Professional Documents
Culture Documents
8, AUGUST 2018
Abstract — For stroke survivors and many other people life by limiting both their psychological and physical capabil-
with upper-extremity impairment, daily life can be difficult ities. Treatments such as Graded Motor Imagery and mirror
without properly functioning arms. Some modern physical therapy have been previously discovered to benefit those
therapy exercises focus on rehabilitating people with these
troubles by correcting patients’ perceptions of their own with diminished motor function [2], [3]. These rehabilitative
body to eventually regain complete control and strength techniques focus on visual feedback for the user to promote the
over their arms again. Augmentative wearable robots, such regaining and restrengthening of the patient’s arm movements.
as the upper-extremity exoskeletons and exosuits, may be This mirror technique can provide therapy for improving a
able to assist in this endeavor. A common drawback in many stroke survivor’s motor system. A technique that creates this
of these exoskeletons, however, is their inability to conform
to the natural flexibility of the human body without a rigid visual demonstration of movement through physical action can
base. We have built one such exosuit to address this chal- consequently achieve the same desired effect as traditional
lenge: Compliant Robotic Upper-extremity eXosuit (CRUX). manual therapy but with a higher degree of precision immune
This robot is a compliant, lightweight, multi-DoF, portable to a practitioner’s inherent human error due to the use of a
exosuit that affords its wearer the ability to augment them- robotic guide [4], [5]. Therapists can use these strategies as
selves in many unconventional settings (i.e. outside of a
clinic). These attributes are largely achieved by using a the primary mode for rehabilitation of a patient or they can
modified tensegrity design situated according to measured employ them in conjunction with other methods as well.
lines of minimal-extension, where a network of tension Wearable robots, such as soft exoskeletons (“exosuits”),
members provide a foundation to apply augmentative forces can promote this treatment through the augmentation and
via precisely placed power-lines. In this paper, we detail the subsequent retraining of the nervous system. The goals of
design process of CRUX, the report on CRUX’s prototypical
composition, and describe the mimetic control algorithm Graded Motor Imagery and mirror therapy stem from the
used. We also discuss the results of three studies that combination of moving the physical self along with reward
illustrate the efficacy of CRUX’s mimetic controller, CRUX’s signals in the brain representing that the action was properly
flexibility and compliance, and the metabolic cost reduction completed. An exosuit can theoretically meet this need by
when users exercise with assistance from CRUX as opposed augmenting movements that would otherwise be manipulated
to without. We conclude this paper with a summary of our
findings, potential use cases for this technology, and the by a physical therapist. However, a critical roadblock that
direction of future related work. impedes the design of many wearable robots, including these
potential exosuits, is creating a device which conforms to the
Index Terms — Exosuits, human-robot interaction,
medical robotics. nuances of the human body, specifically its hybrid soft-rigid
structure and consequential flexibility and compliance.
1534-4320 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
LESSARD et al.: SOFT EXOSUIT FOR FLEXIBLE UPPER-EXTREMITY REHABILITATION 1605
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
1606 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 26, NO. 8, AUGUST 2018
exercises. These designs and others like them have provided we examine one subset of potential CRUX users who may
a foundation upon which other upper-extremity exoskeletons benefit the most from the exosuit. Users who have single-arm
can be developed. For example, the wearable robots from impairment are likely to benefit from or even have already
Tsagarakis et al. [24] and Galiana et al. [25] provide upper undergone Graded Motor Imagery and mirror therapy. Both of
extremity augmentation through the use of compliant mate- these physical therapy techniques demonstrate an immediate
rials. Additionally, exosuits like that of Lessard et al. [26], use case for CRUX. The impaired users we recruited would
Simpson et al. [27], and O’Neill et al. [28] have minimized therefore be able to provide the most meaningful preliminary
the mass and increased the mobility of exosuits to decrease the commentary and feedback on the device given that CRUX was
functional overhead required for compliant augmentation in designed for bilateral mimicry (a necessary feature for both
the upper-extremity. Specifically, the design by Simpson et al. mirror therapy and Graded Motor Imagery). This demographic
features a set of straps that anchor a unilateral device against of users also typically possessed enough upper-extremity
the torso at the waistline and the chest line of the wearer. strength to manually override unwarranted or unsafe actuation
A fabric-based vest is used by ONeill et al. to anchor soft if necessary.
cruciate linkages to the actuator in the users armpit. Both of The primary focuses of CRUX necessary to address single-
these designs use pneumatic actuation to augment users. The arm impairment are its features that augment users without
exosuit by Lessard et al. however, uses a full neoprene jacket significantly sacrificing wearer mobility and flexibility. These
with additional tensile components as a tension-based anchor attributes allow users to wear CRUX, a textile-based robot,
on the user’s torso. Although these devices have provided more like clothing than a traditional exoskeleton or other
important advances towards the goal of improved upper- restrictive wearable robot. As a result, users can theoreti-
extremity therapy, they also highlight the need for further cally don and wear CRUX in “unconventional” settings and
development for new flexible and compliant designs that can environments, such as for highly dexterous and contorting
conform to the nonlinear human body while still providing tasks or simply when at-home. In situations like these, rigid
meaningful augmentation. designs may over-constrain the user, or potentially obstruct
To address this need, we have redesigned and upgraded motion that could be accomplished normally. Flexible robots
CRUX (the Compliant Robotic Upper-extremity eXosuit). We are also more appealing as wearable devices to users than
originally created CRUX for relatively simple compliant, light- rigid robots. Prior studies have shown that intuitive physical
weight augmentation of the arms [26]. Now we additionally interfaces and comfortable fit are significant factors in assistive
introduce the modifications to CRUX that have given it a more device usability and abandonment [29].
