Sunteți pe pagina 1din 14

Design of a biologically inspired lower limb exoskeleton for human gait

rehabilitation
Mingxing Lyu, Weihai Chen, Xilun Ding, Jianhua Wang, Shaoping Bai, and Huichao Ren

Citation: Review of Scientific Instruments 87, 104301 (2016); doi: 10.1063/1.4964136


View online: http://dx.doi.org/10.1063/1.4964136
View Table of Contents: http://aip.scitation.org/toc/rsi/87/10
Published by the American Institute of Physics

Articles you may be interested in


Neutron resonance spin echo with longitudinal DC fields
Review of Scientific Instruments 87, 125110 (2016); 10.1063/1.4972395

Sensitivity of magnetic field-line pitch angle measurements to sawtooth events in tokamaks


Review of Scientific Instruments 87, 11E541 (2016); 10.1063/1.4962058

A novel instrument to measure differential ablation of meteorite samples and proxies: The Meteoric Ablation
Simulator (MASI)
Review of Scientific Instruments 87, 094504 (2016); 10.1063/1.4962751

Mitigation of hard x-ray background in backlit pinhole imagers


Review of Scientific Instruments 87, 11E341 (2016); 10.1063/1.4962192

Invited Article: Concepts and tools for the evaluation of measurement uncertainty
Review of Scientific Instruments 88, 011301 (2017); 10.1063/1.4974274

High resolution imaging of 2D distribution of lithium in thin samples measured with multipixel detectors in
sandwich geometry
Review of Scientific Instruments 88, 023706 (2017); 10.1063/1.4977217
REVIEW OF SCIENTIFIC INSTRUMENTS 87, 104301 (2016)

Design of a biologically inspired lower limb exoskeleton for human


gait rehabilitation
Mingxing Lyu,1 Weihai Chen,1,a) Xilun Ding,2 Jianhua Wang,1 Shaoping Bai,3
and Huichao Ren1
1
School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, China
2
School of Mechanical Engineering and Automation, Beihang University, Beijing 100191, China
3
Department of Mechanical and Manufacturing Engineering, Aalborg University, Aalborg 9000, Denmark
(Received 14 March 2016; accepted 19 September 2016; published online 25 October 2016)
This paper proposes a novel bionic model of the human leg according to the theory of physiology.
Based on this model, we present a biologically inspired 3-degree of freedom (DOF) lower limb
exoskeleton for human gait rehabilitation, showing that the lower limb exoskeleton is fully compatible
with the human knee joint. The exoskeleton has a hybrid serial-parallel kinematic structure consisting
of a 1-DOF hip joint module and a 2-DOF knee joint module in the sagittal plane. A planar 2-DOF
parallel mechanism is introduced in the design to fully accommodate the motion of the human knee
joint, which features not only rotation but also relative sliding. Therefore, the design is consistent with
the requirements of bionics. The forward and inverse kinematic analysis is studied and the workspace
of the exoskeleton is analyzed. The structural parameters are optimized to obtain a larger workspace.
The results using MATLAB-ADAMS co-simulation are shown in this paper to demonstrate the
feasibility of our design. A prototype of the exoskeleton is also developed and an experiment
performed to verify the kinematic analysis. Compared with existing lower limb exoskeletons, the
designed mechanism has a large workspace, while allowing knee joint rotation and small amount of
sliding. Published by AIP Publishing. [http://dx.doi.org/10.1063/1.4964136]

