion
torque of the ankle, but an extension of the knee does not cause a
dorsiflexion torque of the ankle. These relationships are governed
by the equations in section Energy Transfer Kinematics.

2.1.3. Electrical System

The Beta-Prosthesis was designed to integrate with the
CYBERLEGs control and orthosis module systems and therefore
does not have onboard control systems or power. The
CYBERLEGs control and power system is a completely
autonomous system consisting of a wearable backpack housing
the control unit and batteries for power (Giovacchini et al.,
2015). The control unit is a National Instruments sbRIO-9632
embedded single board computer with an integrated FPGA
that is intended to run the prosthesis, the wearable sensory
apparatus (Ambrozic et al., 2014), a pelvis orthosis, and high level
control algorithms (Ronsse et al., 2013; Ruiz Garate et al., 2017)
developed within the consortium.
FIGURE 7 | The ET system schematic showing the direct connection of the The sbRIO computer is connected to the prosthesis through a
ankle to the knee joint. Here the knee joint a is connected through the moment tether cable that carries all of the control signals from the purpose
arm and push rod to the center of the carriage at g, constrained to slide on
¯ A spring KET is connected to the prosthesis at pulley d and connected
axis bc.
made driver board mounted on the prosthesis. This driver board
to a cable that wraps around a pulley at g and back to pulleys d and e. The uses four Maxon 50/5 module driver boards, one for each of
cable then attaches to the ankle at moment arm E which when pulled creates the motors in the prosthesis. The tether also transmits the 36 V
a torque around the ankle joint f. The positions of pulleys d and e are defined supply voltage to be compatible with the other components of the
by p1 and p2 , of which P2y can be adjusted through actuation. The distance
CYBERLEGs system.
between the knee and ankle joints is Lsh . The governing equations can be
found in the Energy Transfer Kinematics (see Supplementary Material).
2.2. Bench Testing Setup
The knee and ankle systems were tested on a custom designed
of removing energy from the joint provides the opportunity to test bench to verify that the system could produce the desired
capture this energy to return to the system at another point of torques and the passive behavior of the knee creates a good
time in the gait cycle. This has been previously done in systems approximation of the desired knee torque, as seen in Figure 8.
such as the passive WalkMECH (Unal et al., 2013), which uses a The test bench allows a motor to drive the output of the prosthesis
spring, push/pullrod, and locking system to deliver energy to the while controlling the relevant joint characteristics, either position
ankle. Systems such as the active MIT CSEA knee simply push of the moment arm or position of the carriage depending on the
regenerated electrical energy from the backdriven drive motor joint to be tested. The load motor used on the test bench was
onto the power bus during times of negative work (Rouse et al., a Maxon RE 50 with a GP 62 A gearbox driven by a Maxon
2014). EPOS2 70/10 motor controller. This motor was connected to the
The ET system of the Beta-Prosthesis intends to capture two output of the joint to be tested using a rotary torque transducer
major sources of negative work of the gait cycle, work done to (ETH Messtechnik DRBK-200-N). The signals for driving the
stop the lower leg swinging forward at the end of swing phase motor and recording the signals were recorded at 100 Hz, with
and negative work done at the knee during late stance that is 16 bit accuracy through the sbRIO to guarantee synchronous data
required to keep the knee from collapsing under the individual collection.
while beginning flexion for the swing phase. To do this, there is
a cable that directly connects the ankle to the knee when it is 2.3. Prosthesis Control System for Walking
locked, and allows the cable to go slack when it is unlocked. A Experiments
schematic of this system can be found in Figure 7, showing how 2.3.1. Low Level Control
the ET cable starts anchored to the prosthesis at pulley d, wraps Low level control of the system of the prosthesis is handled by the
around the end of the knee pushrod at pulley g, then continues ESCON drivers, running a tuned current and velocity feedback
back to pulley d and e to attach to the back of the foot through loop. The velocity signals used by the ESCON are generated by

Frontiers in Neurorobotics | www.frontiersin.org 8 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

the FPGA of the sbRIO running a control loop at 1,000 Hz. discussed in Ambrozic et al. (2014) and Parri et al. (2017).
Position commands are calculated by the real time system at 100 Here the state machine has been modified to include the
Hz and fed to the FPGA. Position commands that are tracked by new WA mechanism positions as well as the KD command
the real time system are calculated by the Top Level Control. positions. The state machine of the WSA contained a number
of levels, the first including a quiet standing, gait initiation,
2.3.2. Top Level Control and gait termination phases. From the gait initiation phase
Control methods for the Beta-Prosthesis utilized a modified the main walking state machine was entered. The walking
Intention Detection (ID) system and Wearable Sensory state was broken into four different sub-states named, from
Apparatus (WSA) controller with a finite state machine as the point of view of the prosthesis, the Early Stance (State
1), Late Stance (State 2), Swing (State 3), and Late Swing
(State 4) phases of the gait cycle as seen in Figure 9. These
states were triggered by a combination of the WSA angular
velocity sensors as well as signals from pressure insoles. Each
of these states was designed to capture a specific gait event,
heel strike, heel off, toe off, or terminal swing phase. The state
machine system along with state transitions is summarized in
Figure 10.
It should also be stated that these position setpoints (locked,
unlocked, half flex, etc.) were tuned for the individual’s
self selected speed and were used only for level ground
walking.

2.4. Subjects
The male subjects (n = 4) had an age range of 63 (SD = 11) years,
a weight range of 61.8 (SD = 2.63) kg, and a height range of
FIGURE 8 | The prosthesis was clamped in a rigid frame with a motor 173.75 (SD = 5.06) cm. Three had amputations due to traumatic
connected to the output of the joint to be tested. This figure shows the injury and one was a dysvascular patient with an average of
prosthesis as it would be positioned in knee actuator or ET system tests. The 11 (SD = 10.67) years since their amputation. Their mobility
frame has a large DC motor attached to the prosthesis through a torque levels varied from were K1 (the dysvascular subject), to K3 (two
sensor, so that the output kinematics of the joint can be varied at the same
time as the prosthesis is actuated.
subjects), as defined by the Medicare Functional Classification
Level.

FIGURE 9 | Normal level ground gait cycle (adapted from Cuccurullo, 2004). The figure shows how the different states correspond to the normal gait cycle, with the
shaded leg representing the prosthesis. Also displayed are frames from a video of an experiment showing the gait cycle of the amputee. Note that these transition
points are not fixed with respect to the gait cycle percentage, but rather to certain measurements made by the WSA (see Figure 10) and therefore can change.
Written informed consent was obtained from the subject for the publication of this image.

Frontiers in Neurorobotics | www.frontiersin.org 9 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 10 | The CYBERLEGs Walking State Machine. The CYBERLEGs WSA was used to create the triggers for state transitions, using the angular velocity,
ω(rad/sec), of the different limb sectors. The pressure insoles were used to determine if the feet were on the ground. Top level systems are shown in circles, and the
positions of each of the knee, ankle, and WA are shown. The exact values for each of these setpoints was determined by empirical trials.