flexible structure and more robust control to achieve more To design an exosuit that achieves these goals, we performed
augmentable actions for users than before. three steps: motion capture to identify lines of minimal exten-
The remainder of this paper is divided into three sections: sion in the upper extremity, deriving a cable map from those
the System Design, the Results and Discussion, and the lines, and prototyping the exosuit.
Conclusion. In the System Design section, we illustrate the
methodology for designing CRUX which uses motion capture
and subsequent cable mapping that reflects the morphology A. Identifying the Lines of Minimal Extension
motion-captured user. We then introduce the mechanics that We define lines of minimal extension as those where the
afford CRUX its multi-DoF actuation for greater arm move- skin shears the least as the body moves. Previous literature
ment. Then we demonstrate an implemented controller that has discussed using lines of non-extension for the design
allows CRUX to mimic arm movement bilaterally, a tech- of counterpressure suits [30], [31]. We differentiate lines of
nique critical to the aforementioned Graded Motor Imagery minimal extension from lines of non-extension to account for
and mirror therapy. In the Results and Discussion section, variations we found between specific test subjects and their
we subsequently discuss the quantitative and qualitative results body types when identifying the lines.
obtained when a group of participants individually use CRUX To identify these lines, we first performed motion capture
to perform specific exercises and test the capabilities of the to quantify the surface morphology of the arms (Figure 4A).
exosuit. These tests cover three specific criteria: the efficacy We began by selecting a characteristic subject to serve as
of the on-board controller and algorithm, the flexibility and a template for CRUX. Although CRUX is primarily designed
compliance of the exosuit, and the metabolic impact of the to assist users with single-arm impairment, the characteristic
exosuit. In the final section of the paper, the Conclusion, subject we chose was unimpaired. This decision was made for
we summarize our design, results, and the impact they may three reasons: to allow CRUX to fit more users, to be safer,
have on future soft wearable robots. and to avoid exhausting our limited pool of impaired users.
Unimpaired users tended to be more flexible than their
impaired counterparts and for that reason, a test which uses
II. S YSTEM D ESIGN
a healthier subject for designing a particular cable map will
When designing CRUX, we identified our target demo- yield a design safer for others to use. Although future exosuits
graphic as users with single-arm impairment. Single-arm may be custom-fit and personalized to one person, we used a
impairment is a prevalent medical issue and exosuits can single exosuit for all of our tests. For that reason, we chose
potentially address this need. Additionally, by studying users to create a template produced by one, healthy individual for
with single-arm impairment, particularly stroke survivors, repeated use.
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
LESSARD et al.: SOFT EXOSUIT FOR FLEXIBLE UPPER-EXTREMITY REHABILITATION 1607
TABLE I
M OVEMENTS P ERFORMED D URING M OTION C APTURE
TO I DENTIFY L INES OF M INIMAL E XTENSION
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
1608 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 26, NO. 8, AUGUST 2018
B. Cable Map
From these aforementioned lines of minimal-extension,
we wanted to identify ideal routes for tension elements to pull
along the body. These tension routes would provide a specific
path for cables to adhere to and pull along when contracted
by a motor. Our priority was developing paths to guide cables
that augmented the largest and most influential muscles and
muscle groups of the upper-extremity (namely, those in the
shoulder, upper arm, and forearm).
To identify these muscles and muscle groups, we consulted
past studies on upper-extremity muscular strength in stroke-
survivors. Ikai and Fukunaga have observed that muscular
cross-sectional area accurately predicts muscle strength [32].
Further principal component analysis and cluster analysis by
Bohannon and Andrews have additionally shown the impact
of stroke-induced paresis on upper-extremity movement in the Fig. 5. Power-lines, composed of flexible aramid cable threaded through
upper arm and forearm [33]. Additionally, Motricity Index, segments of rigid Bowden housing, channel power from the motors
a measure of motor function in muscles, has been found to be to actuate the exosuit. Seven power-lines on each arm dictate user
augmentation by assisting the muscle groups highlighted in Figure 4B.
a valid indicator of paretic upper-extremity strength in stroke The combination of tension members (i.e. the aramid cables and the neo-
survivors [34]. prene base layer) and compression members (e.g. housing segments)
Using these findings, we then isolated relevant arm move- function similarly to tensegrity structures.
ments according to motion primitives: movements achievable
in one, unique degree-of-freedom. We specifically chose to
augment humeral rotation (in and out), elbow flexion and On top of the elastic tensegrity-inspired base layer, cables
extension, wrist pronation and supination, lateral shoulder are pulled along the described routes (Figure 4B), acting
raise, and forward shoulder raise and lower. We chose not to as tendons in our tension-based augmentation exosuit. These
explicitly augment lateral shoulder lowering as gravity already cables are guided along each arm by anchored segments of
provided enough force to achieve this motion when users were rigid Bowden cable housing. This combination of a cable
upright. and its associated housing is called a power-line (Figure 5).
