Home Search Collections Journals About Contact us My IOPscience

Design and control of a bio-inspired soft wearable robotic device for ankle–foot rehabilitation

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2014 Bioinspir. Biomim. 9 016007

(http://iopscience.iop.org/1748-3190/9/1/016007)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 128.237.196.159
This content was downloaded on 16/01/2014 at 16:59

Please note that terms and conditions apply.

 Bioinspiration & Biomimetics
Bioinspir. Biomim. 9 (2014) 016007 (17pp) doi:10.1088/1748-3182/9/1/016007

Design and control of a bio-inspired soft

wearable robotic device for ankle–foot
rehabilitation
Yong-Lae Park 1 , Bor-rong Chen 2 , Néstor O Pérez-Arancibia 3 ,
Diana Young 4,5 , Leia Stirling 6 , Robert J Wood 4,5 , Eugene C Goldfield 5,7
and Radhika Nagpal 4,5
1
Robotics Institute, Carnegie Mellon University, Pittsburgh, PA 15213, USA
2
BioSensics, LLC., Cambridge, MA 02142, USA
3
Department of Aerospace and Mechanical Engineering , University of Southern California,
Los Angeles, CA 90089, USA
4
School of Engineering and Applied Sciences, Harvard University, Cambridge, MA 02138, USA
5
Wyss Institute for Biologically Inspired Engineering, Harvard University, Boston, MA 02115, USA
6
Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge,
MA 02139, USA
7
Boston Children’s Hospital, Boston, MA 02115, USA
E-mail: ylpark@cs.cmu.edu

Received 5 July 2013, revised 23 October 2013

Accepted for publication 22 November 2013
Published 16 January 2014

Abstract
We describe the design and control of a wearable robotic device powered by pneumatic
artificial muscle actuators for use in ankle–foot rehabilitation. The design is inspired by the
biological musculoskeletal system of the human foot and lower leg, mimicking the
morphology and the functionality of the biological muscle–tendon–ligament structure. A key
feature of the device is its soft structure that provides active assistance without restricting
natural degrees of freedom at the ankle joint. Four pneumatic artificial muscles assist
dorsiflexion and plantarflexion as well as inversion and eversion. The prototype is also
equipped with various embedded sensors for gait pattern analysis. For the subject tested, the
prototype is capable of generating an ankle range of motion of 27◦ (14◦ dorsiflexion and 13◦
plantarflexion). The controllability of the system is experimentally demonstrated using a linear
time-invariant (LTI) controller. The controller is found using an identified LTI model of the
system, resulting from the interaction of the soft orthotic device with a human leg, and
model-based classical control design techniques. The suitability of the proposed control
strategy is demonstrated with several angle-reference following experiments.

Keywords: soft robotics, bio-inspired robotics, wearable robotics, active orthosis, pneumatic
artificial muscles (PAMs), soft sensors, feedback control
(Some figures may appear in colour only in the online journal)

1. Introduction assistance during gait but also for preventing the development
of abnormal gaits over time.
Ankle–foot pathologies in patients with neuromuscular In this paper, we present an active soft orthotic
disorders, caused by cerebral palsy (CP), amyotrophic lateral
device—incorporating sensing, control, and actuation—for
sclerosis, multiple sclerosis, or stoke, can result in abnormal
gaits over time, such as drop foot, equinus, and crouch use in treating ankle–foot pathologies associated with
gait [1]. The treatment of these pathologies using active neuromuscular disorders [2]. The design mimics the
orthotics has the potential not only for providing immediate muscle–tendon–ligament–skin architecture in the biological

1748-3182/14/016007+17$33.00 1 © 2014 IOP Publishing Ltd Printed in the UK

Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

degrees of freedom of the joint. To avoid limiting the motion

(a) (b) (c)
of the joint, the applicability of using shape memory alloy
(SMA) wire springs has been investigated to create a prototype
soft actuated knee brace [26]. Although SMA wires showed
potential to be embedded in a flexible structure, achieving
appropriate power management and response time presented
strong challenges for developing a practical orthotic device.
Soft wearable devices using cable-driven actuation [27] and
pneumatic muscle actuation [28] have shown similar concepts
to the proposed system in this paper. However, they were
targeted to upper extremity rehabilitation.
In contrast to prior orthotic designs that either constrain
Figure 1. Design and prototype of bio-inspired active soft orthotic or actuate the ankle joint only in the sagittal plane, the
device. (a) Lower leg muscle anatomy. (b) Design concept. device presented in this paper (figure 2) has multiple artificial
(c) Actual prototype. muscle–tendon units that mimic not only the morphology
but also the functionality of the biological muscles in terms
musculoskeletal system of the human ankle, as shown in of its ability to control varied sagittal and mediolateral
figure 1. ankle motions such as dorsiflexion, plantarflexion, inversion,
There have been various approaches for treating ankle– and eversion. In addition to this biomimetic design, the
foot pathologies. One of the most typical solutions is to wear device uses soft sensors, actuators, and structures that do not
rigid (or semi-rigid) ankle braces, such ankle–foot orthoses constrain the user’s natural degrees of freedom, which is not
(AFOs) [3–6]. These devices improve gait abnormalities by possible in traditional exoskeleton robots. The prototype is
forcing the ankle joint angle to be close to 90◦ . However, made of soft plastics and composite materials that comprise
long-term use of these passive devices causes disuse atrophy a light and flexible, but robust structure. The device is
of muscles and consequently makes the user physically also equipped with various sensors, including strain sensors,
dependent on the device [7, 8]. Furthermore, long-term use inertial measurement units (IMUs), and pressure sensors. The
of passive AFOs may induce neural adaptations with gradual strain sensor and the IMUs provide information on joint
reduction of muscle activity over time [9]. For example, angle, while the pressure sensors detect foot contact with the
previous work shows neural adaptation of motor representation ground. In order to demonstrate the suitability of the proposed
occurs with spinal cord injury or immobilization [10, 11]. approach, we present experimental results by showing the
Thus, passive AFOs can increase the time needed in physical characterization of the individual components, as well as
therapy to compensate for these unwanted long-term effects. demonstrate controllability of the device using a linear time-
Active devices are another approach to address invariant (LTI) controller in several angle-reference following
pathological gaits. While active devices require relatively experiments. Ultimately, we envision a system that not only
complicated systems compared to passive braces, with can increase the efficiency of rehabilitation by enhancing
appropriate control, they not only can provide higher muscle usage but can also potentially provide assistance during
independence, as shown with robotic prosthetic legs [12, 13], gait by increasing safety and stability.
but also can be used for re-educating the neuro-motor The rest of the paper is organized as follows. Section 2
system [14], with previous work showing the efficacy of describes the design and fabrication of the prototype.
adaptive control with active devices [15]. In addition, robotic Section 3 discusses the sensor validation and mechanical
gait training and rehabilitation has demonstrated benefits characterization of the system. Section 4 presents the
on improving abnormal gaits associated with neuromuscular control strategy and experimental cases. We conclude with a
disorders [16, 17]. Several actuation methods have been discussion of future work, which includes potential extensions
employed to obtain active elements. For example, brushless of the prototype.
dc motors have been used for an ankle rehabilitation robot
[18], as well as in series elastic actuators with a rigid plastic 2. Design
AFO frame [15, 19]. Powered AFOs have been proposed
for supplementing weak muscles for both dorsiflexion and The basic design concept was to create an external artificial
plantarflexion using McKibben pneumatic artificial muscle muscle–tendon–ligament structure by mimicking the existing
actuators [20, 21] and bidirectional pneumatic rotary actuators biological musculoskeletal system, as described in figure 1.
[22], combined with carbon fiber reinforced foot and leg This concept allows us to achieve three design requirements:
frames. In [23], a robotic tendon using a spring-based linear no rigid frame structures, no constraints on the natural joint
actuator has been proposed to provide active gait assistance. motions, and non-invasiveness. The prototype is composed of
Active orthotic devices using magnetorheological fluid [24] three physical layers: base, actuation, and sensing. The control
and electrorheolegical fluid [25] have been developed to hardware connects the actuation and sensing layers to enable
provide variable damping characteristics. However, the the execution of complex control rules. The entire prototype,
devices mentioned above all require rigid frames and linkages including electronics and batteries but not including the mass
and mechanical joints, which reduce the number of natural of the air source, weighs approximately 950 g.

2
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

(a)

(b)

Figure 2. Main design of the active soft orthotic device (highlighting key components). (a) Front view of the prototype integrated on a
human leg showing the three anterior muscles (M1: tibialis anterior, M2: extensor digitorum longus, and M3: peroneous tertius). (b) Side
view of the prototype showing the posterior muscle (M4: gastrocnemius) and the strain sensors.