I. INTRODUCTION ALEX III developed by Columbia University has four de-


grees of freedom at each leg (hip adduction/abduction, hip and
Walking is a fundamental motor skill of humans. Damage
knee flexion/extension, and ankle plantar/dorsiflexion).12,13 A
to central and peripheral nervous systems (e.g., in stroke pa-
force-field controller is used and proved to be an effective way
tients or spinal cord injury) often leads to a gait impairment and
to encourage patients to take an active role in the rehabilitation
reduced mobility.1,2 Research has demonstrated that through
process. The WalkTrainer, developed by EPFL, uses electri-
a repetitive rehabilitation treatment, patients may rebuild their
cal muscle stimulation to implement a closed-loop training
damaged nervous system and regain the ability to move.3 Robot
method.14,15 There are three active degrees of freedom at each
assisted training can play a key role during rehabilitation.4,5
leg (hip, knee, and ankle joints).15
The robotic exoskeleton is a wearable robot which can be
All the above robots have complex supporting systems to
used to strengthen physical fitness, aid the rehabilitation train-
help the patient maintain balance or support their body weight.
ing of paralysed patients, or assist people in various daily activ-
There are other kinds of rehabilitation robots that can move
ities.6 Exoskeleton robots have been studied extensively, with
freely. ReWalk, a commercial exoskeleton robot to help para-
the design and development of many focusing on the lower
plegic patients walk in daily life, has two rotational degrees of
limb. The Lokomat robot developed by ETH Zurich and Ho-
freedom at the hip and knee joints, actuated by motors. The
coma is one of the most successful and famous rehabilitation
robot can help patients to sit down, stand up, walk, and climb
robots.7–9 It has four degrees of freedom (left and right hip and
stairs. Different actions are controlled manually.16 The Hybrid
knee joints) actuated by four linear drives. The linear drives are
Assistive Limb (HAL) developed by the University of Tsukuba
built with a motor and a precision ball screw.7 An impedance
adopted a new control method. It measured surface electromy-
control method is used to control the Lokomat.8 LOPES devel-
ography (sEMG) and translated it to the desired joint torque
oped by the University of Twente is another lower-limb reha-
to control actuation.17,18 Other lower-limb rehabilitation robots
bilitation robot.10,11 Compared with the Lokomat, LOPES has
include NaTUre-gaits,19 ReoAmbulator,16 Ekso,20 Vanderbilt
up to eight active degrees of freedom. A cable driven method is
powered orthosis,21 and Rex.19
used in the knee joint design of LOPES. By placing the driving
Besides research on whole lower-limb rehabilitation ro-
motors on the base, the weight of the exoskeleton is reduced.
bots, there is a considerable body of work that focuses on a
Series elastic actuators are also included to improve the robot’s
single joint rehabilitation. This work is also critically impor-
compliance. LOPES also uses an impedance control method.11
tant for the development of lower-limb rehabilitation robots.
Since the knee joint plays an important role in walking, many
a)Authorto whom correspondence should be addressed. E-mail: researchers have focused on knee joint exoskeletons. For
whchen@buaa.edu.cn example, a knee orthosis developed by EPFL is a one degree

0034-6748/2016/87(10)/104301/13/$30.00 87, 104301-1 Published by AIP Publishing.


104301-2 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

FIG. 1. MR images of knee joint motion.

of freedom exoskeleton robot which can assist with knee


flexion/extension.22 Northwestern University also developed
a knee joint exoskeleton.23
Researchers from Charles University in Prague studied the
shape and relative movement of the knee joint with Magnetic
Resonance Imaging (MRI) and found the movement of the knee
joint is not simply rotation around a single point, but that the
center of rotation moves.24 Figure 1(a) shows the knee joint
at different rotation angels (10◦ and 90◦; upper segment is the
femur and the lower segment is the tibia). Figure 1(b) presents
an analysis of knee joint movement. When the rotation angle
is small, the movement of the femur is along the circular arc
CB; when the rotation angle is large, the movement of femur is
along the circular arc BA. The trajectory of the femur and tibia
can be approximated as an ellipse. This study shows that the FIG. 2. Comparing snapshots of an exoskeleton between two different knee
joint models.
movement of the knee joint cannot be simplified to a fixed-axis
rotation (pin joint). Based on the above findings, Professor Kok-
Meng Lee from the Georgia Institute of Technology compared bionic knee joint. Since the mechanism changes its instanta-
the difference between a biological knee joint and simplified pin neous center of rotation during movement, it is similar to the
joint.25 In Figure 2 the solid line represents the movement of a human knee joint. Thus the mechanism has a high degree of
biological knee joint, while the dashed line represents the knee bio-imitability. However, it is not practical to use the complex
simulated as a pin joint. It can be observed that when the rotation mechanism in rehabilitation.28 Yang from Zhejiang University
angle is large, there is a considerable difference between the two designed an adaptive knee joint exoskeleton using a cam mech-
joint types. Therefore the movement of the center of rotation anism to investigate the effects of different exoskeleton designs
cannot be neglected. This work provided guidance for the bionic on the internal joint forces/torque in the knee. The results
knee joint exoskeleton design. showed that an adaptive exoskeleton can minimize internal
Most current lower-limb rehabilitation robots simplified joint forces during the knee-exoskeleton interaction.26
the knee joint as a pin joint. The center of rotation of the knee From the view of human kinematics, we can find that
joint is commonly fixed, which does not match the human knee the complex coupled motion of the knee joint is actuated
joint and limits the workspace of the knee joint to a fixed arc. by several skeletal muscles along the articular surfaces, so
Since a patient walking with an exoskeleton is a closed kine- a parallel mechanism, which is similar to skeletal muscles,
matic loop, it is different to normal human walking without an may be useful in the design of a bionic knee joint. This paper
exoskeleton (open kinematic loop).26 If the movement of the proposes a new model of the human lower limb in the sagittal
exoskeleton is not transparent to natural human movement, it plane, which takes the complex coupled motion of knee joint
will exert extra forces on the knee joint. The extra force will into consideration. A simple parallel knee joint mechanism
lead to patient discomfort and an unnatural gait. for the human lower limb exoskeleton rehabilitation training
To examine the problem of knee exoskeletons that do not robot is designed and tested. The new mechanism can improve
mirror the motion of the human knee joint during walking, the bio-imitability and adaptability of the exoskeleton knee
Mehmet Alper Ergin and his team designed a 3-RRP parallel joint. An experiment is conducted to verify the kinematics and
knee joint mechanism.27 The mechanism allows the motion performance of the exoskeleton.
of the human knee joint and the exoskeleton knee joint to be The paper is organized as follows: Section II presents a
closely matched throughout lower limb flexion and extension. novel model of the human lower limb. Section III describes the
The main drawbacks of the mechanism are that its size and biomechatronic design of lower limb exoskeleton. Kinematic
the lack of coordination with other parts of the lower limb analysis of the exoskeleton robot is presented in Section IV.
exoskeleton design. Dewen Jin proposed a six-bar mechanism Section V focuses on the workspace analysis and structure
104301-3 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