One issue we noticed that hindered the mobility and pota- Since there are seven cables planned for each arm of the
bility of prior upper-extremity exosuit designs was the need exosuit, there are fourteen power-lines outlined during the
for a structural base. Augmentation of the upper-extremity is cable mapping process.
particularly challenging because unlike the lower-extremities Each power-line is at maximum length when the associ-
which have a single central bone (the pelvis) and the ground ated muscle(s) are completely relaxed (i.e. not contracted)
to support movement, the upper-extremities must rely on sup- and at minimum length when the associated muscle(s) are
porting augmentation without a clear base. We address this by completely contracted. Antagonistic power-lines, like their
constructing a tensegrity-inspired base. In traditional tensegrity muscles, are therefore inversely proportional in length. For
structures, compression elements exist as simple rods and example, a biceps power-line at maximum length implies a
tension elements exist as simple cables that attach two ends triceps power-line at minimum length since the biceps and
of two different rods. We diverged from traditional tensegrity triceps muscles are antagonistic to one another. To further
designs and instead used an approach that featured not just elucidate the above invariant, if we describe the expressed
cables along the identified tension routes, but an omni-present length of a power-line as:
tension element which unites them all by connecting to every
rigid element. This omni-present tension element is an elastic L = l + cx (1)
base which is pre-shaped according to human torso and upper- 0≤c≤1 (2)
extremities, to provide distribute applied forces throughout the
entire structure. For the same reason that tensegrity structures where L is the expressed length of a power-line, l is the
maintain their constitution when internal forces are applied, minimum length of the power-line (also written as the sum
the structural integrity of the exosuit as a whole is preserved of lengths of every tube serially composing the power-line),
during augmentation. An additional change in the design of c as the percentage of contraction, and x as the length of the
our exosuit from dogmatic tensegrity structures is the use of exposed cable when the power-line is at maximum length, then
non-rod rigid elements. Allowing diversity in the shape of the for any pair of antagonistic power-lines:
exosuit’s rigid elements prevents unnecessary limitations to
what shape the backplate, housing and anchors must conform. c1 + c2 = 1 (3)
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
LESSARD et al.: SOFT EXOSUIT FOR FLEXIBLE UPPER-EXTREMITY REHABILITATION 1609
These cable mappings provide the layout and construction of shown in Figure 4C supports 5 of the 6 motors as well as the
the physical prototype subsequently created. battery and a 32 kB microcontroller. The sixth motor is located
on the shoulder of the augmented arm. Since our impaired
C. Prototype participants all had right-arm impairment, we mounted this
The parts with which we built CRUX were selected accord- sixth motor above the right shoulder. We found that the
ing to our hybrid soft-rigid design philosophy. Flexibility and movements augmented by this motor (i.e. pronation and
structural compliance were, therefore, achieved by construct- supination) required less anchorage than other movements
ing the exosuit out of mostly compliant materials with rigid augmented by the exosuit and so we placed this motor directly
elements strategically placed to interfere minimally. on the arm to prevent further crowding on the backplate and
1) Base Layer: The base layer of CRUX is neoprene and upper-back.
covers both arms as well as the chest and abdomen. This The battery is a three-cell 3500 mAh lithium polymer power
neoprene is 2 mm thick on the torso and 1 mm thick around source which provides between 11.1 − 12.6 V and up to 50
the arms. Since neoprene is spongy and elastic, users maintain A. This powers the equipped brushed DC micromotors, each
most of their mobility and the exosuit tends to conform to the of which has a stall torque of 88.3 Nm (125 oz-in) and a gear
user’s body shape. This conformability promotes a low-profile ratio of about 1000:1. This arrangement of motors focuses
for the exosuit as well. Both of these attributes are critical on augmenting the right arm of the user. A plunger (not
factors in an exosuit that must tout extreme compliance while visible in Figure 4) is attached to exosuit for use during semi-
simultaneously maintaining the rigidity necessary to anchor autonomous mimetic control.
on-board components and actuation forces. An additional joystick controller can be detached from this
2) Power-Lines: Each power-line consists of a braided- back plate for user manual override control of the exosuit.