2.1. Base foot stability control during the ground contact phase of
walking.
The base layer contains the foot, ankle, and knee braces that The actual prototype worn on a subject’s right leg is shown
make physical contacts with the wearer’s skin. Through this in figure 2. Four off-the-shelf pneumatic artificial muscles
layer, the actuation forces are transmitted to the human body. A (DMSP-10-120-200, 180, 160, and 140, Festo, Esslingen,
flexible commercial knee strap (419 Multi-Action Knee Strap, Germany) were used for actuation, and two off-the-shelf
McDavid, Woodridge, IL 60517, USA), an ankle brace (511 miniature solenoid valves (NEX-2-03-L, Parker Hannifin
Elastic Ankle Support, McDavid, Woodridge, IL 60517, USA), Corp., Cleveland, OH 44124, USA) were installed for air
and a leather shoe (Performa, Vibram, Albizzate, Italy), with injection and release of each muscle. The proximal side of
no rigid components, were modified to be used as the knee, each muscle was anchored to the knee brace, and the distal side
ankle, and foot braces, respectively. was anchored to the foot brace through metal tendon cables
that are flexible but inextensible. Metal hooks were firmly
2.2. Actuation sewn onto the multiple places of the knee and foot braces to
provide anchoring points for the tendon cables. The hooks
The actuation layer is composed of artificial muscles, tendons, made the muscles easy to attach to and detach from the braces.
and ligaments. Four artificial muscles were placed on the lower Since the knee brace has anchoring points only below the knee
leg (three anterior muscles for dorsiflexion, inversion, and joint, the device does not constrain the knee range of motion.
eversion, as well as one posterior muscle for plantarflexion), Muscles 1 and 3 (mimicking the tibialis anterior and peroneous
with their artificial tendons anchored at the knee brace tertius, respectively) provide inversion and eversion functions,
and the foot brace. The artificial muscles were designed respectively, as well as active dorsiflexion, while muscle 2
as counterparts to the biological muscles for dorsiflexion, (mimicking the extensor digitorum longus) provides only
plantarflexion, inversion, and eversion, respectively, so that dorsiflexion. Muscle 4 (mimicking gastrocnemious) provides
the device could provide the supplementary forces to the active plantarflexion. Figure 3 shows possible active ankle
corresponding muscles. The artificial tendons were placed motions generated by different combinations of actuated
close to where the corresponding biological tendons were muscles. This agonist–antagonist muscle architecture not only
located. As the artificial muscles contract, their artificial can create desired ankle motions but also can provide physical
tendons pull the anchors on the foot brace, resulting in support by increasing the stiffness of the ankle joint with co-
a corresponding ankle motion. To prevent the knee brace contraction.
from slipping during actuation, thin non-slip silicone pads Figure 2(a) shows the tendon system. While muscles 1, 3,
were firmly fixed at the inside of the knee brace making and 4 each have only one anchor on the distal side, muscle 2
physical contacts to the wearer’s skin. Since the four artificial has multiple anchors on the foot brace. The differential tendon
muscles can be actuated independently, they can generate mechanism of muscle 2, similar to the foot design of a wall-
active mediolateral motions, such as inversion and eversion, climbing robot [29], distributes the pulling force from one
as well as active sagittal motions, such as dorsiflexion and tendon cable to four anchors. Flexible Teflon R
tubes, sewn
plantarflexion. These mediolateral motions will be useful for onto the foot brace, secure the paths of force transmission

3
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

(a)
(a) (b) c(c)
((c)) (d) (e(e)
(e)) (f ) (g)
(g)
(g) (h)

Figure 3. Actively assisted ankle motions: (a) dorsiflexion (b) with all three anterior muscles contracted; (c) plantarflexion (d) with only
muscle 4 contracted; (e) inversion ( f ) with only muscle 1 contracted; (g) eversion (h) with only muscle 3 contracted.

(a) (b) (a) (b)

θ θ
θ
a2 θ2 a3 a1
θ1 a2
θ3 a4
a1
a3

(c) a3 Figure 5. (a) Strain sensor prototype with dimensions. (b) Prototype
r3
in action showing resistance change with strain (unit: ).
a2 r2 φ2 r4 a4
φ1
r1 2.3. Sensing
a1
The sensors are the outermost layer of the prototype and are
Figure 4. Actuation system configuration and its parameters composed of three sensing modalities. We are interested in
(ai : anchor location of tendon i, Fi : contraction force of muscle i information about the configuration of the leg and how the
connected to tendon i, θi : vertical angle of tendon i, and φi :
horizontal angle of tendon i.). (a) Side view of foot to show sagittal user is interacting with the ground. Thus we focus on strain,
motions. (b) Front view of foot to show mediolateral motions. (c) inertial, and pressure sensors.
Top view of foot.
2.3.1. Strain sensor. Two custom-built strain sensors were
by allowing the cables to smoothly slide through with muscle used for measuring the ankle joint angles, as shown in
actuation. In addition to the teflon tubes, the artificial ligaments figure 2(b). One was attached on the dorsal side of the
made of nylon straps were added to constrain the lateral ankle for measuring sagittal angle changes, and the other
motions of the cables while allowing only axial motions. was placed on the medial side for detecting mediolateral
Figure 4 shows how the actuation system of the prototype motions. Although there are different types of commercial
applies torques to the ankle based on the foot and ankle strain gauges, none of them are designed for accurately
models in [6]. Dorsiflexion is achieved by the cooperation measuring large strain on a flexible surface. Figure 5 shows
of three anterior muscles, and plantarflexion by the posterior the design and the actual prototype of the custom-built
strain sensors. Microchannels, filled with a liquid metal
muscle. Assuming the force directions at the tendon anchors
alloy, Eutectic Gallium–Indium (EGaIn), were embedded in
are perpendicular to the sagittal plane, their torques can be
a silicone elastomer (EcoFlex0030, Smooth-On, Inc., Easton,
calculated as
PA 18042, USA) sheet, as described in [30]. When the material
τdorsi = F1 r1 sin θ1 cos φ1 + F2 r2 sin θ2 + F3 r3 sin θ3 cos φ2 , experiences axial strain, the overall channel length increases
and the cross-sectional areas of the channels decrease resulting
(1)
in increased electrical resistance of the microchannel. Since
the microchannels are filled with liquid, the strain sensor is
τplantar = F4 r4 . (2) highly flexible and stretchable. The channel size is 250 μm ×
In the same way, the torques for mediolateral motions are 250 μm, and the overall thickness of the sensor is 1.5 mm. The
nominal resistance at rest is 10.3 . The experimental gauge
τinversion = F1 r1 sin θ1 sin φ1 , (3) factor of the strain sensor is 3.4. The gauge factor did not show
any noticeable difference with a relatively mild temperature
τeversion = F3 r3 sin θ3 sin φ2 . (4) change (between 20 and 40 ◦ C), which was acceptable for an
indoor rehabilitation device as its current form. However, a
Based on the configurations of the artificial tendons and thorough investigation on the performance with more extreme
the dimensions of the prototype, the maximum torques temperature changes is one of our on-going research areas to
the system can apply are 110 Nm, 53 Nm, 20 Nm, and make the device more mobile and assistive in the future. More
21 Nm for dorsiflexion, plantarflexion, inversion, and eversion, details on this strain sensor, such as design, fabrication, and
respectively. calibration, can be found in [31, 32].

4
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

2.3.2. Inertial measurement unit. Two off-the-shelf 9- (a) (b)

degree-of-freedom IMUs (SEN-09623, 9DOF Razor IMU,
Sparkfun Electronics, Boulder, CO 80301, USA) were used
for measuring the orientations of the lower leg and the
foot. These IMUs provide sensor data from three different
types of microelectromechanical systems based sensors: 3D
accelerometer, 3D angular rate gyroscope, and low-field 3D
magnetometer. An on-chip sensor fusion algorithm, provided
by the manufacturer of the IMU boards, derives the 3D
orientation of each IMU using the sensor inputs. From the
Figure 6. Sensor validation experiments. (a) Experimental setup
two IMUs, one mounted on the foot and the other on the lower with a strain sensor, two IMUs, and a goniometer. (b) Sensor
leg, the ankle joint angles, as Euler angles, can be computed validation result with ankle joint angle measurement. More details
by calculating the rotation matrices, as described in [33]. on this characterization can be found in [2].
While it is possible to measure joint angles using only
IMUs (using two per joint), joint angle measurement using the 50 Hz for actuation by accessing all the sensor data using its
strain sensor, with inertial measurement employed only for own MCU (Atmega1280, Atmel, San Jose, CA 95131, USA).
initial calibration, offers the significant advantage of lower Finally, the actuation stage carries out control commands
power consumption. IMU subcomponents, specifically the generated at the previous stage by controlling solenoid valves
gyroscopes that enable angle estimation, require higher power through pulse-width-modulation (PWM). More details on the
consumption than the strain sensor. In this implementation, control hardware can be found in [2].
a single strain sensor consumes approximately 0.625 mW of
power, while the accelerometers and gyroscopes included in 3. Characterization
just one of the IMUs in this design consume approximately
36 mW. Future implementations of this sensor system may The prototype was characterized in two functional categories,
enable power savings by including only one IMU to provide sensing and actuation, to understand how the device performs
absolute angle reference, as well as sleeping the IMU after given the non-rigid nature of all of the components. Power
initial strain sensor calibration. Reducing the number of IMUs consumption is also briefly discussed at the end of this section.
required for joint angle measurement in favor of the strain All the testing was done with one human subject, and the
sensing method also provides design advantages, as the strain testing procedure was reviewed by the Harvard Medical School
sensors may be more easily and comfortably embedded within Committee on Human Studies.
soft architectures.
3.1. Sensor validation
2.3.3. Pressure sensor. Foot pressure sensing is required to
identify ground contact events. While the strain sensors and Three types of sensing were characterized: sagittal angle
IMUs measure the ankle joint angle, the foot pressure sensors sensing, mediolateral angle sensing, and foot pressure sensing.
detect relative pressure distribution on the foot from ground
contact. The pressure sensors determine which part of the foot 3.1.1. Sagittal joint angle sensing. For joint angle
was in contact with the ground in real time. Four off-the- measurement, the dorsal strain sensor was selected as a reliable
shelf force sensitive resistors (FSRs) (FlexiForce-A201-25lb, source of angle information during operation. However, since
Tekscan Inc., Boston, MA 02127, USA) were embedded in a the strain sensor requires calibration, each time a user wears
shoe insole. Although the pressure sensors are currently used the device, due to the change of the sensor location, the IMUs
as on–off switches, they can also be used to measure actual are used for calibrating the strain sensor by establishing a linear
foot pressures with accurate calibration if necessary. mapping between the strain sensor readings and joint angles
measured by the IMUs. The performance of the strain sensor
2.4. Control hardware and the IMUs were evaluated using a commercial optical
goniometer (PASPORT Goniometer Sensor PS-2137, PASCO,
The control hardware (electronics), including batteries, were Roseville, CA 95747, USA).
attached to a modified thigh brace (473 Thigh Sleeve, A validation experiment was conducted to examine the
McDavid, Woodridge, IL 60517, USA), containing multiple performance of the dorsal strain sensor and the IMUs with
micro-controller units (MCUs) to support tasks divided into the goniometer used as ground truth. The two IMUs were
four major stages: sensing, signal processing, control, and attached to the goniometer, and the goniometer was mounted
actuation. to the interior part of the subject’s right foot. The strain
The sensing stage samples all the sensors, including sensor was attached to the anterior part of the ankle, as
strain sensors, IMUs, and a pressure sensor, at 50 Hz using shown in figure 6(a). Once all the sensors were mounted,
independent MCUs (Atmega328P, Atmel, San Jose, CA the subject freely moved the ankle in plantarflexion and
95131, USA). The signal processing stage applies a different dorsiflexion for 20 min. A representative 10 s segment from
algorithm to each sensor type to process different types of the experiment is shown in figure 6(b). The result showed that
sensor data. Then, the control stage runs a control loop at both the strain sensor and IMUs provided accurate joint angle