3. EXPERIMENTS compared to estimated maximum torque. Results can be found

in Figure 11 (Right).
3.1. Ankle Bench Testing
The ankle was placed in the test bench described in section 3.3. Knee With ET Bench Testing
2.2 with the ankle output connected to the torque transducer. The prosthesis was placed in the test bench setup with the
The output motor was used to drive the ankle angle along the output motor of the test bench connected to the knee joint.
Winter trajectory at 1.5 s/stride while the ankle moment arm was The ET system was then commanded to lock and unlock from
commanded to provide the ankle torque for the given time in approximately 43% to 64% of the gait cycle, corresponding to the
the gait cycle. Although the ankle is capable of providing 130 time of negative work of the knee joint. The ET cable tension
Nm of torque during the gait cycle, the test bench motor was was measured with a load cell while the knee and the ankle were
only capable of driving the output to 90 Nm due to limitations commanded to track respective position trajectories. The ankle
in the current the external motor driver was able to provide. motor was commanded in output position mode, pinning the
Therefore a Winter torque/angle trajectory was selected with a kinematics of the ankle to the Winter trajectory regardless of the
peak torque of around 90 Nm. Results can be found in Figure 11 torque required.
(Left).
3.4. Preliminary Walking Experiments
As part of the preliminary validation of the prosthesis system,
3.2. Knee Actuator Bench Testing four amputees were subject to a 3 min treadmill experiment. Each
In this experiment the knee was locked at a fixed angle and subject attached the prosthesis to their existing socket through a
connected to the test bench torque transducer. The carriage was standard prosthesis pyramid adapter and a visual alignment was
commanded to deliver the maximum joint torque. This was performed by a prosthesis technician. After a short session of
repeated three times for both flexion and extension every 10◦ approximately 15 min for tuning of the prosthesis state machine
of knee flexion, from full extension to full flexion. The actual parameters from section 2.3.2 over a level ground 10 meter long
joint torque was recorded and maximum measured torque was catwalk, the subject was asked to walk at a self selected pace

Frontiers in Neurorobotics | www.frontiersin.org 10 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 11 | Ankle actuator torque using moment arm control (Left). The moment arm tracks a position trajectory to create the output torque (red) creating the
Winter Target output torque (gray). The test was run to a peak of only 90 Nm because of limitations in the bench test motor. The maximum flexion (blue) and extension
(red) torques for the knee actuator are shown in the (Right) figure. This does not include the torques created by the WA spring mechanism.

on the treadmill for 3 min to investigate the behavior of the 4.2.2. Knee Actuator Torque Testing
knee joint. This self selected speed was determined by slowly Figure 11 (Right) shows the estimated maximum torque (solid
increasing the speed of the treadmill in 0.2 km/h increments until blue and red lines) and experimental values (dotted blue and
the subject felt it was too fast to sustain. The speed was then red lines) achieved by the actuator and compared to the normal
reduced until the subject was comfortable. Subjects were asked torques during walking (black).
to minimize the use of the handrails, although to not remove
their hands from the rails in case of an emergency. Subjects 4.2.3. Knee With Energy Transfer
for the experiments were selected and approved by the Ethical Figure 12 shows the results of the bench tests of the prosthesis
Committee of the Don Gnocchi Foundation (FDG), Florence while using the energy transfer system. The ET system was able
Italy, and experiments were conducted under FDG supervision. to transfer approximately 8.2 J/stride from the knee to the ankle
Data was recorded from the prosthesis during the treadmill during this trial.
testing period.
During the validation with subjects, the ET system was not 4.3. Preliminary Walking Experiments
used for testing. This was because the control of the ET system A representative example of a subject walking on the treadmill
would not allow for kinematics that varied greatly from the for 126 strides at a speed of 2.2 km/h and average stride duration
desired Winter kinematics, leading to high forces as well as of 1.76 s/stride (SD = 0.07) are presented. The torque angle
ineffective gait. This was mitigated though the use of a modified relationships for the ankle (left) and knee (right) are found in
KD trajectory, allowing the subjects to walk. This is discussed in Figure 13 and shown compared to the typical behavior of a
greater detail in section 5.2.1. microprocessor controlled C-Leg (Segal et al., 2006). The average
energy injection from the ankle was about 2.8 J/stride.
Figure 14 shows the average behavior of the actuators and
4. RESULTS joints during a walking trial on the treadmill as well as the
timing of the state machine transitions during the gait cycle. Also
4.1. Design Results
shown here are the two absolute time based transitions 1t1 and
A list of prosthesis design values and prosthesis characteristics
1t2 which were experimentally determined delays between the
are found in Table 1. These values were used in simulation and
beginning of State 2 and the unlocking of the WA and between
during the testing session.
State 4 and the locking of the WA. The state transitions are
determined by the WSA.
4.2. Bench Testing Figure 14B shows the ankle moment arm desired position
4.2.1. Ankle Torque/Angle Testing and actual position. The ankle was only commanded to –10◦ of
The results of the ankle torque/angle characteristics from the plantarflexion, as opposed to the calculated –20◦ for full plantar
bench testing device can be found in Figure 11. The actuator torque. This was as requested by the amputees.
provides a torque output that is well with the standard deviation Figure 14C shows the desired and actual knee carriage
of the Winter ankle torque/angle behavior. position. Because the knee does not have the energy transfer

Frontiers in Neurorobotics | www.frontiersin.org 11 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