aramid cable stretched along one of the tension routes. We When the joystick is enabled, the user can manually control
define the cable routes of CRUX as the physical paths of a two motors simultaneously. The joystick is programmable
power-line along a tension route identified during the motion to perform two arbitrary pairs of antagonistic movements
capture stage of our system design. The cables in CRUX (e.g. biceps/triceps movement and pronation/supination). The
are rated to 50 lbs-force (220 N), but other limitations in two movement pairs can be reprogrammed quickly using an
the exosuit prevent forces that large from being experienced external laptop, such as in the case of user testing for multiple
by the user regardless. There is never more than one cable types of arm movement. Whenever more than two movements
strung through any portion of housing as that would induce were required to be manually controlled by the exosuit,
unnecessary friction on the cables inside. For the entirety of the joystick-interface was replaced by a laptop connected to
the power-line, the cable is above the neoprene base layer, but the microcontroller via USB cable.
for some portions of the power-line, the cable is additionally The on-board microcontroller consists of an Arduino Pro
surrounded by Bowden cable housing adhered to the base mini development board with custom designed circuit board
layer via neoprene cement. Having Bowden cable housing and two stackable motor driver modules. On the circuit board,
along the entire power-line would inhibit the flexibility of the motor driver modules are connected via I2C. Additionally,
exosuit at the joints, where the majority of the morphological the microcontroller supports the input from external controllers
changes in the upper-extremity occur. To counteract this, (including the aforementioned joystick and plunger-button)
Bowden cable housing was placed only in segments. The and the IMU network.
regions of the power-line without Bowden cable housing are 4) IMU Network: CRUX’s IMU network is a collection of
therefore as flexible as the naked cable. The exact method for four wireless standalone IMU nodes where each node senses
determining the location and length of the segments was an using a 3DOF accelerometer, 3DOF gyroscope, and 3DOF
iterative process. For each identified cable route, we started magnetometer. The sensors on these IMU nodes are Bosch
with Bowden housing whose length reached from one distal Sensortec BNO055’s. The manufacturer guarantees that there
joint to the next joint proximally. After lightly adhering the is never more than 0.5◦ in alignment error between the internal
Bowden housing to the base layer, a user donned the prototype gyroscope and the labeled surface on any particular model.
and moved their arms according to the largest achievable range This error falls far below what would be significant for
of motion they could. Until the achievable range of motion our application. To avoid gimbal lock, each IMU internally
matched the user’s range of motion without the Bowden cable represents rotations mathematically as quaternions instead of
housing, the Bowden cable housing was slightly trimmed Euler angles and simply outputs the translated roll, pitch, and
and the process repeated for every segment of Bowden cable yaw. Because our particular use of these IMUs is limited to a
housing and for every cable route. The exact movements used narrow workspace (i.e. only what an arm can reach), we justify
to test the range of motion were biceps curls and shoulder that the error accumulated through the rounding of floating-
rotations. These exercises were selected since they illustrated point quaternion angles is insignificant.
movement according to the outlined motion primitives. The The IMU nodes operate on a wireless network so that
result of this iterative process is the longest lengths of Bowden they afford the system the capability to monitor (sniff) the
housing to provide necessary rigidity without interfering with transmitted data within the network without creating any
user movement. interruptions or lag.
3) Back Plate: On the back of the exosuit is a modular To minimize the difference between IMU orientation and
backplate constructed out of thin fiber glass. The backplate upper-extremity orientation, we kept the housing modules as
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
1610 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 26, NO. 8, AUGUST 2018
Q l = (θ1 , α1 , φ1 , θ2 , α2 , φ2 ) (4)
flat as possible. This low-profile keeps the IMUs close to the Q f = (θ3 , α3 , φ3 , θ4 , α4 , φ4 ) (5)
surface of the exosuit and consequently, the skin of the user.
The user’s leader arm is equipped with a plunger button and
The IMUs and their housing are placed on the exosuit such
accompanying display LED. When pressed, actuation of the
that the pose data for each arm can be distinguished uniquely.
follower arm is possible based on the state. The follower arm
Two IMU nodes are placed on each arm at the forearm (distal)
is moved to mirror the position of the leader arm, such that
and above the biceps (proximal). By placing one IMU on each
Q f becomes equivalent to Q l . However, actuation only occurs
forearm and one IMU on each upper arm, the orientation
when the threshold signal is high. The threshold is defined as:
of both halves of both arms can be continuously tracked
(Figure 6). As a result, knowing the poses of the pair of IMUs Q T h = |Q f − Q l | (6)
on an arm invariably allows one to mathematically reconstruct
the pose of both halves of the arm. Markings on the neoprene and the threshold signal will go high only when:
indicate both the position and the orientation expected of
L ≤ Qt h ≤ H (7)
each IMU attached. These on-suit markings correspond to
matching annotations on the IMU housing boxes. The specific In our implementation, L = 5◦ and H = 40◦ . This threshold
orientation of these IMUs was checked and corrected each allows for smoother and more calculated movements of the
time the exosuit was used for the sake of consistency. In slave arm. Although selecting threshold of 40◦ sacrifices finer
our user tests, we oriented the upper arm IMUs such that adjustments and movements during mimicry than a lower
the housing box was velcroed on the lateral side and the threshold would, we chose a high threshold to ensure wary
forearm IMUs such that the housing box was velcroed on the users felt safe and confident when using the exosuit. Concerns
dorsal side. A feedback acquisition system is used for the IMU for safety were one of the top reasons potential recruited
network via DWM1000 wireless transceiver modules. A prior participants for the user study were hesitant to try CRUX when
version of CRUX used IMUs in a similar fashion to this, but taught about the technology. As a result of the high value, this
did not incorporate any explicit attachment guides or housing threshold also protects against any unintended actuation from
that are necessary for the mimetic algorithm. the user or overloading of the motors. As a result, however,
5) Mimetic Control: Although an exosuit controlled manually the slave and master arm differ minimally in pose due to the
via joystick or laptop can express every possible augmentable 5◦ lower bound of the threshold signal.