5
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

(a)

(b)

Figure 7. Mediolateral ankle motions and strain sensor calibration

result.

measurements during the experiment. The mean error of the

IMU measurement was 0.1◦ ± 2.9◦ , with a maximum error of
9.1◦ . The mean error of the strain-derived angle was 0.3 ◦ ± Figure 8. Mechanical characterization results. (a) Maximum
1.6◦ , with a maximum error of 4.3◦ . dorsiflexion and plantarflexion with various PWM duty cycles.
The strain sensor experienced approximately 73% strain (b) Repeated sagittal ankle motions with the PWM duty cycles
during full dorsiflexion and plantarflexion tests. To fully shown in (a).
utilize the calibration region (0–100% strain), the sensor
was stretched with 20% pre-strain and installed from a full 3.2. Actuation system characterization
dorsiflexion position. Also, the two ends of the sensor were
firmly attached to the ankle brace with non-stretchable nylon The mechanical system was characterized to evaluate the
straps to minimize any possible sliding effect, a potential error response time, linearity, and repeatability of the actuation
source, between the sensor and the ankle brace. This type of system. The prototype was installed in the subject’s right leg
strain sensor requires approximately 3N of tensile load for (figure 2). The subject sat on a rigid bench hanging the foot in
100% strain according to [34], which is negligible compared the air and was able to move the leg and the foot without any
to the contraction force of the pneumatic muscles. constraint.
Due to the nonlinearity of pneumatic muscles,
proportional control of the actuators was not as straightforward
3.1.2. Mediolateral joint angle sensing. In addition to sagittal as for other linear actuators, such as dc motors. One method
motions, mediolateral motions were characterized (figure 7) for proportional control is to use proportional valves that
using a pre-stretched strain sensor placed at the side of the control the input air pressure of the muscles by changing input
ankle. Two anterior muscles, M1 and M3, were actuated voltage [35]. Another method is PWM control of binary on/off
alternately creating inversion and eversion motions while the solenoid valves [36, 37]. In our design, binary valves and the
other two muscles, M2 and M4, were fully relaxed, and PWM method were implemented to take advantage of the
the strain sensor signal was measured. Since the goniometer compact form factor of the binary valves. All the experiments
was not able to measure these rotational angles, we used were done using lab-bench air supply. The source air pressure
professional video analysis software (ProAnalyst, Xcitex, was 570 kPa with a flow rate of 19 Lpm during the experiments.
Cambridge, MA 02141, USA) to provide reference angles. The characterization experiment was divided into three
The result showed a linear sensor response for measuring phases: dorsiflexion, hold, and plantarflexion. During the
mediolateral ankle angles. dorsiflexion phase, the system fully contracted all three
anterior muscles while releasing the posterior muscle to reach
3.1.3. Foot pressure sensing. The FSR pressure sensors the maximum dorsiflexion angle. Then, the system held the
were characterized in order to determine the threshold for dorsiflexion angle by closing all the valves. Finally, the system
detecting the ground contact of the foot. With no load, the fully contracted the posterior muscle while releasing the
resistance of the pressure sensors was at or above 1 M. anterior muscles to reach maximum plantarflexion. In this
With a 143 lb adult standing on the insole, the resistance was experiment, we were able to achieve different contraction and
consistently less than 500 k. Thus, 500 k was selected as release speeds of the pneumatic muscles by changing the duty
the threshold for the subject tested. This threshold value is cycle of the PWM controller. The result, shown in figure 8(a),
easily tunable for different users. shows the maximum achievable range of ankle motion was

6
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

Table 1. Pneumatic muscle geometries and gas consumptions when such as the ankle stiffness of a given individual or the actuator
fully contracted (air pressure: 570 kPa). speed, are unknown or difficult to estimate. These complexities
Radius Length Volume Air CO2 make modeling from first principles very difficult. Also, in
Muscle (cm) (cm) (cm3 ) (g) (g) general, it is not uncommon that highly complex systems
exhibit nonlinear dynamics that are not easy to model with
M1 1 13.5 42.4 0.31 0.47
M2 1 12 47.7 0.28 0.42 a priori information only. Further difficulties arise from the
M3 1 10.5 33.0 0.24 0.37 injection of unexpected voluntary or involuntary disturbances
M4 1 15 47.1 0.34 0.51 by the human subject, which demand a control system
Total 1.17 1.77 capable of rejecting low- and medium-frequency disturbances.
A third set of challenges reflects the fact that, in general,
pneumatic actuators are not easy to model and control. In
27◦ (14◦ dorsiflexion and 13◦ plantarflexion) for the case study
particular, although the actuators of the kind considered here
subject.
are lightweight, compliant, and capable of generating large
We also tested repeated dorsiflexion and plantarflexion.
forces, it has been argued that pneumatic actuators have
The system carried out the characterization program
continuously for about 2 min. The result, figure 8(b), shows not been widely used in rehabilitation and assistive robotics
the system could achieve the same angles reliably. While the because they are difficult to control [38]. McKibben pneumatic
target angles were obtained, the angle held gradually decreased artificial muscles, for example, are very challenging to model
during the holding phase due to air leakage of the valves. Thus, from first principles or characterize dynamically, and thus,
feedback control is necessary for more accurate joint angle they have been primarily modeled using a static approach
control, which will be discussed in section 4. [39, 40]. This might explain that, to date, the pneumatic robotic
systems in which sophisticated control strategies have been
implemented are composed of rigid moving parts [38, 41, 42],
3.3. Power consumption
and that, when artificial muscles have been used, the control
3.3.1. Electric power. The system consumes relatively small strategies are mostly model-free, tuning-based or based on
electric power as the actuation is pneumatically powered. heuristic methods such as fuzzy logic [43–45].
Three rechargeable lithium-ion batteries (3.7V-900mAh, There exists an extensive literature on first-principles-
Sparkfun Electronics, Boulder, CO 80301, USA) provide based static modeling of McKibben artificial muscles, in
enough power to operate the entire system for more than 2 h. which the objective has been to find relationships between
The control hardware, excluding the solenoid valves, requires the actuator tension, length, and air pressure [39, 40, 46]. This
2.5 W, and the valve circuits require 1.5 W on average. line of research is useful for understanding the capabilities and
limitations of the actuators. However, static modeling is not
3.3.2. Pneumatic gas consumption. Pneumatic power sufficient for devising trajectory-following high-performance
consumption can be analyzed based on the pressure and the controllers. An alternative approach follows from observing
volume of the compressed air consumed for actuation using that, under controlled experimental conditions, the McKibben
the ideal gas law. The volume of air can be calculated from actuator exhibits a behavior that can be modeled as a second-
the geometry of the contracted muscles. Assuming the gas is order system that maps the effective force generated by the
consumed isothermally, and the molar masses of air and CO2 contractile element to the resulting actuator displacement,
are 29 g and 44 g, respectively, the four muscles consume the provided that the actuation pressure remains constant [43, 47].
total of either 1.17 g of air or 1.77 g of CO2 in one full actuation. This kind of model depends on three parameters that can be
Table 1 summarizes the geometry and gas consumption of interpreted as a mass, a spring coefficient and a damping
each muscle when contracted with pressure of 570 kPa. For coefficient, which depend on the actuation pressure. Thus,
portability, a commercially available cylinder (5GAR9, W. W. assuming small variations of the actuation pressure and using
Grainger Inc., Lake Forest, IL 60045, USA) that contains 567 g appropriate sensors, a control strategy to follow a desired
of liquid CO2 at 827 kPa can be carried with the device. One actuation displacement. However, for angular trajectory
cylinder can provide more than 320 full actuation cycles of all following, this is not practical because we still need a way
four muscles assuming isothermal gas consumption. to exactly specify and generate the required effective force.
Furthermore, estimating only two parameters for a fixed-
4. Control structure low-order model almost certainly would not capture
the complex dynamics of the system proposed here.
Once the basic characterization of the sensing and actuation Considering the issues described above, we avoid
systems interacting with the human subject is completed, modeling the robotic-human system from first principles and
a model-based feedback controller is designed such that instead treat the system as a black box for which we estimate
the orthotic system is used to track desired ankle angular an input–output LTI dynamic mapping, using a subspace
trajectories. The development of such a controller poses several system identification technique. Once a model of the system
challenges, which need to be resolved in order to meet the is found through system identification, a model-based single-
desired requirements of the orthotic system as a rehabilitation input–single-output (SISO) LTI controller is designed. The
tool. Numerous complexities arise as the combined human- idea of treating this complex robotic configuration, which
orthotic system is a complex plant where relevant parameters, mechanically interacts with a human leg, as a black box is