TABLE 1 | Selected prosthesis characteristics used in simulation and for final position to help the amputee with ground clearance during the
design. initial swing phase. The knee joint flexes to around 50 degrees
Property Value Units
and the knee carriage position is moved back to extended at the
end of the swing phase. The knee WA locks at the end of swing
Active DOF 2 knee, ankle to guarantee a safe heel contact.
System Voltage 36 VDC Figure 14E shows the state machine state during the gait
Ankle Moment Arm Length (B) 50 mm cycle. The state machine state was determined by the WSA, and
Ankle Linkage Length (A) 50 mm therefore did not happen at exactly the same time during the gait
Ankle Actuator Spring Constant (k) 130 N/mm cycle on every step, but because the gait was rather consistent,
Ankle Actuator Spring Weight 113.4 g the transitions happen at approximately the same time. This is
Ankle Parallel Spring Constant 94.20 N/mm seen as the sloping averages in the graph. The states roughly
Ankle Parallel Spring Weight 60 g correspond to early stance (Value 1), late stance (Value 2), early
Ankle Gear Ratio 320:1 - swing (Value 3), and late swing phases (Value 0) of the prosthesis.
Ankle Winding Voltage 48 V
Ankle Max Torque 130 (@15A) Nm 4.4. Prosthesis Energy Consumption
Ankle Continuous Torque (no parallel 30.6 Nm During Trials
spring)
From previous power consumption measurements, the total
Ankle Range of Motion –30 to 20 deg device consumes about 65 J/stride. Of this, the WA requires about
Shoe Size 42 EU 10 J/stride regardless of the speed of the gait. Electrical model
Knee Gear Ratio 5.8:1 - estimations show that during this trial the ankle uses about 19
Knee Ball Screw Ratio 2 mm/turn J/stride while the knee uses about 23 J. These values seem to be
Knee Max Torque ∼ 70 Nm considerably less than older versions of the Vanderbilt prosthesis
Knee Continuous Torque ∼ 55 Nm (91 J/stride), although this was at 5.1 km/h and 87 steps/min with
Knee Range of Motion 0 to 95 deg a higher joint work output (Sup et al., 2009) and was considerably
Baseline Spring Constant (KBL ) 10.7 N/mm (each spring) higher than the CSEA knee (3.6 J/stride) which had a net negative
Baseline Spring Length 64 mm joint work and was able to partially power its own electronics
Baseline Spring Diameter 16 mm through regeneration (Rouse et al., 2014). These values should be
Baseline Spring Mass 19.7 g taken only as general guidelines due to the significantly different
Extension Spring Constant (KEX ) 89.1 N/mm (each spring) testing conditions and subject behaviors in each study.
Extension Spring Length 32 mm
Extension Spring Diameter 16 mm
Extension Spring Mass 15.6 g
5. DISCUSSION
Weight Acceptance Spring Constant 300 N/mm
Testing of the device was divided into two different sections,
Weight Acceptance Spring Rest Length 38 mm
bench testing and the patient trials. Bench testing of the device
Weight Acceptance Spring Diameter 32 mm
shows the device can work as designed when the external
Weight Acceptance Spring Weight 90 g
kinematics are imposed, with good ankle and knee torque
Prosthesis Overall Mass ∼5 kg
approximation of the biomechanical data at speeds close to
- Knee Module Mass 1,736 g
the actual speeds of the tests. Patient trials showed that all
- WA Module Mass 440 g four patients were able to walk with the prosthesis using the
- Ankle Module Mass 1,850 g CYBERLEGs WSA over level ground and on the treadmill with
- Electronics 300 g minimal (<1 h) training. Overall these tests show it is possible to
Prosthesis Height (Overall, longest) 500 mm find walking gaits that allow for ambulation with actuators that
Knee to Ankle Length (Shortest) 355 mm utilize soft series elastic springs, although the overall behavior
Knee to Ankle Length (Longest) 395 mm was not the same as averaged normal gait as defined by the
Knee to Pyramid Top Length 30 mm Winter biological data. An example of the typical gait can be
seen in the Video in Supplementary Material which shows that
qualitatively smooth gait progression and a small energy injection
at the ankle was possible.
mechanism the knee actuator is used to provide flexion and From the patient experiments, it was apparent that ET design
extension torques when necessary. Even though the ET system and control was insufficient for gaits greatly differing from the
was not included in the initial trials, the KD system of the average kinematics, leading to the loss of ET function and in
prosthesis was able to actively compensate for this missing some cases interfering with the gait cycle. Due to this loss of
component, at the expense of a less electrically efficient gait cycle. ET function, the knee actuator control was highly modified
Figure 14D shows the WA motor position. At the heel strike compared to the original design. This resulted in gait that
beginning of the gait, the WA is in the locked position. During was closer in behavior to a passive knee than the average
40–80% of the gait cycle, the knee motor setpoint is set to a flexed biomechanical data. This was exacerbated by short training

Frontiers in Neurorobotics | www.frontiersin.org 12 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 12 | Bench test results of the prosthesis with ET system. (A) The knee angle of the prosthesis (red) while using the ET in the current configuration. Note that
while the ET is locked, the knee angle deviates slightly from normal Winter angles (blue). (B) The time dependent behavior of the torque shows the knee torque well
within the standard deviation of the Winter data for almost all of the gait cycle. Using just the passive systems of the prosthesis can provide a good approximation of
normal knee torque. (C) The knee power (red) shows the shift in the power peak to better align with the peak in the ankle power (dashed black). By bending the knee
earlier, the power of the knee is better suited to transfer to the ankle during the maximum pushoff. (D) Shows the torque of the ankle provided only by the ET system.
This torque would replace ankle motor torque, reducing electrical consumption. Right Top graph: Passive behavior of the knee actuator including ET. Using only the
BL (blue), WA (green), and ET (red) provides the knee with a good approximation of the normal Winter knee torque. The knee actuator carriage is kept constant. The
green line is the output torque of the combination of the BL and WA while the WA is locked, the red line is the combination of the BL and ET. Right Bottom: The force
in ET cable during the time the ET is engaged. The force in the cable peaks at around 700 N, which is applied at the ankle, reducing the ankle motor torque.

FIGURE 13 | The experimental mean torque-angle characteristics of the ankle (left) and knee (right) during amputee experiments (Blue) at 2.2 km/h compared to the
Winter reference (4.8 km/h) (Red). All 126 strides of a representative subject are shown in gray traces, illustrating the variation between steps. Also shown are data
from a widely prescribed C-Leg microcontroller prosthesis (black) (Segal et al., 2006). The desired ankle pushoff torque was much lower than the normal walking at
the request of the subjects, in part due to the lower walking speed of the tests. Average energy injection from the ankle was about 2.8 J per stride. The knee behavior
of the powered prosthesis ended up providing a microcontroller-like torque angle characteristic, mainly because of the lack of stance flex.

Frontiers in Neurorobotics | www.frontiersin.org 13 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 14 | Knee angle, Knee Drive carriage position, and WA position during treadmill walking. The desired position of the WA and KD carriage are shown in red,
here the gait state machine determines when the position setpoint should be changed, the position of which was determined experimentally through tests over level
ground. The state changes do not happen at exactly the same time during the gait cycle on every step because of variances in the signals given to the WSA, resulting
in curved averaged state values. 1t1 and 1t2 are experimentally determined delays between the beginning of State 2 and the unlocking of the WA and between State
4 and the locking of the WA.

periods that were insufficient to allow the needed level of testing with the ET system also shows that it is possible to transfer
familiarity with a device that operates dramatically different from energy (at least 8.2 J/stride) from the knee to the ankle, at least
their currently prescribed prostheses. under specific kinematic and loading conditions.