arm movement theoretically afforded by the exosuit, passive The state machine consists of three states: off, ready, and
control can reduce mandatory input, making the exosuit easier mimic (Figure 7). The off state is the initial state and does
to operate. Operational difficulty is an important factor to con- not allow the user to activate the suit even if the button is
sider, especially since CRUX is designed to be used by those pressed. In the off state Q l is not within Q t h ; the threshold
who are disabled by upper-extremity impairment. Both the signal is low and so the suit cannot be activated. But once
lack of dexterity and lack of operational experience observed Q l is withing the threshold a transition to the ready state will
in many people with upper-limb impairment could prevent an occur. In the ready state the suit is waiting to be activated by
otherwise qualified user from reaping the rehabilitative benefits the user pressing the button. Once the button is pressed the
of CRUX. For this reason, a passively controlled mode for state will transition to the mimic state, where the LED will
CRUX can lower the necessary level of skill to operate the light up indicating to the user the suit is active and actuation
exosuit. of the slave arm will occur.
The passively controlled mode described in this paper is a This mimetic controller is built upon a simple closed-loop
mimetic algorithm. This algorithm synchronizes the position feedback system. Leading state is tracked by using the median
of one arm to the measured position of the other arm. of the previous five samples. Although a more complicated
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
LESSARD et al.: SOFT EXOSUIT FOR FLEXIBLE UPPER-EXTREMITY REHABILITATION 1611
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
1612 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 26, NO. 8, AUGUST 2018
Fig. 8. A user demonstrates a lateral arm raise with CRUX. The user’s leading arm (left) dictates the movement of their following arm (right). The
following arm, lagging behind the leading arm slightly, updates its position using micromotors installed on the backplate. IMUs on the exosuit provide
sensed orientation for CRUX’s controller.
Fig. 9. A user demonstrates a biceps curl with CRUX. The user’s leading arm (left) dictates the movement of their following arm (right). The following
arm, lagging behind the leading arm slightly, updates its position using micromotors installed on the backplate. IMUs on the exosuit provide sensed
orientation for CRUX’s controller.
Figure 10 illustrates the pitch of two of the four different design and reassured users. By t = 8s, the user has relaxed
IMUs on CRUX as a characteristic user learns to use CRUX their control slightly and can recognize the pulling sensation
for assisting their left arm while performing a lateral arm of CRUX in their upper-arm (proximally). After a few seconds
raise and lower (Figure 8). Each pair of plots (labeled T1 , of acclimating to the pull along CRUX’s power-lines, the user
T2 , and T3 ) illustrates a 17s time period during which the demonstrates very shaky augmentation, enough to demonstrate
user learned a critical lesson indicative of that user’s progress some degree of control in assisting their following arm.
in mastering CRUX. The time period of 17 seconds was At T3 , the user demonstrates intuitive mastery of CRUX.
chosen because it was the shortest time period that concisely When the user lifts their leading arm, the following arm
illustrated the highlighted lessons learned by that user. T1 , T2 , mimics the precise movement with only a short delay of
and T3 occur shortly after one another such that the end of T3 around 1s. The user demonstrates a relatively large range of
occurs roughly one minute after the start of T1 . operation (around 20◦ in proximal pitch and 60◦ in distal
At T1 , the user wears CRUX for the first time. Unsure pitch), which is necessary for many arm movements in mirror
of what sensation to expect when augmented, the user sub- therapy. We note one interesting observation in T3 at t = 12s:
consciously resists movement in their following arm as they the user’s leading arm rapidly moves (perhaps to scratch their
move their leading arm. Realizing their rigidity, the user at nose) but the following arm does not mimic the movement.
roughly t = 10s overcompensates and consciously forces their This discrepancy can be attributed to the built-in safety feature
following arm to perfectly mimic the leading arm. which prevents unintended movement of the following arm.