7
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

experimentally validated by evaluating the performance of the 0.5Ts

proposed control strategy for several predetermined angular
references. It is important to mention that classical and modern
synthesis techniques have been recently used in the design and
State v [1 0]T
implementation of controllers for active hard robotic prosthetic
systems [38, 41, 42, 48–50]. In the work presented in this
paper, we expand the use of model-based control strategies T
Ts 2Ts
to the soft architecture described in the previous sections of [0 0]
T ime (sec)
this paper. The specific solution proposed in this research is
described in the following subsections. Here, R, R++ , and Z+ T
denote the sets of real, strictly positive real, and non-negative [0 1]
integer numbers, respectively. The variable t is used to index
discrete time, i.e., t = {kTs }∞ +
k=0 , with k ∈ Z and Ts ∈ R, 0.25Ts
where Ts is the sampling-and-hold time. The variable τ denotes
continuous-time. z−1 denotes the delay operator, i.e., for a Figure 9. Example of an input signal u(t ) written in terms of the
signal x, z−1 x(k) = x(k − 1) and conversely zx(k) = x(k + 1). state v. In this case u(0) = 0.5 and u(1) = −0.25.
In this case, z is also the complex variable associated to the
z-transform. processor (DSP) output and the state v, the input u(t ) can be
defined directly in terms of the state v. We can think of this
4.1. Inputs, outputs, and system identification as a one-to-one mapping between u(t ) and the history of v
From an upper-level input–output perspective we can identify over the valve-cycle starting at time t. In order to describe this
three major subsystems of interest that compose the robotic mapping, we say that a valve-cycle starts when v changes from
system: the solenoid valves (and the associated electronic low- [0 0]T to one of the other two allowable states in (5), i.e.,
level controller), the four pneumatic muscles, and the strain a valve-cycle starts when one and only one of the two valves
sensor, used to measure the ankle joint angle. Here, we focus on is opened. Some time after one of the valves is opened, this
sagittal plane control (dorsiflexion and plantarflexion), which is closed and remains in this position until a new cycle starts.
implies that the set of physical valves can be represented as a Here, the duration of each valve-cycle is fixed at Tv , which
subsystem composed of two abstract valves (an injection valve is chosen to coincide with the sample-and-hold time used for
and a release valve), each of which is in one possible state: signal processing and control, Ts . Therefore, we completely
open or closed. Thus, the set containing the three anterior define the valve-cycle starting at the discrete time t as follows
muscle injection valves and the posterior muscle release below.
valve is modeled as one abstract injection valve. Similarly,
Definition 1. In terms of the input signal u(t ) ∈ [−1, 1], the
the set containing the three anterior muscle release valves
valve-cycle at time t is defined according to the rules.
and the posterior muscle injection valve is modeled as one
abstract release valve. The ankle joint angle is the controlled • If u(t )  0, the corresponding valve-cycle starts with
variable, which is measured with the custom-built strain v = [1 0]T and stays this way for an amount of time
sensor, previously described. equal to |u(t )|Ts , then the state switches to v = [0 0]T
Since there are eight physical valves and the subject foot and stays this way for the rest of the valve-cycle.
is allowed to rotate in space, the original problem is multiple- • If u(t ) < 0, the corresponding value-cycle starts with
input–multiple-output (MIMO). However, the complexity of v = [0 1]T and stays this way for an amount of time
the problem is greatly reduced by simplifying the movement equal to |u(t )|Ts , then the state switches to v = [0 0]T
of the foot to the sagittal degree of freedom and by defining and stays this way for the rest of the valve-cycle.
the control signal in an intelligent manner, so that the open-
loop mapping becomes SISO. For sagittal plane motions, the
Definition 1 allows one to formulate a SISO LTI
four actuators work synchronously, where the three anterior
identification problem, and consequently, to use classical
muscles always expand or contract simultaneously, while the
model-based controller design techniques. Here, we illustrate
posterior muscle performs synchronously the opposite action.
definition 1 through a simple example. Figure 9 shows an input
Thus, four of the eight solenoid valves are thought of as the
signal u = {u(0), u(1))}, with u(0) = 0.5 and u(1) = −0.25.
abstract element valve 1, and the other four solenoid valves
Since u(0) > 0, the valve-cycle corresponding to Time = 0 s
correspond to the abstract element valve 2. We label the states
starts with v = [1 0]T , and since |u(0)| = 0.5, it follows that
of valve 1 and valve 2, v1 and v2 , respectively. When a valve
is closed, its value is 0, and when open, its value is 1. Thus, v = [1 0]T for 0.5Ts s, and then, v = [0 0]T for the rest
the state-space of the entire system is given by of the valve-cycle. Similarly, since u(1) < 0, the valve-cycle
        corresponding to Time = 1 s starts with v = [0 1]T , and
v 0 0 1 since |u(1)| = 0.25, it follows that v = [0 1]T for 0.25Ts s,
v= 1 ∈ , , , (5)
v2 0 1 0 and then, v = [0 0]T for the rest of the valve-cycle.
where the vector [1 1]T is not an allowable state. Since In this way, the signal u(t ) simultaneously conveys
there are no dynamic relationships between the digital signal two pieces of information at each discrete time t. The

8
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

Pneumatic
u(t) PWM w(τ ) Pneumatic Actuators Strain A/D & y(t)
Generator Valves Interacting Sensor Transformation
with Human Leg

Figure 10. Upper-level physical description of the system from an input–output perspective, according the definitions of u(t ) and y(t ). Note
that y(t ) is a physical variable (angular position), whereas u(t ) is not a physical entity but a mathematical tool used for controller design.

d(t)
u(t) y(t)
P (z) +

Figure 11. Idealized system dynamics. P(z): discrete-time

open-loop plant; u(t ): input signal; y(t ): measured sagittal ankle
angle; d(t ): output disturbance, representing the aggregated effects
of all the disturbances affecting the system.

sign of u(t ) provides the state, v, with which the valve-

cycle starts at time t, and the magnitude of u(t ) determines
how long the state v remains different from [0 0]T during
the corresponding valve-cycle. The whole process can be
regarded as a mapping that transforms the discrete-time signal
u(t ) to a two-dimensional analogous PWM signal, w(τ ) =
[w1 (τ ) w2 (τ )]T , where w1 activates valve 1 and w2 activates
valve 2. Naturally, the choice for the system output is the
sagittal ankle joint angle measured by the strain sensor, labeled
as y(t ).
The definition of u(t ) and the upper-level idealization
Figure 12. Input signal used in the LTI system identification of the
in figure 10 allows one to think of the system as a black plant P. Upper plot: complete input sequence used in the system
box that maps the digital input u(t ) to the digital output y(t ), identification. This is white noise, where each data point is the
where the sampling-and-hold time for signal processing and realization of a uniform distribution. The compactness of the plot
control is Ts = Tv . Thus, assuming an LTI system, P, an (the blue rectangular spot) is a reflection of the high frequency
equivalent block-diagram is depicted in figure 11, where d(t ) content of the signal. Bottom plot: close-up of the input signal used
in the system identification. This segment shows 20 s of the signal.
is an output disturbance representing the aggregated effects of
all the disturbances affecting the system. The system described
by this block-diagram can be estimated using well-known on a rigid bench while his leg hangs freely, as described in
subspace system identification methods that yield state-space section 3. Figures 12 and 13 show the input and output signals
realizations of LTI systems, i.e., {AP , BP , CP , DP } matrices used in the LTI system identification of the plant P.
describing the system according to The resulting identified dynamics of P(z), labeled as
P̂(z), are shown in figure 14. The original 48th-order model
xP (t + 1) = AP xP (t ) + BP u(t ), (6) identified is shown along with reduced order models with
orders 12, 4, and 2, respectively. Since the highest frequencies
y(t ) = CP xP (t ) + DP u(t ) + d(t ), (7) in recorded ankle joint angle signals are smaller than 2 Hz [55],
where, xP (t ) ∈ R is the n-dimensional vector state of the state-
n it is clear that the system dynamics can be represented by the
space realization {AP , BP , CP , DP }, describing P. A realization low-pass second-order LTI filter in figure 14. The identified
is called minimal if the system described by (6)–(7) is both systems have been normalized so that the respective dc gain is
observable and controllable. The main advantage of using 0 dB and the cutoff frequency of P̂ is approximately 0.1 Hz.
subspace identification methods is that the order of the system A cutoff frequency of approximately 2 Hz has been reported
to be identified does not have to be assumed a priori, and for McKibben type actuators [21], which indicates that in the
therefore, the resulting realizations are always minimal. case considered here, the cutoff frequency of 0.1 Hz reflects the
Using the algorithm in [51], with the implementation effect of the mechanical impedance of the user’s limb, but more
described in [52], the system in figure 11 is identified with importantly, the slow response of the solenoid valves used to
the use of 50 000 samples generated using a uniform white- generate the PWM signals. In order to reduce the order of the
noise signal input u(t ) ∈ [−1, 1], at a sample-and-hold rate of system, a state-space realization of the identified 48th-order
50 Hz, i.e., a sampling time Ts = 0.02 s. We use white noise model is balanced [56], and then, a certain number of states,
to excite the system over the entire sampled spectrum, 0 Hz relatively less observable and controllable than the others, are
to the Nyquist frequency, 25 Hz [53, 54]. In the experiments discarded, as explained in [56]. Here, the identified models
performed for identifying the system, the same subject sits were validated using the standard practice of simultaneously

9
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

Figure 15. Implementation of the control loop at the machine level,

using five basic computing functions. The control loop is scheduled
to run every 20 ms. (The total time of each loop is 20 ms.)

d(t)
yd (t) uK (t) y(t)
+ K(z) P̂ (z) +
−

Figure 16. Idealized closed-loop configuration used in the design of

the controller K(z). The signal uK (t ) is not directly applied to the
system P(z), but used in the definition of the control strategy in (8).