5.1. Bench Testing 5.1.1. Ankle Bench Testing

Even though the tests were slightly limited in speed and The ankle has a bit of hysteresis primarily coming from friction
magnitude compared to the original torque and angle targets, in the parallel spring mechanism, which can be seen in Figure 11.
we have been able to show that the power consumption of the Overall the ankle was able to reliably reproduce the torque angle
knee actuator alone (i.e., without the ET system) does reduce the characteristics of the normal Winter data.
energy consumption of the motors compared to a direct drive
system (Geeroms et al., 2017, 2018). This is due to the division of 5.1.2. Knee Bench Testing
the torque angle characteristics of the knee into different sections, The right side of Figure 12 shows predicted and actual knee
utilizing the WA system for the high stiffness required after heel actuator torque measurements while bench testing the knee
strike and energy storage of the BL spring during stance and actuator. The estimated torque values come from a maximum
swing phases. Ankle behavior also was able to utilize the energy 2000 N axial force on the actuator carriage. To reach this high
capture and return of the ankle SEA and parallel springs. Bench torque the spring must be fully compressed for both the extension

Frontiers in Neurorobotics | www.frontiersin.org 14 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

and flexion torques. The low torque region of the flexion torque motor primitives methods that have been developed within the
is where the carriage does not have enough displacement to fully CYBERLEGs consortium (Garate et al., 2016; Ruiz Garate et al.,
compress the spring. The actual joint torques of the knee are well 2017).
estimated by the predictions. The ankle was able to inject energy into the gait cycle, but this
amount was small during these tests. The ankle used a very small
5.1.3. ET Bench Testing pushoff angle, at the request of the subjects. On average the ankle
Using the force in the cable and the moment arm of the ET on the performed about 2.8 J of work per step, which is far from the 17 J
ankle, the torque on the ankle can be computed. Multiplying this of the target walking, but better than the slightly negative total
with the displacement of the ankle gives the energy transferred work provided by passive devices.
from the knee to the ankle. In this example the value is 8.2 J, The question then arises if users settle into these ideal
which is about 80% of the available energy from the knee and torque and angle characteristics if the prosthesis is able to do
about half of the required energy for pushoff, a considerable so on the test bench. From the initial patient trials with the
savings if it can be replicated in walking individuals. prosthesis, it was clear that the subjects did not have a Winter-
The top graph in Figure 12 shows the torque-angle like average gait torque/angle progression, with joint torque and
characteristics of the full prosthesis knee including the KD, angle characteristics varying greatly from the original targets.
WA, and ET in the test bench. The contributions of each system
is illustrated by a different color in the graph. When the graph is 5.2.1. Issues With the ET System
only blue, the BL spring is the only component providing knee The largest deviation from the original design was the inability to
torque. When green, the WA and the BL systems combine to utilize the ET system. The ET system was designed with specific
create the knee torque. When red, the ET system and the BL required output kinematics to correctly harvest knee negative
combine to create the displayed torque. The bottom graph of work and deliver that work to the ankle. Deviations from these
the Figure shows the force in the cable of the ET system during kinematics caused a dramatic rise in the cable system past the
activation. This cable pulled on the ankle with a moment arm of ET design criteria. For example under normal conditions the
approximately 6 cm, resulting in an energy transfer of around ET could expect to see approximately 600 N of tension during
8.2 J. walking (Heins et al., 2018). If the kinematics were changed to
The knee trajectory was slightly modified to allow the best those of a standard C-Leg walker, these forces could increase
fit between knee and ankle torques. This also had the effect of to higher than 4000N. The deviant kinematics, in particular the
shifting the knee power a bit earlier which allowed the energy timing between knee and ankle motions, and increased forces
transfer to align better with the ankle pushoff. In normal walking, prohibited the ET system from unlocking at the correct times.
the negative work from the knee is slightly after the pushoff of As a result, the prosthesis ankle was not able to dorsiflex at the
the ankle, as shown in Figure 12C. Also if one wants to track the beginning of the swing phase, increasing the chance of stumbling
Winter ankle power exactly, if the ET transfers normally, there and resulting in an unsafe situation for the amputee.
is a requirement of the ankle motor to absorb the late energy of Removing the ET modified the behavior of the KD because
the ET system, as discussed in Heins et al. (2018). By shifting the original torque angle behavior of the prosthesis depended on
it earlier more energy can be transferred without needing to this system to provide extension torque during the late stance.
dissipate mistimed energy at the ankle. In normal walking the This modification of the KD required the actuator to move
net work output of the knee is around –14 J, while with our much further distance than originally intended to provide this
modified kinematics the output work was around –10 J, but the extension torque, which limited the peak velocity of the knee.
time shifting of the negative work allowed more energy transfer. It is possible that if proper feedback control for the ET
mechanism was created, then it might be possible to safely use
5.2. Walking Trials the system by limiting the tension in the cable, although because
Four subjects were able to walk on level ground after a short the ET was not designed to actuate under load, this was not
training session. Two of the four were able to increase their implemented. In addition if the ET is then a fully active system
self selected walking velocity by approximately 0.2 km/h over rather than a clutched system as designed, there must be a
their own prosthesis. Figure 14 shows the averaged results for controller that can watch the knee and ankle joints and predict
the experiment. Figure 14A shows the angle of the knee joint periods of time when the knee would dissipate power and the
during walking. As the amputees were not used to walking with ankle provide kinematics that were suitable to receive this energy,
this prosthesis and did not train for a considerable amount of if those periods exist. Another way of solving this issue is to
time, they did not use the stance flex as it was designed to be strictly control the output kinematics, which would guarantee the
used and preferred a straight leg during walking. Based on subject knee and ankle relationships. This method would not necessarily
feedback during the tuning session, the WA motor was given a result in a reduction of motor electrical consumption or a suitably
sufficiently high setpoint which effectively locked the knee joint stable gait. Modifications to the ET system must be made to
in extension. This effect is seen during the first half of the gait further examine this aspect of the design in walking trials.
cycle, as there is no knee flexion. Adapting this for different
terrain and speeds are a couple of focuses of future study for the 5.3. Control System Modifications
state machine controller. Other types of controllers can also be The control system of the prosthesis was designed to change the
implemented on the device and will be investigated, such as the configuration of the prosthesis in order for the output torque to

Frontiers in Neurorobotics | www.frontiersin.org 15 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