At T2 from t = 2s to t = 6s, the user manually overrides the In a more serious case with stronger motors and a weaker user,
exosuit to move their follower arm. The user then immediately this movement could accidentally damage the user, so CRUX
and manually re-corrects their follower arm at t = 6s to correctly intervenes at t = 12s to prevent the movement.
be within the predetermined hysteresis threshold for mimetic The data in Figure 10 demonstrate the proof-of-concept
control so they can resume mimetic movement assisted by that CRUX can augment mimetic movement in users who
the exosuit. The correction performed by the user has to have learned to operate the exosuit. After around a minute
be done manually as it is beneath the threshold to trigger of wearing CRUX, the user featured in Figure 10 learned to
mimetic movement (40◦ ). The rule followed by the exosuit operate CRUX at level sufficient to augment arm movements
is to apply a force via the motors to move the follower arm popular in mirror therapy. Formally measuring device mastery
until it reaches the pose of the leader arm (within 5◦ ). This is challenging, but we can confidently say that every user
rule, however, does not engage until there is a difference in tested was able to comfortably operate CRUX within minutes.
pose by ≤ 40◦ . As previously mentioned, although 40◦ is a Occasionally, users would get momentarily stuck, but after
large threshold, it further ensures safety in this prototypical quick instruction from a study proctor, were able to resume
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
LESSARD et al.: SOFT EXOSUIT FOR FLEXIBLE UPPER-EXTREMITY REHABILITATION 1613
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
1614 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 26, NO. 8, AUGUST 2018
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
LESSARD et al.: SOFT EXOSUIT FOR FLEXIBLE UPPER-EXTREMITY REHABILITATION 1615
TABLE III
M EDIAN H EART R ATE P ERCENT I NCREASES
A FTER E XERCISING
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
1616 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 26, NO. 8, AUGUST 2018
exosuit assisted them with the assigned tasks. The remaining system. The difficulty we foresee in these potential designs is
2 users said that they “Neither Agree nor Disagree” that the accomplishing this without over-encumbering the user. These
exosuit assisted their assigned movements. No user said that designs must also maintain proper balance when actuating.
they would “Disagree” or “Strongly Disagree” that the exosuit Any applied imbalance could contort the user dangerously
helped them. Of the 7 users who had previously completed with enough force.
some physical therapy, all 7 agreed or strongly agreed that such Additionally, the control of exosuits requires further study
a device would have been useful during their therapy. Multiple and examination. In particular, the development of additional
users commented that they did not feel the exosuit weighing control interfaces, the optimization of the algorithms behind
them down during the tests. The most common complaints them, and the user testing to support them all require further
when using the exosuit were confusion when operating the investigation. One particular autonomous control method, pro-
joystick while simultaneously exercising (during the metabolic prioception, can yield an exosuit which interprets and reacts
test) and overheating while moving around in the exosuit. to user intention. This problem is difficult to solve, however,
These data illustrate that CRUX can satisfy the need because people tend to have such diverse kinematics and
required of an exosuit for Graded Motor Imagery and mirror redundant workspaces that the decision to preemptively actuate
therapy: the ability to mechanically assist the user and provide an exosuit according to a predicted motion is marred with a
positive reinforcement. high likelihood of actively antagonizing a user’s movement.
Additionally, there is no guarantee that users who don an
IV. C ONCLUSIONS exosuit are comfortable with a non-deterministic robot acting
CRUX enables portable, multi-DoF actuation for the upper- out of their perceived control.
extremity, including both the forearm and the upper arm. To summarize, CRUX encapsulates flexible and compli-
In previous iterations of CRUX, the ability to control the ant augmentation for the upper extremity with a degree of
exosuit akin to Graded Motor Imagery and mirror therapy had capability that allows for tasks such as bilateral mimicry to
not been included. The current version of CRUX features both be achieved by a user. Future iterations of this exosuit will
hardware and software upgrades such as the additional frame- hopefully extend these principles and usher in even greater
work to better support IMUs and the mimetic algorithm to possibilities for portable, lightweight wearable robots.
implement the new controller. Additionally, we provide studies
and evaluations which provide evidence that CRUX augments
ACKNOWLEDGMENT
users without inhibiting their flexibility or compliance and can
therefore function as a rehabilitative device. The authors would like to thank Leonard Norton for pro-
CRUX augments users with power-lines superimposed on viding advice on the physical therapy movements that were
a flexible base. Along these power-lines, tension forces are translated into exosuit supported movements. Additionally,
directed through a tensegrity-inspired foundation. This foun- the authors would like to thank the Cabrillo College Stroke
dation creates a network upon which localized augmentation Center for their assistance with the user studies.
can occur, affording CRUX its portability and its compliance.
These considerations are crucial if CRUX is to comply with R EFERENCES
the natural flexibility and redundancy possessed by humans
as well as the practical need for wearable devices that can be [1] Centers for Disease Control and Prevention, “Prevalence of stroke–
United States, 2006–2010,” MMWR, Morbidity Mortality Weekly Rep.,
transported to out-of-clinic settings. vol. 61, no. 20, p. 379, 2012.