Figure 13. Output signal used in the LTI system identification of the
plant P. Upper plot: complete output sequence used in the system sampled, the control signal u(t ) computed, and then using the
identification. Bottom plot: close-up of the output signal used in the information in u(t ), PWM commands are sent to the solenoid
system identification. This segment shows 20 s of the signal. valves that activate the actuators. The algorithms processing
the signals required for the control loops are run at 50 Hz,
Identified Model of Plant P limited by the response time of the solenoid valves (20 ms).
20
When the sensor signal is converted to an angle, a
0 reference input is either generated from a predefined trace
Magnitude (dB)

or computed in real time from an input function. Both the

-20
measured joint angle and the reference are then fed into a
-40 2nd Order higher level control algorithm, to be discussed in the next
4th Order part of this subsection, which produces actuation parameters
-60 12th Order according to definition 1 for the solenoid valves. The PWM
48th Order
-80
duty cycles of the valves are then set according to the control
180 algorithm output in order to contract or release the pneumatic
artificial muscles. Figure 15 illustrates the implementation of
the control loop at the machine level, using five computing
Phase (deg)

90

functions.
0

-90 4.2.2. Upper level controller design. At an upper-level, using

definition 1, controller design means that we are interested in
-180
10
-3
10
-2
10
-1
10
0 1
10
finding rules for generating the input signal u(t ) so that the
Frequency (Hz) output signal y(t ) follows a reference yd (t ) and the control
error ey (t ) = yd (t ) − y(t ) is minimized according to some
Figure 14. Bode plot of the identified model P̂ of P. A 48th-order
model is originally identified, then reduced order models with 12, 4 metric. With this in mind, we propose a nonlinear control law
and 2 states are found, using the balanced reduction method. These with the form ⎧
systems have been normalized so that the respective dc gain is 0 dB. ⎨1 if uK (t )  1
u(t ) = uK (t ) if − 1 < uK (t ) < 1 (8)
⎩
exciting the estimated model and the physical system with the −1 if uK (t )  −1
same signal and then compare the respective outputs [53, 54]. with
uK (t ) = [K(z)ey ](t ), (9)
4.2. Controller implementation and design
where K(z) is an LTI controller designed using the information
4.2.1. Control software The system
implementation. contained in P̂, as defined in subsection 4.1. A depiction of the
identification presented in subsection 4.1 and the controller idealized perfectly linear model used for controller design is
design described in this subsection are implemented and shown in figure 16. Considering that the defined input–output
tested on the main DSP (8-bit Atmel Atmega1280 MCU system P relies on definition 1 and that P̂ was found around
clocked at 8 MHz with 8 KB RAM). In order to implement a specific operating point, K must be designed not only to
and run the closed-loop algorithm, the strain sensor signal is improve the system performance but also to meet minimum

10
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

Estimates of So and T Sinusoidal Reference (Freq. 0.4 Hz; Amp. 2.5 )

(a) 5
20

10

Angle (deg)
2.5
Magnitude (dB)

-10 0

-20 Reference yd(t)

-2.5
-30 Measured Angle y(t)
-5
-40 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30
Estimate of So
-50 Time (sec)
Estimate of T
-60 Signal u(t) for Sinusoidal Reference (Freq. 0.4 Hz; Amp. 2.5 )
180 1

Input [-1:1]
0.5
Phase (deg)

90

0
0
-0.5

-90 -1
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30

-180 Time (sec)

-3 -2 -1 0 1
10 10 10 10 10
Frequency (Hz) (b) Sinusoidal Reference (Freq. 0.4 Hz; Amp. 10 )
12.5
10
Figure 17. Bode plots of the estimated output sensitivity and 7.5
Angle (deg)
complementary sensitivity functions Ŝo = (1 + P̂K)−1 and 5
2.5
T̂ = P̂K(1 + P̂K)−1 of So and T , respectively. Ŝo is plotted in blue 0
-2.5
and T̂ is plotted in green. The cutoff frequency of the high-pass filter -5 Reference yd(t)
Ŝo is between 4 and 5 Hz, which indicates that the idealized nominal -7.5
-10 Measured Angle y(t)
control system in figure 16 is capable of rejecting low-frequency -12.5
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30
disturbances generated by involuntary movements of the human
Time (sec)
subject. The shape of T̂ (z) demonstrates that the controller K(z)
significantly increases the bandwidth of the system. Signal u(t) for Sinusoidal Reference (Freq. 0.4 Hz; Amp. 10 )
1
Input [-1:1]

standards of robustness with respect to plant uncertainty. In this 0.5

case, plant uncertainty is expected as the device interacting 0

with a human leg results in highly complex dynamics. The -0.5

controller K must also be designed such that, in closed loop, the
-1
system rejects low-frequency disturbances, desired trajectories 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30
can be followed, and the gain and phase margins are large Time (sec)
enough to ensure stability and performance robustness to plant
Figure 18. First set of experiments. In both experimental cases, (a)
uncertainties. and (b), the upper plot shows the time-series of the reference signal
The first two capabilities required from the resulting yd (t ) and the measured angular position y(t ), the controlled
closed-loop system are examined using the output disturbance variable. In both cases, the bottom plot shows the time-series of the
sensitivity and complementary sensitivity functions corresponding signal u(t ). In case (a), the reference is a sinusoid
with amplitude 2.5◦ and frequency 0.4 Hz. In case (b), the reference
1 PK is a sinusoid with amplitude 10◦ and frequency 0.4 Hz.
So = , T = , (10)
1 + PK 1 + PK
which can be estimated as Ŝo = (1 + P̂K)−1 and T̂ = P̂K(1 + frequency of 0.4 Hz, with amplitudes of 2.5◦ , 5◦ , 7.5◦ and
P̂K)−1 , respectively. Bode plots of both estimated plants, for 10◦ , while the subject is sitting on a bench with his foot
the system parameters defined in subsection 4.1, are shown in hanging. The experiments corresponding to amplitudes of
figure 17. The gain and phase margins are computed from an 2.5◦ and 10◦ are shown in figure 18. The performance of
estimate L̂ = P̂K of the loop-gain L = PK. The corresponding each of the amplitudes considered here (2.5◦ , 5◦ , 7.5◦ and
values are 5.8 dB and 45.9◦ , approximately, which indicates 10◦ ) is shown in table 2. The performance is measured using
that the resulting closed-loop system is robustly stable. Note the experimental standard deviation (ESD) of the control
that the control signal u(t ) is not a physical entity, but a error signal, expressed as a percentage of the reference angle
mathematical tool useful for controller design. range. In both cases in figure 18, the upper plot shows
the time-series of the chosen reference and the resulting
4.3. Experimental results measurement, and in each case, the bottom plot shows the
corresponding control signal u(t ). Note that the results in
The suitability of the proposed approach is demonstrated case 1(a) are in close agreement with what is predicted
through six different sets of experiments. The first set is by the estimated complementary function T̂ in figure 17,
shown in figure 18 and table 2. In this experiment, the in terms of performance. This consistency is explained by
system is required to follow sinusoidal references with a fixed the fact that for a sinusoidal reference with an amplitude of

11
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

Table 2. Performance measured as the experimental standard deviation (ESD) of the control error, expressed as a percentage of the angle
range 100% STD(Control error)
angle range
.

Signal amplitude
Low High
Experimental case 2.5◦ 5◦ 7.5◦ 10◦ (figure 23(a)) (figure 23(b))

Case 1 (0.4 Hz sine) 7.2% 7.3% 7.1% 7.4% N/A N/A

Case 2 (0.4 Hz sine + disturb.) 11.5% 10.6% 10.4% 8.6% N/A N/A
Case 3 (0.8 Hz sine) 11.7% 10.7% 14.2% 16.0% N/A N/A
Case 4 (1.0 Hz sine) 14.5% 17.2% 17.7% 20.2% N/A N/A
Case 5 (Walking gait) N/A N/A N/A N/A 12.9% 17.2%
Case 6 (Square wave) 18.5% 20.9% 24.2% 26.4% N/A N/A

2.5◦ , the behavior of the system can be considered LTI, as (a) Control Signal u(t) for Sinusoidal Reference (0.4 Hz; 2.5 )
uK (t ) rarely saturates and u(t ) ≈ uK (t ), as evidenced from the 1

bottom plot in figure 18(a). In the case shown in figure 18(b),

Input [-1:1]
0.5
it is clear that the measured signal y(t ) closely follows the
reference yd (t ). However, as shown in table 2, as the amplitude 0

of the reference increases, a progressive deterioration in -0.5

performance can be observed, caused by the saturation of the
signal uK (t ), according to the law in (8). The saturation of uK (t ) -1
15 15.2 15.4 15.6 15.8 16 16.2 16.4 16.6 16.8 17
reflects the physical inability of the robotic orthotic system to Time (sec)
contract and expand at the speeds required by the linear control (b) Section of the Valve State v
law defined by K(z). The notion that the saturation of uK (t ) is [0 1]’