match some quasi-static torque target assuming that the output generally lowered as the gait becomes more symmetric. Currently
kinematics of the system would then converge to the normal only powered transtibial powered prostheses have been shown to
gait kinematics. This method makes a of assumptions, most reduce energy consumption of the user to normal levels during
importantly first that if the average joint torques are obtained, the level ground walking. These devices are capable of providing
person would naturally have kinematics which are about normal torque/angle characteristics much closer to normal ankle
and second that deviations away from normal joint torques behavior than conventional prostheses (Herr and Grabowski,
caused by external disturbances would be sufficiently handled 2012). But it seems that additional energy asymmetrically injected
by the natural impedance of the prosthesis and the control of into the gait cycle could reduce this further (Caputo and Collins,
the person using the prosthesis. The control method was not 2014), although at increasingly diminishing returns. This would
designed to be a classical impedance or torque based system using mean there should be torque/angle characteristics that are more
output feedback to generate a specific stiffness or trajectory. On metabolically efficient than the Winter targets for a given walking
the bench this works well because the output kinematics of the condition, even if assistance is asymmetrically applied. This is
system are constrained by the output motors, and therefore the doubly true when considering prosthesis design when the inertial
joint torques and kinematics are as expected. properties of the leg can be custom tailored. In simulation,
When both the torque and kinematics are unconstrained, the relaxing the symmetry constraint has shown that it should be
person tends to walk very differently than expected, particularly possible to reduce amputee cost of transport lower than walking
at the knee. While ankle kinematics and torques were somewhat with a biological leg (Handford and Srinivasan, 2016, 2018).
normal for the low input power that was desired, the knee Although symmetric gaits may not be optimal for energy
behavior was far different. To find gait cycles that were capable consumption, there is an increased chance of osteoarthritis in
of safely walking, the ET needed to be disabled. Because of the the sound leg in transtibial and transfemoral amputees, and there
lack of the ET system during the walking trials, the control are reasons to believe that increasing joint kinematic symmetry
of the system was modified so that the KA could replace the generally leads to reduced detrimental loading, particularly peak
functionality of the WA system. This included changing the state force and peak knee external adduction in the contralateral
machine to follow a different trajectory than that with the ET limb (Morgenroth et al., 2011; Grabowski and D’Andrea, 2013).
enabled, ultimately implementing a very simple flexion/extension Whether the reduction of these forces actually reduces incidents
motion for the knee, as was determined from feedback from the of osteoarthritis has not been proven, and also it hasn’t
subjects. For each of the 4 states of the gait cycle, the subjects were been proven that restoring torque/angle characteristics of the
asked what behavior they would like to have from the prosthesis amputated limb to normal will minimize these forces in a global
and the position threshold for the state was determined. sense.
Even though natural kinematics and torques do not
5.3.1. Normal Torque/Angle Characteristics as Target necessarily minimize metabolic energy consumption or minimize
The Beta-Prosthesis was designed with the intention of providing injury, one thing that is certain is that the more normal and
the normal torque and kinematics of a leg, as determined by symmetric the gait kinematics the more natural and unassuming
average healthy gait. With this particular design and controller it looks, which is a large part of the functionality of a daily
there is an implicit assumption that the prosthesis has a similar worn prosthesis. It also provides a familiar starting target for
mass and moment of inertia as an average human leg, because the design of prosthetic limbs which are designed to replace
the target torques are dependent on these aspects. This prosthesis normal limbs. Possibly designs based on non-anthropomorphic
has been built with this as a constraint, but tends to lead to principles will allow the discovery of other solutions in the future
a relatively heavy prosthesis and associated problems, such as (LaPrè and Sup, 2013), in much the same way carbon ESR blades
socket pistoning. From discussions with the subjects during the revolutionized prostheses for running.
trials, while walking while powered this extra weight is not Regardless, the current control of the prosthesis does
noticed until the prosthesis performs poorly or is not actuated, not attempt to force the Winter kinematics output at the
and then the weight is highly detrimental. knee and results in an asymmetric gait. This was shown
It should be noted that even without the normal biological to increase the metabolic rate of the participants (10 ±
torque/angle joint progression the patients were able to walk 9%) when compared to their conventional prostheses. As the
at speeds equal to or above those while using their every day subjects become more familiar with the device, the control
prosthesis. So how necessary is it that the prosthesis really becomes more refined, and we are able to better apply torques
track the normal gait characteristics? Indeed some extremely with more accurate timing, we expect an increase in gait
fast transfemoral amputee sprinters find that the design of their symmetry and to eventually reduce metabolic consumption
passive prostheses may not need a knee joint at all, relying on the (Malcolm et al., 2013).
prosthesis design to generate the pushoff and using the hips to
provide ground clearance. 5.3.2. Comparison of Kinetics and Kinematics to
Volumetric oxygen measurements with almost all current Normal and C-Leg
prostheses are generally 10–30 percent higher than normal Figure 15 shows the knee and ankle torque and kinematics of
walking, and 50–100 percent higher at maximal speeds of both the CYBERLEGs prosthesis and the C-Leg as a function of
walking (Genin et al., 2008). Many have suggested this is because stride percentage, as was done in Figure 13. In these Figures it is
of dealing with gait asymmetry, and energy consumption is a bit clearer to see how the experimental ankle torque essentially

Frontiers in Neurorobotics | www.frontiersin.org 16 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

FIGURE 15 | Comparison of the prosthesis behavior vs. the target Winter and C-Leg data (Segal et al., 2006). Kinematics are shown in the upper two graphs while
the bottom graphs show the joint torque. Note that because the measurement of joint torque was done using the actuator, the blue dotted line is only an estimate of
the knee joint torque based on the behavior of the prosthesis.

followed the biological torque up to the maximum that the There is no feedback of the output trajectory, output
subjects requested. It is also clear the ankle had an early pushoff, impedance, or output torque to compensate for deviation from
as well as a large dorsiflexion during the swing phase, both of the target torque/angle in this method. This is actually similar
which were requested by the subjects. The knee joint however has to a rest position microcontroller controlled system, where the
behavior much closer to the microcontroller knee, with the knee rest position of a spring is changed during different phases in
remaining on the full extension endstop during the stance phase. the gait cycle, although here the position can be changed while
Because the knee torque of the prosthesis is measured through loading and unloading. It was theorized that if the position of
the actuator displacement, an estimated blue dotted line has been the motor side of the spring was placed close enough to the
shown on the knee torque graph which better represents the correct position, the loading characteristics of the output could be
total knee torque during the stance phase. The major difference slightly modified by the walker and they would find the best way
between the CYBERLEGs prosthesis and the C-Leg is a powered to walk with the device, resulting in near normal kinematics and
extension phase at the end of swing phase instead of a braking joint torque. In this way, neither the kinematics or the torque are
flexion torque. The subjects felt best knowing the knee would be fully constrained. Results show that this tends to work well in the
at full extension at the end of swing phase, presumably because ankle, the users seem to be able to load and unload the ankle in a
it is difficult to judge how far the knee is bent without visual or biologically similar fashion, albeit with reduced energy injection,
sensory feedback such as the leg hitting full extension. They are but with the modified knee control, the knee did not prove to be
also familiar and trained to use this method of gait with their as well-behaved.
current passive prostheses. Because the ET was designed to utilize a very specific
knee/ankle torque and kinematics relationship, the lack of
5.3.3. Gait Improvements constraint in the control of the kinematics allowed the device
The current prosthesis control uses motor position setpoints to attain unsafe conditions, and could not be used as designed.
which change the position of the motor side of the SEA based on a It is possible that in a system that is designed to retain angle
heuristic rule-based state machine. This method requires that the relationships between the knee and ankle the ET system would
dynamic and contact forces of the user are somewhat near to the work as designed, although because of the actuator effort to keep
normal values from which the targets were derived because both kinematic accuracy, it isn’t guaranteed that this would be useful
the generated kinematics and torques of the joints are completely toward prosthesis energy reduction. Another option would be
dependent on these external forces. to determine a gait suitable for an individual without the use of