Through tendon-based Bowden cables, motors supplement [2] G. L. Moseley, “Graded motor imagery is effective for long-standing
movement in a particular arm (dictated by the modular complex regional pain syndrome: A randomised controlled trial,” Pain,
vol. 108, nos. 1–2, pp. 192–198, 2004.
backplate attached). We have demonstrated that a mimetic [3] A. Cacchio, E. De Blasis, V. De Blasis, V. Santilli, and G. Spacca,
controller can pull and release these cables to move a user’s “Mirror therapy in complex regional pain syndrome type 1 of the upper
arm according to the position of their opposing arm. This par- limb in stroke patients,” Neurorehabil. Neural Repair, vol. 23, no. 8,
pp. 792–799, 2009.
ticular controller has potential applications in fields including [4] D. Ezendam, R. M. Bongers, and M. J. A. Jannink, “Systematic review
physical therapy where patients are required to perform tedious of the effectiveness of mirror therapy in upper extremity function,”
and precise mirrored actions for rehabilitation. Disab. Rehabil., vol. 31, no. 26, pp. 2135–2149, 2009.
Although we have not completed a long-term study mea- [5] P. S. Lum, C. G. Burgar, P. C. Shor, M. Majmundar, and
M. Van der Loos, “Robot-assisted movement training compared with
suring the full efficacy of CRUX, data collected from our conventional therapy techniques for the rehabilitation of upper-limb
studies illustrate that users were immediately able to move motor function after stroke,” Arch. Phys. Med. Rehabil., vol. 83, no. 7,
their arms more easily once they learned how to operate CRUX pp. 952–959, Jul. 2002.
[6] S. Lessard, J. Bruce, E. Jung, M. Teodorescu, V. SunSpiral, and
during their respective tests. Literature in traditional physical A. Agogino, “A lightweight, multi-axis compliant tensegrity joint,” in
therapy highlights mobility as a key resource in patients’ Proc. IEEE Int. Conf. Robot. Automat. (ICRA), Stockholm, Sweden,
recovery post-injury/stroke, so we believe that the increase May 2016, pp. 630–635.
[7] G. Scarr and H. Harrison, “Examining the temporo-mandibular joint
in arm mobility afforded by CRUX will be an improvement from a biotensegrity perspective: A change in thinking,” J. Appl.
upon traditional rehabilitation training. Biomed., vol. 15, no. 1, pp. 55–62, 2017.
Future designs can extend these results by adding further [8] L. B. Baltaxe-Admony et al., “Simulating the human shoulder through
active tensegrity structures,” in Proc. ASME Int. Design Eng. Tech.
degrees of freedom to match the enormous level of redun- Conf. Comput. Inf. Eng. Conf. (IDETC/CIE), Charlotte, NC, USA,
dancy and fluidity expressed by the human musculoskeletal Aug. 2016, p. V006T09A027.
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.
LESSARD et al.: SOFT EXOSUIT FOR FLEXIBLE UPPER-EXTREMITY REHABILITATION 1617
[9] S. Lessard et al., “A bio-inspired tensegrity manipulator with multi-DOF, [24] N. Tsagarakis, D. G. Caldwell, and G. A. Medrano-Cerda, “A 7 DOF
structurally compliant joints,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. pneumatic muscle actuator (pMA) powered exoskeleton,” in Proc. 8th
Syst. (IROS), Oct. 2016, pp. 5515–5520. IEEE Int. Workshop Robot Human Interaction (RO-MAN), Sep. 1999,
[10] M. Solomonow, A. Guzzi, R. Baratta, H. Shoji, and R. D’Ambrosia, pp. 327–333.
“EMG-force model of the elbows antagonistic muscle pair. The effect [25] I. Galiana, F. L. Hammond, R. D. Howe, and M. B. Popovic, “Wearable
of joint position, gravity and recruitment,” Amer. J. Phys. Med., vol. 65, soft robotic device for post-stroke shoulder rehabilitation: Identifying
no. 5, pp. 223–244, 1986. misalignments,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst.,
[11] J. Gordon and C. Ghez, “EMG patterns in antagonist muscles during Oct. 2012, pp. 317–322.
isometric contraction in man: Relations to response dynamics,” Exp. [26] S. Lessard et al., “CRUX: A compliant robotic upper-extremity exosuit
Brain Res., vol. 55, no. 1, pp. 167–171, 1984. for lightweight, portable, multi-joint muscular augmentation,” in Proc.
[12] K. R. Saul et al., “Benchmarking of dynamic simulation predic- Int. Conf. Rehabil. Robot. (ICORR), Jul. 2017, pp. 1633–1638.
tions in two software platforms using an upper limb musculoskeletal [27] C. S. Simpson, A. M. Okamura, and E. W. Hawkes, “Exomuscle:
model,” Comput. Methods Biomech. Biomed. Eng., vol. 18, no. 13, An inflatable device for shoulder abduction support,” in Proc. IEEE
pp. 1445–1458, 2015. Int. Conf. Robot. Automat. (ICRA), May/Jun. 2017, pp. 6651–6657.