caused by a speed constraint follows from simply noticing that,

for a fixed frequency, the required contraction and expansion
State v

[0 0]’
speeds of the artificial muscles are directly proportional to the
amplitude of the reference yd (t ), i.e.,
sd (t ) = AgAd ω cos(ωt ), (11) [1 0]’
15 15.2 15.4 15.6 15.8 16 16.2 16.4 16.6 16.8 17
where sd (t ) is the desired speed of contraction and expansion, Time (sec)
ω is a fixed frequency, Ag ∈ R++ is a geometric constant, and
Figure 19. Upper plot: short section of the time-series of u(t ) from
Ad is the amplitude of the angular reference. Note that the
the experimental case 1(a). Bottom plot: resulting valve’s state v(τ ),
speed of a McKibben actuator depends on the actuator itself, corresponding to the section of u(t ) in the upper plot.
but also on the actuation pneumatic valves, which in this case
are the main cause of speed limitation. effect of the force disturbance acting on the system. Recall that
The last plots associated with case 1 are shown figure 19, here, performance is measured using the ESD of the control
which is an example of the mapping from u to v. An interesting error signal as a percentage of the reference angle range.
thing to notice in figure 19 is that, as described in definition 1, Interestingly, in case 2, as can be observed in figure 20, as the
if the signal u does not hit the saturation limits, the idealized percentage performance in table 2 improves with amplitude,
valve’s state v changes during a time cycle. In other words, in the absolute control error increases with amplitude. The first
the system considered here, saturation means that the idealized phenomenon is explained by the saturation of uK (t ) caused
valve’s state v stays constant over the whole time cycle. by the speed constraint of the actuators, already discussed
In the second set of experiments, the objective is to test above. The second phenomenon can be explained by noticing
the system when subjected to a significant constant force that the most notable effect produced by the force disturbance
disturbance. In this case, a large weight (500 g) counteracting is a high frequency oscillation over the measured signal yd (t ),
the dorsiflexion of the subject’s leg is hanged at the distal which has an amplitude that stays approximately constant as
end of the foot, which experimentally simulates the case the reference angle’s amplitude is varied. The existence of
of patients with highly contracted gastrocnemius muscles, a this high frequency content is the direct result of the orthotic
typical characteristic of a drop foot symptom in CP. The chosen system’s slow response speed. Overall, the capability of the
references are the same used in the first set of experiments. In closed-loop system to reject disturbances, according to what
both cases shown in figure 20, it is clear that the measured is predicted by the estimate of Ŝo in figure 17, is demonstrated
signal y(t ) closely follows the reference yd (t ), and overall, in figure 20(a). This indicates that the chosen control strategy
the achieved performance is good enough to suggest that the is adequate for this kind of system and that the limitations in
robotic orthotic system could be employed as a rehabilitation performance reflect physical limitations of the hardware. The
device. Note that in this case, as shown in table 2, for performances of all the variants of case 2 are shown in table 2.
each amplitude considered (2.5◦ , 5◦ , 7.5◦ , 10◦ ), the resulting Figures 21 and 22 show the experimental sets 3 and 4,
performance is degraded with respect to case 1, due to the which are aimed at testing how well the system performs at

12
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

(a) Sinusoidal Ref. (0.4 Hz; 2.5 ) and Const. Disturbance (a) Sinusoidal Reference (Freq. 0.4 Hz; Amp. 2.5 )
5 5
Angle (deg)

Angle (deg)
2.5 2.5

0 0

-2.5 Reference yd(t) -2.5 Reference yd(t)

Measured Angle y(t) Measured Angle y(t)
-5 -5
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30
Time (sec) Time (sec)
Signal u(t) for Sinusoidal Ref. (0.4 Hz; 2.5 ) and Const. Disturbance Signal u(t) for Sinusoidal Reference (Freq. 0.4 Hz; Amp. 2.5 )
1 1

Input [-1:1]
0.5
Input [-1:1]

0.5

0 0

-0.5 -0.5

-1 -1
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30

Time (sec) Time (sec)

(b)
Sinusoidal Ref. (0.4 Hz; 10 ) and Const. Disturbance (b) Sinusoidal Reference (Freq. 0.4 Hz; Amp. 10 )
12.5 12.5
10 10
7.5 7.5
Angle (deg)
Angle (deg)

5 5
2.5 2.5
0 0
-2.5 -2.5
-5 Reference yd(t) -5 Reference yd(t)
-7.5 -7.5
-10 Measured Angle y(t) -10 Measured Angle y(t)
-12.5 -12.5
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30

Time (sec) Time (sec)

Signal u(t) for Sinusoidal Ref. (0.4 Hz; 10 ) and Const. Disturbance Signal u(t) for Sinusoidal Reference (Freq. 0.4 Hz; Amp. 10 )
1 1
Input [-1:1]

0.5
Input [-1:1]

0.5

0 0

-0.5 -0.5

-1 -1
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30
Time (sec) Time (sec)

Figure 20. Second set of experiments. In both experimental cases, Figure 21. Third set of experiments. In both experimental cases,
(a) and (b), the system is subjected to the force constant disturbance. (a) and (b), upper plot shows the time-series of the reference signal
In both cases, (a) and (b), upper plot shows the time-series of the yd (t ) and the measured angular position y(t ), the controlled
reference signal yd (t ) and the measured angular position y(t ), the variable. In both cases, the bottom plot shows the time-series of the
controlled variable. In both cases, the bottom plot shows the corresponding signal u(t ). In (a), the reference is a sinusoid with
time-series of the corresponding signal u(t ). In case (a), the amplitude 2.5◦ and frequency 0.8 Hz. In (b), the reference is a
reference is a sinusoid with amplitude 2.5◦ and frequency 0.4 Hz. In sinusoid with amplitude 10◦ and frequency 0.8 Hz.
case (b), the reference is a sinusoid with amplitude 10◦ and
frequency 0.4 Hz.
in (11). The resulting performances of all the variants of cases
3 and 4 are also shown in table 2.
higher required speeds, demonstrate the controller capabilities Figure 23 shows the experimental set 5. In case 5(a),
and expose situations where the limitations of the hardware the reference is an experimentally-recorded amplitude-scaled
design come into play. In case 3(a) and (b), the references are walk signal. As expected from the experimental cases already
sinusoids with a fixed frequency of 0.8 Hz, with amplitudes shown and from figure 17, it is clear that the measured
signal y(t ) closely tracks the reference yd (t ). However,
2.5◦ and 10◦ , respectively. In case 4(a) and (b), the references
despite the fact that the amplitude of the desired output yd (t )
are sinusoids with a fixed frequency of 1.0 Hz, with amplitudes
has a relatively small amplitude, the signal uK (t ) saturates
2.5◦ and 10◦ , respectively. The results in figures 21(a) and repeatedly. This phenomenon is due to the higher frequency
22(a) show that the proposed control strategy is adequate, content, in comparison to the signals considered in the
as predicted by Ŝo and T̂ . However, figures 21(b) and 22(b) previously presented cases, in this new reference. References
unequivocally demonstrate that the system is subjected to a with higher frequency content imply higher requirements of
hard physical speed constraint. This follows from noticing contraction and expansion speeds from the artificial muscles.
that the desired speed of contraction and expansion sd (t ) is As expected, the saturation phenomenon becomes worse when
directly proportional to the reference frequency ω, as shown the walking reference signal is scaled to a magnitude similar

13
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

(a) Sinusoidal Reference (Freq. 1.0 Hz; Amp. 2.5 ) (a) Amplitude-Scaled Walk Reference
5 7.5

5
Angle (deg)

Angle (deg)
2.5
2.5

0 0

-2.5
-2.5 Reference yd(t) Reference yd(t)
-5
Measured Angle y(t) Measured Angle y(t)
-5 -7.5
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30

Time (sec) Time (sec)

Signal u(t) for Sinusoidal Reference (Freq. 1.0 Hz; Amp. 2.5 ) Signal u(t) for Amplitude-Scaled Walk Reference
1
1

Input [-1:1]
0.5
Input [-1:1]

0.5

0 0

-0.5 -0.5

-1 -1
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30

Time (sec) Time (sec)

(b) Sinusoidal Reference (Freq. 1.0 Hz; Amp. 10 ) Amplitude-Scaled Walk Reference
12.5
(b) 15
12.5
10
10
Angle (deg)

7.5 Angle (deg) 7.5

5 5
2.5 2.5
0 0
-2.5 -2.5
-5
-5 Reference yd(t) Reference yd(t)
-7.5
-7.5 -10
-10 Measured Angle y(t) -12.5 Measured Angle y(t)
-12.5 -15
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30

Time (sec) Time (sec)

Signal u(t) for Sinusoidal Reference (Freq. 1.0 Hz; Amp. 10 ) Signal u(t) for Amplitude-Scaled Walk Reference
1 1
Input [-1:1]

0.5 0.5
Input [-1:1]

0 0

-0.5 -0.5

-1 -1
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30

Time (sec) Time (sec)

Figure 22. Fourth set of experiments. In both experimental cases, Figure 23. Fifth set of experiments. In both experimental cases,
(a) and (b), upper plot shows the time-series of the reference signal (a) and (b), upper plot shows the time-series of the reference signal
yd (t ) and the measured angular position y(t ), the controlled yd (t ) and the measured angular position y(t ), the controlled
variable. In both cases, the bottom plot shows the time-series of the variable. In both cases, the bottom plot shows the time-series of the
corresponding signal u(t ). In (a), the reference is a sinusoid with corresponding signal u(t ). In case (a), the reference is an
amplitude 2.5◦ and frequency 1.0 Hz. In (b), the reference is a amplitude-scaled walking signal ranging from −1◦ to 5◦ ,
sinusoid with amplitude 10◦ and frequency 1.0 Hz. approximately. In case (b), the reference is an amplitude-scaled
walking signal ranging from −10◦ to 12.5◦ , approximately.
to those in the technical literature showing normal walking
patterns [55], as shown in figure 23(b). Despite the saturation
tracking constant signals, which adds evidence to the notion
issue, the experiments in figure 23 present compelling evidence
that the proposed robotic system has a substantial potential
on the robotic orthotic system capabilities to be employed in
as a rehabilitation device. Finally, notice that the saturation
rehabilitation and on the potential capabilities as a walking
phenomenon appears in both cases, more dramatically
assistive device. The resulting performances of case 4 are
underlined in figure 24(b). The resulting performances of all
shown in table 2.
the variants of case 6 are shown in table 2.
Figure 24 shows the experimental set 6. In case 6(a),
the reference is a 0.1 Hz-square-wave with an amplitude of
2.5◦ . In case 6(b), the reference is a 0.1 Hz-square-wave 5. Discussion and future work
with an amplitude of 10◦ . These cases clearly show the
capabilities and limitations of the robotic system. From both The main contribution of this work is the biologically inspired
cases it follows that the maximum angular speed achievable design and development of a soft wearable robot. The
by the system is 14◦ s−1 , approximately. Also, as predicted by following features make the proposed system novel compared
figure 17, it follows that the control strategy is adequate for to prior wearable assistive devices.