Frontiers in Neurorobotics | www.frontiersin.org 17 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

the ET and then adding back ET capability if the gait allows for patterns that utilize the average torque/angle characteristics.
negative energy of the knee to be transmitted. Because the prosthesis does not impose either torque or trajectory
When it was decided that the ET portion of the knee could upon the user, they tend to find gait patterns that are very
not be used, a new trajectory for the position of the knee carriage different from the average biomechanical data. This may be due
was created based on feedback from the people doing the trials, to training and unfamiliarity with the prosthesis, it may have
without full regard to the actual torque/angle characteristics of to do with the nature of the socket interface, inaccuracy of the
the knee. The subjects also did not use the stance flex WA system, control timing, or a combination of other reasons. When the user
and preferred to utilize the end stop of the knee as much as deviates from the average biomechanical trajectories, the energy
possible during stance. This behavior may, in part, be to the way saving functions of the prosthesis are reduced and the device
conventional sockets are set and how people are trained to use functions similarly to other powered prostheses under evaluation
prostheses. Knee hyperextension is often used for knee “stability” today.
during the stance phase using conventional prostheses and it is We conclude with a summary of points learned while
possible with a modified socket alignment this tendency could developing this prosthesis:
be reduced. Control and setup were the main reasons that the
– Bench testing showed the quasi-static stiffness based prosthesis
behavior of the prosthesis resembled that of a passive prosthesis.
can reproduce average walking knee and ankle joint torques
It is clear that a refined, better tuned, control system with
when the output of the prosthesis was constrained with
clearer goals in system constraints will be required to produce
external motors. Under these conditions the ET was found
more normal knee torque/angle characteristics. The top level
to be capable of transferring energy from the knee to the
state machine system was not the most adaptable system that
ankle and a considerable energy consumption reduction of the
could have been chosen for this task, although it was sufficient
motors was found.
to obtain preliminary walking gait. Improvements to this will
– The prosthesis was used in a preliminary validation
need to include much more training of the user to utilize the
experiment with four amputee subjects and through
WA correctly as well as adding in a better position trajectory
modification of the main actuator behavior, the prosthesis was
of the knee carriage to provide expected knee torque. For
able to create a stable gait cycle with all subjects.
topics such as gait symmetry and metabolic consumption, better
– Using the quasi-stiffness estimations from average
performance is needed than this control provided. For a fairly
biomechanical data for the stiffness of the ankle springs
complete discussion on different control methods in prostheses
creates behaviors that resemble the average biological data in
and exoskeletons, readers should refer to Tucker et al. (2015)
walking trials.
which provides a large array of different methods that may be
– Using the quasi-stiffness estimations for the stiffness of the
implemented or examine online optimization methods (Kim
knee springs did not provide sufficiently average kinematics
et al., 2017; Zhang et al., 2017) to achieve better performance
and torques during walking trials. Even though it is possible
from the control system.
to generate average torque and kinematics in the prosthesis it
does not mean the person using it will choose to walk with
6. CONCLUSIONS AND FUTURE WORK average torque and kinematics without stabilizing constraints.
– Energy transfer from the knee to the ankle is possible under
We have created a new, active, combined ankle-knee prosthetic ideal conditions.
system which achieves as much as possible with a passive – Because of deviations in the knee and ankle joint kinematics
approach, using springs that were chosen to match the biological during walking tests, the tests had to be run without the use of
quasi-stiffness of normal gait. These springs are locked and the ET system. These mismatches stem from a combination
unlocked during the gait cycle and combined with an energy of the prosthesis control, which does not constrain the
harvesting system to passively provide the majority of the kinematics, the ET control, which was treated as a locked or
required torque angle characteristics during normal walking, unlocked clutch in this implementation, as well as the way the
while maintaining versatility by providing active actuation. subjects interact with the prosthesis, preferring behaviors that
Under ideal conditions the prosthesis worked on the bench as were not like average biomechanical data.
designed, which also showed a lower motor electrical cost than – In order to overcome differences in kinematics, the motor
most current designs. In particular, the capability of the energy must actuate in a different manner than average biomechanical
transfer system to reduce both the knee and ankle motor work data would suggest which reduces the efficacy of the
was considerable. In the ideal case the prosthesis performs quite quasi-stiffness approach in reducing energy consumption,
well compared to the current array of powered prostheses. particularly in the knee.
In reality there are two major issues with the prosthesis. First – The use of low stiffness springs in the knee determined by
these savings are all but eliminated by the implementation, where quasi-stiffness trajectories limit the ability of the actuator
it takes approximately 10 J/stride to capture a similar amount of to modify the behavior of the knee due to low actuator
work in both the WA and ET systems. More efficient, and in the bandwidth, although solutions can be found that provide
case of the ET more controllable, locking mechanisms should be stable gait.
found to better utilize these systems. Second is that the people – A simple state machine system with a number of
wearing the prosthesis do not seem to be able to find walking experimentally tuned variables to set thresholds for actuation

Frontiers in Neurorobotics | www.frontiersin.org 18 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

and timing was implemented and work sufficiently to experiments at VUB. SH helped with data analysis and derived
provide basic gait functions. These thresholds were primarily governing torque equations. BV supervised experiments at
determined by feedback from the patients, and resulted in a VUB and proofread the manuscript. MM provided the WSA
powered ankle actuation that was similar to biological ankle system and assisted with top level control. RM presided over
function at a reduced amplitude and knee behavior similar to the experimental sessions at FDG, gained ethical approvals,
current microcontroller devices. and lead patient recruitment. NV supervised the experimental
– It was determined that a much longer training period must sessions. DL supervised the hardware design and proofread the
be allowed for the users before measurements. Because the manuscript.
prosthesis behaves quite differently to a standard prosthesis,
the user must learn to have high trust the device and they must FUNDING
have a detailed understanding of the behavior of the device and
how it can be utilized. Training alone may improve kinematics, This study was partly funded by the European Commission under
although it is not the only issue. the CYBERLEGs project (Grant #287894), within the Seventh
– New gait detection and control methods should be able to Framework Programme (FP7-ICT-2011-7), and the CYBERLEGs
better utilize the passive aspects of the prosthesis, but how this Plus Plus project (Grant #731931), within the H2020 framework
can best be accomplished is a focus of future work. (H2020-ICT-25-2016-2017). JG received a Ph.D. grant from
Flanders Innovation & Entrepreneurship (VLAIO).
ETHICS STATEMENT
ACKNOWLEDGMENTS
This study was carried out in accordance with the
recommendations of Fondazione Don Carlo Gnocchi. Ethical We would like to thank Marnix De Boom and Marc Luypaert
approval for the protocol was obtained through FDG. All for their work building the Beta-Prosthesis. Thank you to
subjects gave written informed consent in accordance with the Carlos Rodriguez-Guerrero for assistance in proofreading the
Declaration of Helsinki. manuscript. We would also like to thank our test subjects for their
participation in our project. www.cyberlegs.eu.
AUTHOR CONTRIBUTIONS
SUPPLEMENTARY MATERIAL
LF designed and constructed the hardware, conducted
experiments, analyzed data, and wrote the manuscript. JG The Supplementary Material for this article can be found
designed and constructed the hardware, conducted experiments, online at: https://www.frontiersin.org/articles/10.3389/fnbot.
analyzed data, and proofread the manuscript. RJ-F supervised 2018.00080/full#supplementary-material