[13] S. L. Delp et al., “OpenSim: Open-source software to create and analyze [28] C. T. O’Neill, N. S. Phipps, L. Cappello, S. Paganoni, and C. J. Walsh,
dynamic simulations of movement,” IEEE Trans. Biomed. Eng., vol. 54, “A soft wearable robot for the shoulder: Design, characterization, and
no. 11, pp. 1940–1950, Nov. 2007. preliminary testing,” in Proc. Int. Conf. Rehabil. Robot. (ICORR),
[14] C. J. Walsh, K. Endo, and H. Herr, “A quasi-passive leg exoskeleton for Jul. 2017, pp. 1672–1678.
load-carrying augmentation,” Int. J. Humanoid Robot., vol. 4, no. 03, [29] R. E. Cowan, B. J. Fregly, M. L. Boninger, L. Chan, M. M. Rodgers,
pp. 487–506, 2007. and D. J. Reinkensmeyer, “Recent trends in assistive technology for
[15] W. van Dijk, H. van der Kooij, and E. Hekman, “A passive exoskeleton mobility,” J. Neuroeng. Rehabil., vol. 9, no. 1, p. 20, 2012.
with artificial tendons: Design and experimental evaluation,” in Proc. [30] A. S. Iberall, “The use of lines of nonextension to improve mobility in
IEEE Int. Conf. Rehabil. Robot. (ICORR), Jun./Jul. 2011, pp. 1–6. full-pressure suits,” Defense Tech. Inf. Center, Fort Belvoir, VA, USA,
[16] S. Toxiri, J. Ortiz, J. Masood, J. Fernández, L. A. Mateos, and Tech. Rep. AD0610519, 1964.
D. G. Caldwell, “A powered low-back exoskeleton for industrial han- [31] D. Newman, J. Hoffman, K. Bethke, J. Blaya, C. Carr, and B. Pitts,
dling: Considerations on controls,” in Wearable Robotics: Challenges “Astronaut bio-suit system for exploration class missions,” Nat. Iranian
and Trends. Cham, Switzerland: Springer, 2017, pp. 287–291. Amer. Council, Washington, DC, USA, Tech. Rep. NIAC Phase 2, 2005,
[17] K. Naruse, S. Kawai, H. Yokoi, and Y. Kakazu, “Development of p. 833.
wearable exoskeleton power assist system for lower back support,” [32] M. Ikai and T. Fukunaga, “Calculation of muscle strength per unit cross-
in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst., vol. 4, Oct. 2003, sectional area of human muscle by means of ultrasonic measurement,”
pp. 3630–3635. Int. Zeitschrift Angew. Physiol. Einschliesslich Arbeitsphysiol., vol. 26,
[18] H. In, B. B. Kang, M. Sin, and K. J. Cho, “Exo-glove: A wearable robot no. 1, pp. 26–32, 1968.
for the hand with a soft tendon routing system,” IEEE Robot. Autom. [33] R. W. Bohannon and A. W. Andrews, “Relationships between impair-
Mag., vol. 22, no. 1, pp. 97–105, Mar. 2015. ments in strength of limb muscle actions following stroke,” Perceptual
[19] C. N. Schabowsky, S. B. Godfrey, R. J. Holley, and P. S. Lum, “Devel- Motor Skills, vol. 87, pp. 1327–1330, Dec. 1998.
opment and pilot testing of HEXORR: Hand exoskeleton rehabilitation [34] R. W. Bohannon, “Motricity index scores are valid indicators of paretic
robot,” J. Neuroeng. Rehabil., vol. 7, no. 1, p. 36, 2010. upper extremity strength following stroke,” J. Phys. Therapy Sci., vol. 11,
[20] A. B. Zoss, H. Kazerooni, and A. Chu, “Biomechanical design of the no. 2, pp. 59–61, 2001.
Berkeley lower extremity exoskeleton (BLEEX),” IEEE/ASME Trans. [35] W. Ren, R. W. Beard, and E. M. Atkins, “A survey of consensus
Mechatronics, vol. 11, no. 2, pp. 128–138, Apr. 2006. problems in multi-agent coordination,” in Proc. Amer. Control Conf.,
[21] M. Wehner et al., “A lightweight soft exosuit for gait assistance,” in Jun. 2005, pp. 1859–1864.
Proc. IEEE Int. Conf. Robot. Automat., May 2013, pp. 3362–3369. [36] A. Rodriguez-Angeles and H. Nijmeijer, “Cooperative synchronization
[22] D. Baechle, “MAXFAS: A mobile arm exoskeleton for firearm aim of robots via estimated state feedback,” in Proc. IEEE Int. Conf. Decis.
stabilization,” Ph.D. dissertation, Univ. Delaware, Newark, NJ, USA, Control, vol. 2, Dec. 2003, pp. 1514–1519.
2013. [37] K. P. Dowd et al., “A systematic literature review of reviews on
[23] J. C. Perry, J. Rosen, and S. Burns, “Upper-limb powered exoskeleton techniques for physical activity measurement in adults: A DEDIPAC
design,” IEEE/ASME Trans. Mechatronics, vol. 12, no. 4, pp. 408–417, study,” Int. J. Behav. Nutrition Phys. Activity, vol. 15, no. 1, p. 15,
Aug. 2007. 2018.
Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on April 24,2023 at 19:07:16 UTC from IEEE Xplore. Restrictions apply.