14
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

(a) Square-Wave Reference (0.1 Hz; 2.5 ) rehabilitation without limiting 3D motion of the foot. The
5 performance of the device when applying a feedback controller
was also demonstrated. The current prototype demonstrated
Angle (deg)

2.5
full integration of sensing, actuation, and power to create a
0 nearly untethered system.
-2.5 Reference yd(t) While the system consumes very little electric power and
Measured Angle y(t) could run on batteries for several hours, the device still relies
-5
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 on an air source connection for the pneumatic muscles. Thus,
Time (sec) one area of on-going work is the investigation of solutions
Signal u(t) for Square-Wave Reference (0.1 Hz; 2.5 ) that allow complete untethered operation. Available options
1 include portable air compressors and compressed air canisters.
For the long-term success of such orthotic devices, we are
Input [-1:1]

0.5
also investigating relevant clinical requirements and potential
0
control strategies that would work seamlessly with the user’s
-0.5 motion. The goal is to achieve a fully untethered wearable
-1
system to provide a new level of mobility and active assistance.
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30 This will open a rich space for future rehabilitation techniques
Time (sec)
both inside and outside of the clinic.
(b)
Square-Wave Reference (0.1 Hz; 10 ) While the system goals for rehabilitation—seated ankle
12.5
10
motions—were achieved, the design components could be
adapted for increased performance. From an actuation
Angle (deg)

7.5
5
2.5 perspective, for example, we could use higher bandwidth
0
-2.5 actuators and valves. From a control perspective, in order
-5 Reference yd(t)
-7.5 to track angular positions that result from fast walking
-10 Measured Angle y(t) and running, more sophisticated algorithms are required. In
-12.5
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30
Time (sec) particular, it is possible to represent trajectories similar to the
one in figure 23 by linearly combining families of sinusoidal
Signal u(t) for Square-Wave Reference (0.1 Hz; 10 ) functions. Then, algorithms specialized in tracking families
1
of sinusoidal signals, like multiple-period repetitive control
[57], can be used to synthesize feedback controllers. Also, it
Input [-1:1]

0.5

0
is possible to use adaptive schemes like the one in [58].
We presented a methodology for synthesizing controllers
-0.5
for a biologically inspired pneumatically-actuated soft orthotic
-1 device. The proposed approach relied on system identification
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25 27.5 30
Time (sec)
and classical LTI design tools. This is surprising and
useful, considering that the system to be controlled was
Figure 24. Sixth set of experiments. In both experimental cases, highly complex, in which electronic, mechanical, and
(a) and (b), upper plot shows the time-series of the reference signal pneumatic components interact with a human body. Based
yd (t ) and the measured angular position y(t ), the controlled on experimental data, we presented evidence that by proper
variable. In both cases, the bottom plot shows the time-series of the
corresponding signal u(t ). In case (a), the reference is a 0.1 choice of the inputs and outputs, classical and modern
Hz-square-wave with an amplitude of 2.5◦ . In case (b), the reference system-theoretic tools can be implemented for designing high-
is a 0.1 Hz-square-wave with an amplitude of 10◦ . performance controllers for complex systems.
In this paper, the prototype was tested for seated motions,
(i) Bio-inspired actuation system. While most of previous which have importance for rehabilitation, as a first proof of
ankle–foot orthotics regard an ankle joint as a simple principle. From this initial testing, we can develop a controller
mechanical pin joint, the proposed device mimics a that can assist walking gaits. However, gait assistance is
biological muscle—tendon–ligament architecture with beyond the scope of our current work. Also, while force
multiple muscles to create various natural ankle motions. control may be more appropriate for assisting and supporting
(ii) Soft strain sensors. Hyperelastic strain sensors were dynamic walking—our ultimate goal—the position control
developed and implemented to measure joint angles. This result demonstrates the feasibility of our current prototype
type of soft sensors makes the device easily wearable and for use in ankle rehabilitation activities that often consist
conformable to complicated 3D body shapes. of various types of seated motions, such as those shown in
(iii) Completely soft structure. The proposed system does not [59, 60]. These types of seated motions can be easily recorded
contain any rigid frame structures that are easily seen in and replayed using our current device. Nevertheless, force
other active orthoses, resulting in no constraints on the control with dynamic walking should be investigated as the
natural degrees of freedom of the ankle joint. device is upgraded with embedded force sensors in the future.
The current design showed that such an orthotic device The subject fully relaxed the ankle while neither
has the potential to provide active assistance for ankle intentionally assisting nor resisting the controller’s actuation

15
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

during the control experiments. Although slight unintentional References

muscle (or body) reaction to the actuation was reported by
the subject during the experiments, it was not strong enough [1] Winters T F, Gage J R and Hicks R 1987 Gait patterns in
to disrupt the controller. However, this muscle reaction must spastic hemiplegia in children and young adults J. Bone
be taken into account as an internal disturbance for further Joint Surg. 69 437–41
[2] Park Y-L, Chen B, Young D, Stirling L, Wood R J, Goldfield E
improvement of the control algorithm. An electromyography and Nagpal R 2011 Bio-inspired active soft orthotic device
is a potential tool to measure the activation of the wearer’s for ankle foot pathologies Proc. IEEE/RSJ Int. Conf. on
biological muscles, which could be treated as disturbance in Intelligent Robots and Systems (San Francisco, CA) pp
the control loop. 4488–95
Although the first set of experimental data presented in this [3] Blanton S, Grissom S P and Riolo L 2002 Use of a static
paper focuses only on sagittal motions and SISO control, the adjustable ankle-foot orthosis following tibial nerve block
to reduce plantar-flexion contracture in an individual with
current device provides future opportunities for more complex brain injury Phys. Ther. 82 1087–97
control strategies. For example, a cooperative controller could [4] Mavroidis C et al 2011 Patient specific ankle-foot orthoses
be designed to achieve more natural ankle motions when using rapid prototyping J. Neuroeng. Rehabil. 8 1
controlling individual muscles. Another possible strategy is to [5] Miller F 2005 Cerebral Palsy 1st edn (Berlin: Springer)
implement a supervised learning algorithm to allow the device [6] Yamamoto S, Ebina M, Iwasaki M, Kubo S, Kawai H
to assist the user more dynamically without overpowering and Kayashi T 1993 Comparative study of mechanical
characteristics of plastic AFOs J. Prosthet. Orthot. 5 59–64
the natural muscles. This would require the system to learn [7] Appell H-J 1990 Muscular atrophy following immobilisation:
the user’s motion patterns and calculate correct timings to a review Sports Med. 10 42–58
provide timely actuation in order to seamlessly assist the user’s [8] Geboers J F, van Tuijl J H, Seelen H A and Drost M R 2000
intended motions. Effect of immobilization on ankle dorsiflextion strength
Finally, future work will cover various areas. For testing, Scand. J. Rehabil. Med. 32 66–71
we plan to perform a clinical study with the system using a [9] Geboers J F, Drost M R, Spaans F, Kuipers H and Seelen H A
2002 Immediate and long-term effects of ankle-foot
clinically appropriate population. For design improvement,
orthosis on muscle activity during walking: a randomized
we are currently under development of a more generalized study of patients with unilateral foot drop Arch. Phys. Med.
form, as shown in [31], to address the issue of individual- Rehabil. 83 240–5
dependence. The new design with modularized muscle units [10] Bruelmeier M, Dietz V, Leenders K L, Roelcke U, Missimer J
that are lightweight and compact will provide the capability and Curt A 1998 How does the human brain deal with a
of resizing and reconfiguring, which will allow us to easily spinal cord injury? Eur. J. Neurosci. 10 3918–22
[11] Green J B, Sora E, Bialy Y, Ricamato A and Thatcher R W
quantify the manufacturing and characterization cost and the
1998 Cortical sensorimotor reorganization after spinal cord
applicability. This new design is also expected to solve several injury: an electroencephalographic study Neurology
potential problems of the current prototype, such as trans- 50 1115–21
mission inefficiency, hardware misalignment, and kinematic [12] Herr H M and Wilkenfeld A 2003 User-adaptive control of a
constraints. Optimal power control and increased safety is magnetorheological prosthetic knee Ind. Robot. 30 42–55
another area that we will investigate for a wide range of users. [13] Lambrecht B G A and Kazerooni H 2009 Design of a
semi-active knee prosthesis Proc. IEEE Int. Conf. on
Robotics and Automation (Kobe, Japan) pp 639–45
6. Conclusions
[14] Krebs H, Hogan N, Durfee W and Herr H 2005 Rehabilitation
A bio-inspired soft wearable robotic device was designed and robotics, orthotics, and prosthetics Textbook of Neural
Repair and Rehabilitation ed M E Selzer, S Clarke,
prototyped using soft actuators and sensors. The prototype L G Cohen, P W Duncan and F H Gage (Cambridge:
showed the capability of assisting sagittal ankle motions Cambridge University Press)
with actuation, providing a range of 27◦ for our test subject. [15] Blaya J A and Herr H 2004 Adaptive control of a
The prototype also demonstrated the capability for feedback variable-impedance ankle-foot orthosis to assist drop-foot
control of ankle joint angle using an identified LTI model of the gait IEEE Trans. Neural Syst. Rehabil. Eng. 12 24–31
system. We believe that this bio-inspired design, utilizing soft [16] Banala S K, Kim S H, Agrawal S K and Scholz J P 2009
Robot assisted gait training with active leg exoskeleton
sensors, actuators, and structures, will open a rich space for (ALEX) IEEE Trans. Neural Syst. Rehabil. Eng. 17 2–8
future wearable robot and soft robot technologies for human [17] Patritti B L, Straudi S, Deming L C, Benedetti M G,
rehabilitation and assistance. Nimec D L and Bonato P 2010 Robotic gait training in an
adult with cerebral palsy: a case report PMR 2 71–5
Acknowledgments [18] Roy A, Kerbs I, Williams D J, Bever C T, Forrester L W,
Macko R M and Hogan N 2009 Robot-aided
neurorehabilitation: a novel robot for ankle rehabilitation
This work was supported by the Wyss Institute for Biologically
IEEE Trans. Robot. 25 569–82
Inspired Engineering and National Science Foundation (NSF) [19] Herr H M and Kornbluh R D 2004 New horizons for orthotic
grant CNS 0932015. Any opinions, figures, and conclusions and prosthetic technology: artificial muscle for ambulation
or recommendations expressed in this material are those of the Proc. SPIE 5385 1–9
authors and do not necessarily reflect the views of the NSF. [20] Ferris D P, Czerniecki J M and Hannaford B 2005 An
The authors would like to thank James Niemi for ankle-foot orthosis powered by artificial pneumatic muscles
J. Appl. Biomech. 21 189–97
his support and feedback in this research. We also thank
[21] Gordon K E, Sawicki G S and Ferris D P 2006 Mechanical
Dr Chandana Paul for her suggestions and technical inputs performance of artificial pneumatic muscles to power an
to this project. ankle-foot orthosis J. Biomech. 39 1832–41