optimization. Section VI introduces the control systems. plane, there is both rolling and sliding between the contact
Co-simulation results and experimental investigation are surface of the femur and tibia, which can be observed through
shown in Section VII and Section VIII in order to support the MRI.24 So the instantaneous rotation center of the knee joint
design phase. Finally, conclusions are presented in Section IX. is not fixed, but moves in the sagittal plane during knee flexion
and extension. When the angle of knee flexion and extension
is large, the sliding displacement of rotation center is not
II. KINEMATIC MODEL OF THE HUMAN LOWER LIMB negligible.30
Modeling the human lower limb is the foundation of Based on previous research, we developed a lower limb
bionic exoskeleton design. In this section, we will describe model in the sagittal plane, as shown in Figure 3. We take hu-
the complex coupled motion of the knee joint and establish man knee complex coupled motion into consideration. There
a biological human lower limb kinematic model. are two motions additional to rotation θ 2, one along the femur
The human lower limb has three joints, namely the hip, k y and one perpendicular to femur k x . The kinematics of the
knee, and ankle joints. The hip joint is a ball and socket joint human lower limb is therefore described by the following
with a center of rotation position that is nearly fixed. For the equation:
ankle joint, its rotation angle is small, and the change of the TB A θ 1, θ 2, k x , k y , θ 3 = TBH (θ 1) TH K θ 2, k x , k y TK A (θ 3) ,
 
rotation center is also nearly fixed. In the sagittal plane, the
(1)
hip and ankle joints can thus be considered as rotation joints
which have a fixed axis. In comparison, the human knee joint where TBH (θ 1) is the hip joint transformation matrix related
to the body coordinate, TH K θ 2, k x , k y is the knee joint trans-

is one of the most complex mechanical systems in the human
body and plays a critical role during gait. The lower femur formation matrix related to the hip coordinate, TK A (θ 3) is the
and upper tibia interface shape is irregular during flexion.29 In ankle joint transformation matrix  related to the knee coordi-
terms of the kinematics, the knee joint cannot be simplified nate, and TB A θ 1, θ 2, k x , k y , θ 3 is the ankle joint transforma-
to a single rotation axis. Even during flexion in the sagittal tion matrix related to the body coordinate. The result is

 cos (θ 1 − θ 2 + θ 3) sin (θ 1 − θ 2 + θ 3) 0 l t + k y sin (θ 1) − k x cos (θ 1) + l s sin (θ 1 − θ 2) 




 − sin (θ 1 − θ 2 + θ 3) cos (θ 1 − θ 2 + θ 3) 0 l t + k y cos (θ 1) + k x sin (θ 1) + l s cos (θ 1 − θ 2)


  
TB A θ 1, θ 2, k x , k y , θ 3 =   . (2)
 0 0 1 0 
 0 0 0 1 

The position vector of the end point P expressed in the ankle robot produce motion similar to the human lower limb, we
joint coordinate system is given by A P. Let the position vector prefer that the robot has the same configuration of degrees of
of P with respect to the body coordinate system be B P. Then freedom (DOFs). The designed robot has three active degrees
we can relate A P to B P by the following transformation:
B
P = TB A θ 1, θ 2, k x , k y , θ 3 · A P.