REFERENCES Geeroms, J., Flynn, L., Jimenez-Fabian, R., Vanderborght, B., and Lefeber,
D. (2017). Design and energetic evaluation of a prosthetic knee joint
Ambrozic, L., Gorsic, M., Geeroms, J., Flynn, L., Molino Lova, R., Kamnik, actuator with a lockable parallel spring. Bioinspiration Biomimetics 12:026002.
R., et al. (2014). Cyberlegs: a user-oriented robotic transfemoral prosthesis doi: 10.1088/1748-3190/aa575c
with whole-body awareness control. Rob. Autom. Mag. IEEE 21, 82–93. Geeroms, J., Flynn, L., Jimenez-Fabian, R., Vanderborght, B., and Lefeber, D.
doi: 10.1109/MRA.2014.2360278 (2018). Energetic analysis and optimization of a maccepa actuator in
Au, S., and Herr, H. (2008). Powered ankle-foot prosthesis: the importance an ankle prosthesis. Auton. Rob. 42, 147–158. doi: 10.1007/s10514-017-
of series and parallel motor elasticity. IEEE Rob. Autom. Mag. 15, 52–59. 9641-1
doi: 10.1109/MRA.2008.927697 Genin, J. J., Bastien, G. J., Franck, B., Detrembleur, C., and Willems, P. A. (2008).
Caputo, J. M., and Collins, S. H. (2014). Prosthetic ankle push-off work reduces Effect of speed on the energy cost of walking in unilateral traumatic lower limb
metabolic rate but not collision work in non-amputee walking. Sci. Rep. 4:7213. amputees. Eur. J. Appl. Physiol. 103, 655–663. doi: 10.1007/s00421-008-0764-0
doi: 10.1038/srep07213 Gilbert, B., and Lambry, D. (2013). Joint Actuation Mechanism for a
Cuccurullo, S. (ed.). (2004). Physical Medicine and Rehabilitation Board Review. Prosthetic and/or Orthotic Device Having a Compliant Transmission. Patent No.
New York, NY: Demos. US8435309. Available online at: https://patents.google.com/patent/US8435309
Flynn, L., Geeroms, J., Jimenez-Fabian, R., Vanderborght, B., Vitiello, N., and Giovacchini, F., Vannetti, F., Fantozzi, M., Cempini, M., Cortese, M., Parri, A., et al.
Lefeber, D. (2015). Ankle-knee prosthesis with active ankle and energy transfer: (2015). A light-weight active orthosis for hip movement assistance. Rob. Auton.
Development of the CYBERLEGs Alpha-Prosthesis. Rob. Auton. Syst. 73, 4–15. Syst. 73, 123–134. doi: 10.1016/j.robot.2014.08.015
doi: 10.1016/j.robot.2014.12.013 Goršic, M., Kamnik, R., Ambrožic, L., Vitiello, N., Lefeber, D., Pasquini, G., et
Flynn, L. L., Geeroms, J., van der Hoeven, T., Vanderborght, B., and Lefeber, D. al. (2014). Online phase detection using wearable sensors for walking with a
(2018). Vub-cyberlegs cybathlon 2016 beta-prosthesis: case study in control of robotic prosthesis. Sensors 14, 2776–2794. doi: 10.3390/s140202776
an active two degree of freedom transfemoral prosthesis. J. Neuroeng. Rehabil. Grabowski, A. M., and D’Andrea, S. (2013). Effects of a powered ankle-foot
15:3. doi: 10.1186/s12984-017-0342-y prosthesis on kinetic loading of the unaffected leg during level-ground walking.
Garate, V. R., Parri, A., Yan, T., Munih, M., Lova, R. M., Vitiello, N., et al. J. Neuroeng. Rehabil. 10, 1–12. doi: 10.1186/1743-0003-10-49
(2016). Walking assistance using artificial primitives: a novel bioinspired Hafner, B. J., Sanders, J. E., Czerniecki, J. M., and Fergason, J. (2002). Transtibial
framework using motor primitives for locomotion assistance through a energy-storage-and-return prosthetic devices: a review of energy concepts and
wearable cooperative exoskeleton. IEEE Rob. Autom. Mag. 23, 83–95. a proposed nomenclature. Bull. Prosthet. Res. 39, 1–11. Available online at:
doi: 10.1109/MRA.2015.2510778 https://www.rehab.research.va.gov/jour/02/39/1/pdf/hafner.pdf

Frontiers in Neurorobotics | www.frontiersin.org 19 December 2018 | Volume 12 | Article 80