16
Bioinspir. Biomim. 9 (2014) 016007 Y-L Park et al

[22] Shorter K A, Kogler G F, Loth E, Durfee W K [39] Chou C-P and Hannaford B 1996 Measurement and modeling
and Hsiao-Wecksler E T 2011 A portable powered of McKibben pneumatic artificial muscles IEEE Trans.
ankle-foot orthosis for rehabilitation J. Rehabil. Res. Dev. Robot. Autom. 12 90–102
48 459–72 [40] Tondu B and Lopez P 2000 Modeling and control of Mckibben
[23] Hollander K W, Ilg R, Sugar T G and Herring D 2006 An artificial muscle robot actuators IEEE Control Syst. Mag.
efficient robotic tendon for gait assistance J. Biomech. Eng. 20 15–38
128 788–92 [41] Belforte G, Eula G, Appendino S and Sirolli S 2011 Pneumatic
[24] Svensson W and Holmberg U 2008 Ankle-foot-orthosis interactive gait rehabilitation orthosis: design and
control in inclinations and stairs Proc. IEEE Int. Conf. on preliminary testing Proc. Inst. Mech. Eng. H 225 158–69
Robotics Automation and Mechatronics (Chengdu, China) [42] Wolbrecht E T, Chan V, Reinkensmeyer D J and Bobrow J E
pp 301–6 2008 Optimizing compliant, model-based robotic assistance
[25] Nikitczuk J, Weinberg B, Canavan P K and Mavroidis C 2010 to promote neurorehabilitation IEEE Trans. Neural Syst.
Active knee rehabilitation orthotic device with variable Rehabil. Eng. 16 286–97
damping characteristics implemented via an [43] Balasubramanian K and Rattan K S 2003 Fuzzy logic control
electrorheological fluid IEEE/ASME Trans. Mechatronics of a pneumatic muscle system using a linearizing control
15 952–60 scheme Proc. 22nd Int. Conf. North American Fuzzy
Information Processing Society (Chicago, IL) pp 432–6
[26] Stirling L, Yu C, Miller J, Wood R J, Goldfield E and Nagpal R
[44] Balasubramanian K and Rattan K S 2005 Trajectory tracking
2011 Applicability of shape memory alloy wire for an
active, soft orthotic J. Mater. Eng. Perform. 20 658–62 control of a pneumatic muscle system using fuzzy logic
Proc. Annual Meeting North American Fuzzy Information
[27] Galiana I, Hammond F L, Howe R D and Popovic M B 2012 Processing Society (Ann Arbor, MI) pp 472–7
Wearable soft robotic device for post-stroke shoulder [45] Thanh T D C and Ahn K K 2006 Nonlinear PID control to
rehabilitation: Identifying misalignments Proc. IEEE/RSJ improve the control performance of 2 axes pneumatic
Int. Conf. on Intelligent Robots and Systems (Vilamoura, artificial muscle manipulator using neural network
Portugal) pp 317–22 Mechatronics 16 577–87
[28] Ueda J, Ming D, Krishnamoorthy V, Shinohara M [46] Bertetto A M and Ruggiu M 2004 Characterization and
and Ogasawara T 2010 Individual muscle control using an modeling of air muscles Mec. Res. Commun. 31 185–94
exoskeleton robot for muscle function testing IEEE Trans. [47] Reynolds D B, Repperger D W, Phillips C A and Bandry G
Neural Syst. Rehabil. Eng. 18 399–50 2003 Modeling the dynamic characteristics of pneumatic
[29] Kim S, Spenko M, Trujillo S, Heyneman B, Santos D muscle Ann. Biomed. Eng. 31 310–7
and Cutkosky M R 2008 Smooth vertical surface climbing [48] Au S K, Weber J and Herr H 2007 Biomechanical design of a
with directional adhesion IEEE Trans. Robot. 24 65–74 powered ankle–foot prosthesis Proc. IEEE 10th Int. Conf.
[30] Park Y-L, Majidi C, Kramer R, Berard P and Wood R J 2010 on Rehabilitation Robotics pp 298–303
Hyperelastic pressure sensing with a liquid-embedded [49] Au S K, Weber J and Herr H 2009 Powered ankle–foot
elastomer J. Micromech. Microeng. 20 125029 prosthesis improves walking metabolic economy IEEE
[31] Park Y-L, Chen B, Majidi C, Wood R J, Nagpal R Trans. Robot. 25 51–66
and Goldfield E 2012 Active modular elastomer sleeve for [50] Eilenberg M F, Geyer H and Herr H 2010 Control of a
soft wearable assistance robots Proc. IEEE/RSJ Int. Conf. powered ankle-foot prosthesis based on a neuromuscular
on Intelligent Robots and Systems (Vilamoura, Portugal) pp model IEEE Trans. Neural Syst. Rehabil. Eng. 18 164–73
1595–602 [51] Cho Y M, Xu G and Kailath T 1994 Fast identification of
[32] Park Y-L, Chen B and Wood R J 2012 Design and state-space models via exploitation of displacement
manufacturing of soft artificial skin using embedded structure IEEE Trans. Autom. Control 39 2004–17
microchannels and liquid conductors IEEE Sensors J. [52] Pérez Arancibia N O, Gibson S and Tsao T-C 2004 Adaptive
12 2711–8 control of MEMS mirrors for beam steering Proc. ASME
[33] Slabaugh G 1999 Computing Euler angles from a rotation Int. Mechanical Engineering Congress and Exposition
matrix Technical Report City University London (Anaheim, CA) p 60256
www.soi.city.ac.uk/∼sbbh653/publications/euler.pdf [53] Ljung L 1999 System Identification (Upper Saddle River, NJ:
[34] Mengüc Y, Park Y-L, Martinez-Villalpando E, Aubin P, Prentice-Hall)
Zisook M, Stirling L, Wood R J and Walsh C J 2013 Soft [54] Van Overschee P and De Moor B 1996 Subspace Identification
wearable motion sensing suit for lower limb biomechanics for Linear Systems (Norwell, MA: Kluwer)
measurements Proc. IEEE Int. Conf. on Robotics and [55] Rose J and Gamble J G 2006 Human Walking (Philadelphia,
Automation (Karlsruhe, Germany) pp 5289–96 PA: Lippincott Williams & Wilkins)
[56] Dullerud G E and Paganini F 2000 A Course in Robust Control
[35] Shin D, Sardellitti I, Park Y-L, Khatib O and Cutkosky M 2010
Theory (New York: Springer)
Design and control of a bio-inspired human-friendly robot
[57] Pérez-Arancibia N O, Tsao T-C and Gibson J S 2010 A new
Int. J. Robot. Res. 29 571–84
method for synthesizing multiple-period adaptive-repetitive
[36] Jien S, Hirai S, Ogawa Y, Ito M and Honda K 2009 Pressure controllers and its application to the control of hard disk
control valve for Mckibben artificial muscle actuators with drives Automatica 46 1186–95
miniaturized unconstrained pneumatic on/off valves Proc. [58] Sacks A, Bodson M and Khosla P 1996 Experimental results
IEEE/ASME Int. Conf. on Advanced Intelligent of adaptive periodic disturbance cancellation in a high
Mechatronics (Singapore) pp 1383–8 performance magnetic disk drive ASME J. Dyn. Syst. Meas.
[37] Kerscher T, Zoellner J M, Dillmann R, Stella A Control 118 416–24
and Caporaletti G 2006 Model and control of joints driven [59] Connelly D M and Vandervoort A A 2000 Effects of isokinetic
by fluidic muscles with the help of advanced automatic strength training on concentric and eccentric torque
algorithm generation software Climbing Walking Robots development in the ankle dorsiflexors of older adults
4 245–52 J. Gerontol. A Biol. Sci. Med. Sci. 55 B465–72
[38] Wolbrecht E T, Leavitt J, Reinkensmeyer D J and Bobrow J E [60] Fowler E G, Ho T W, Nwigwe A I and Dorey F J 2001 The
2006 Control of a pneumatic orthosis for upper extremity effect of quadriceps femoris muscle strengthening exercises
stroke rehabilitation Proc. 28th IEEE EMBS Annual Int. on spasticity in children with cerebral palsy Phys. Ther.
Conf. (New York, NY) pp 2687–93 81 1215–23

17