Flynn et al. The CYBERLEGs Beta-Prosthesis

Handford, M. L., and Srinivasan, M. (2016). Robotic lower limb prosthesis design Pfeifer, S., Pagel, A., Riener, R., and Vallery, H. (2014). Actuator with angle-
through simultaneous computer optimizations of human and prosthesis costs. dependent elasticity for biomimetic transfemoral prostheses. IEEE/ASME
Nat. Sci. Rep. 6:19983. doi: 10.1038/srep19983 Trans. Mechatronics 20, 1384–1394. doi: 10.1109/TMECH.2014.2337514
Handford, M. L., and Srinivasan, M. (2018). Energy-optimal human walking with Plooij, M., Mathijssen, G., Cherelle, P., Lefeber, D., and Vanderborght, B. (2015).
feedback-controlled robotic prostheses: A computational study. IEEE Trans. Lock your robot: a review of locking devices in robotics. IEEE Rob. Autom.
Neural Syst. Rehabil. Eng. 26, 1773–1782. doi: 10.1109/TNSRE.2018.2858204 Mag. 22, 106–117. doi: 10.1109/MRA.2014.2381368
Heins, S., Flynn, L., Geeroms, J., Lefeber, D., and Ronsse, R. (2018). Torque control Pratt, J., Krupp, B., and Morse, C. (2002). Series elastic actuators for high fidelity
of an active elastic transfemoral prosthesis via quasi-static modelling. Rob. force control. Ind. Robot 29, 234–241. doi: 10.1108/01439910210425522
Auton. Syst. 107, 100–115. doi: 10.1016/j.robot.2018.05.015 Ronsse, R., De Rossi, S., Vitiello, N., Lenzi, T., Carrozza, M., and Ijspeert,
Herr, H. M., and Grabowski, A. M. (2012). Bionic ankle-foot prosthesis normalizes A. (2013). Real-time estimate of velocity and acceleration of quasi-
walking gait for persons with leg amputation. Proc. R. Soc. Lon. B 279, 457–464. periodic signals using adaptive oscillators. Rob. IEEE Trans. 29, 783–791.
doi: 10.1098/rspb.2011.1194 doi: 10.1109/TRO.2013.2240173
Hitt, J. K., Bellman, R., Holgate, M., Sugar, T. G., and Hollander, K. W. (2008). “The Rouse, E. J., Mooney, L. M., and Herr, H. M. (2014). Clutchable series-elastic
sparky (spring ankle with regenerative kinetics) project: design and analysis of actuator: implications for prosthetic knee design. Int. J. Rob. Res. 33, 1611–
a robotic transtibial prosthesis with regenerative kinetics,” in 2007 Proceedings 1625. doi: 10.1177/0278364914545673
of the ASME International Design Engineering Technical Conferences and Ruiz Garate, V., Parri, A., Yan, T., Munih, M., Molino Lova, R., Vitiello, N., et
Computers and Information in Engineering Conference, DETC2007, vol. 5 PART al. (2017). Experimental validation of motor primitive-based control for leg
C (Las Vegas, NV), 1587–1596. exoskeletons during continuous multi-locomotion tasks. Front. Neurorob.
Jimenez-Fabian, R., Flynn, L., Geeroms, J., Vitiello, N., Vanderborght, B., and 11:15. doi: 10.3389/fnbot.2017.00015
Lefeber, D. (2015). Sliding-Bar MACCEPA for a powered ankle prosthesis. J. Segal, A. D., Orendurff, M. S., Klute, G. K., McDowell, M. L., Pecoraro, J. A., Shofer,
Mech. Rob. 7, 1–2. doi: 10.1115/1.4029439 J., et al. (2006). Kinematic and kinetic comparisons of transfemoral amputee
Jimenez-Fabian, R., Geeroms, J., Flynn, L., Vanderborght, B., and Lefeber, gait using c-leg R and mauch sns R prosthetic knees. J. Rehabil. Res. Dev. 43,
D. (2017). Reduction of the torque requirements of an active ankle 857. doi: 10.1682/JRRD.2005.09.0147
prosthesis using a parallel spring. Rob. Auton. Syst. 92, 187–196. Shamaei, K., Sawicki, G. S., and Dollar, A. M. (2013). Estimation of quasi-stiffness
doi: 10.1016/j.robot.2017.03.011 and propulsive work of the human ankle in the stance phase of walking. PLoS
Kim, M., Ding, Y., Malcolm, P., Speeckaert, J., Siviy, C. J., Walsh, C. J., et ONE 8:e59935. doi: 10.1371/journal.pone.0059935
al. (2017). Human-in-the-loop bayesian optimization of wearable device Staros, A. (1957). The sach (solid-ankle cushion-heel) foot. Orthop. Prosth. Appl. J.
parameters. PLoS ONE 12:e0184054. doi: 10.1371/journal.pone.0184054 11, 23–31.
LaPrè, A. K., and Sup, F. (2013). “Redefining prosthetic ankle mechanics: Non- Sup, F., Varol, H. A., Mitchell, J., Withrow, T. J., and Goldfarb, M.
anthropomorphic ankle design,” in 2013 IEEE 13th International Conference on (2009). Preliminary evaluations of a self-contained anthropomorphic
Rehabilitation Robotics (ICORR) (Seattle, WA), 1–5. transfemoral prosthesis. IEEE/ASME Trans. Mechatr. 14, 667–676.
Lawson, B. E., Mitchell, J., Truex, D., Shultz, A., Ledoux, E., and doi: 10.1109/TMECH.2009.2032688
Goldfarb, M. (2014). A robotic leg prosthesis: design, control, and Tucker, M. R., Olivier, J., Pagel, A., Bleuler, H., Bouri, M., Lambercy, O., et al.
implementation. IEEE Rob. Autom. Mag. 21, 70–81. doi: 10.1109/MRA.2014. (2015). Control strategies for active lower extremity prosthetics and orthotics:
2360303 a review. J. Neuroeng. Rehabil. 12, 1–29. doi: 10.1186/1743-0003-12-1
Lenzi, T., Cempini, M., Hargrove, L., and Kuiken, T. (2018). Design, development, Unal, R., Klijnstra, F., Burkink, B., Behrens, S. M., Hekman, E. E. G., Stramigioli,
and testing of a lightweight hybrid robotic knee prosthesis. Int. J. Rob. Res. 37, S., et al. (2013). “Modeling of walkmech: a fully-passive energy-efficient
953–976. doi: 10.1177/0278364918785993 transfemoral prosthesis prototype,” in Rehabilitation Robotics (ICORR), 2013
Lenzi, T., Sensinger, J., Lipsey, J., Hargrove, L., and Kuiken, T. (2015). “Design IEEE International Conference on (Seattle, WA), 1–6.
and preliminary testing of the RIC hybrid knee prosthesis.” in Proceedings of Van Ham, R., Vanderborght, B., van Damme, M., Verrelst, B., and Lefeber, D.
the Annual International Conference of the IEEE Engineering in Medicine and (2007). MACCEPA, the mechanically adjustable compliance and controllable
Biology Society, EMBS (Milan), 1683–1686. equilibrium position actuator: Design and implementation in a biped robot.
Malcolm, P., Derave, W., Galle, S., and De Clercq, D. (2013). A simple exoskeleton Rob. Auton. Syst. 55, 761–768. doi: 10.1016/j.robot.2007.03.001
that assists plantarflexion can reduce the metabolic cost of human walking. Winter, D. A. (2009). Biomechanics and Motor Control of Human Movement, 4th
PLoS ONE 8:e56137. doi: 10.1371/journal.pone.0056137 Edn. Hoboken, NJ: John Wiley and Sons.
Matthys, A., Cherelle, P., Van Damme, M., Vanderborght, B., and Lefeber, D. Zhang, J., Fiers, P., Witte, K. A., Jackson, R. W., Poggensee, K. L., Atkeson, C. G.,
(2012). “Concept and design of the hekta (harvest energy from the knee and et al. (2017). Human-in-the-loop optimization of exoskeleton assistance during
transfer it to the ankle) transfemoral prosthesis,” in 2012 4th IEEE RAS EMBS walking. Science 356, 1280–1284. doi: 10.1126/science.aal5054
International Conference on Biomedical Robotics and Biomechatronics (BioRob)
(Rome), 550–555. Conflict of Interest Statement: The authors declare that the research was
Mauch, H. (1968). Stance control for above-knee artificial legs-design conducted in the absence of any commercial or financial relationships that could
considerations in the s-n-s knee. Bull. Prosthet. Res. 20, 61–72. be construed as a potential conflict of interest.
Morgenroth, D. C., Segal, A. D., Zelik, K. E., Czerniecki, J. M., Klute, G.
K., Adamczyk, P. G., et al. (2011). The effect of prosthetic foot push- Copyright © 2018 Flynn, Geeroms, Jimenez-Fabian, Heins, Vanderborght, Munih,
off on mechanical loading associated with knee osteoarthritis in lower Molino Lova, Vitiello and Lefeber. This is an open-access article distributed under the
extremity amputees. Gait Posture 34, 502–507. doi: 10.1016/j.gaitpost.2011. terms of the Creative Commons Attribution License (CC BY). The use, distribution
07.001 or reproduction in other forums is permitted, provided the original author(s) and
Parri, A., Martini, E., Geeroms, J., Flynn, L., Pasquini, G., Crea, S., et al. (2017). the copyright owner(s) are credited and that the original publication in this journal
Whole body awareness for controlling a robotic transfemoral prosthesis. Front. is cited, in accordance with accepted academic practice. No use, distribution or
Neurorob. 11:25. doi: 10.3389/fnbot.2017.00025 reproduction is permitted which does not comply with these terms.

Frontiers in Neurorobotics | www.frontiersin.org 20 December 2018 | Volume 12 | Article 80