(3)

Hence, given the hip joint θ 1, the knee joint θ 2 and its related
motion k x and k y , the ankle joint θ 3, the thigh length l t , the shank
length l s , and the position of point P can be computed by Eq. (3).
Note that the planar motions of knee joint k x and k y depend on
the patient’s physiological structure and are coupled with the
knee joint θ 2. This means the planar motions of knee joint k x
and k y can be expressed as k x = f x (θ 2) , k y = f y (θ 2). So the
knee joint has two active degrees of freedom. In the case when
the knee joint angle θ 2 is different, k x and k y are different.

III. STRUCTURAL DESIGN OF LOWER


LIMB EXOSKELETON
The above analysis shows that there is one active degree
of freedom at the hip joint, two active degrees of freedom at
the knee joint, and one active degree of freedom at the ankle
joint in the sagittal plane. In order to make the rehabilitation FIG. 3. Bionic model of lower limb.
104301-4 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

the length of linear drivers is appropriately selected, a large


workspace can be achieved. This will be discussed in Section V
in detail.
When the planar 2-DOF parallel mechanism RPRPR is
used to produce the motion of human knee, the exoskeleton
knee joint will be a virtual joint. There are three mechanisms
available which can be seen in Figure 5. The first design
scheme is to place both upper ends of the knee linear actuators
on the exoskeleton thigh. The second design scheme is to fix
one upper end of a knee linear actuator on the exoskeleton
thigh and the other on the exoskeleton pelvis. The third is
to place both upper ends of the linear knee actuators on the
exoskeleton pelvis. All of the mechanisms consist of two linear
actuators at the knee joint and one linear actuator at the hip
joint. The lower ends of the two knee linear actuators are
connected with a shaft. When the mechanisms are at upright
FIG. 4. Planar 2-DOF parallel mechanism.
posture, the linear actuators are located at the rear side of the
leg. The two actuators are staggered to avoid interference.
of freedom in the sagittal plane, namely one DOF at the hip Comparing the three kinds of design schemes, we can
joint and two DOFs at the knee joint. There is another negative find that the first one is a serial-parallel mechanism which is
joint at the ankle. composed of a 1-DOF rotary hip joint and a planar 2-DOF
Motion in sagittal plane is the most important during hu- parallel knee joint. The second is a pure parallel mechanism
man walking. Besides, the ranges of hip and ankle adduc- which has two branched chains. The third is decoupling mech-
tion/abduction and rotation motion are quite small compared anism, which has a parallel manipulator in the knee joint.
to the flexion/extension motion.32 In order to provide the right All can produce a planar 2-DOF motion in the knee joint.
gait cycle training for patients and not to make the control too Obviously, the parallel mechanism possesses the advantages
complex, we only consider the motion in the sagittal plane of high stiffness, low inertia, and large payload capacity. Com-
and simplify the hip joint and ankle joint as a 1-DOF hinge, paring the three strategies, we can find the last two schemes
similar to most other designs. Unlike most previous exoskeleton are more force efficient. However, their workspace is very
designs that have not taken into account the center of rota- small. To achieve the required workspace, the stroke of the
tion change of the knee joint, we incorporated 2-DOF planar linear knee actuators will be very large, which is not feasible
mechanism RPRPR at the knee joint, as shown in Figure 4. in practice. What’s more, both the second and third designs are
In the mechanism RPRPR, R stands for a revolute joint and P not biomimetic. Based on the above consideration, we finally
stands for a prismatic joint. In our design, the two prismatic chose the serial-parallel mechanism (design (a)) for the lower
joints work as the active joints and three revolute joints act limb exoskeleton.
as passive joints. In Figure 4, the blue parts are the prismatic Figure 6 shows the final design of the lower limb exoskel-
joints, namely the actuators. It is easy to recognize that the eton. The exoskeleton is designed for the rehabilitation of
mechanism can realize 2-DOF planar motion which is suit- paraplegic or hemiplegic patients between 1.60 and 1.80 m
able to simulate the movement of the knee joint. As long as tall, with a maximum body weight of 100 kg. It is conceived

FIG. 5. The lower limb exoskeleton of three different configurations.


104301-5 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

only one leg of the exoskeleton is developed. The total mass


of the exoskeleton is about 14 kg, including the 3 linear driving
actuators. The mechanical structure is designed to allow active
and passive movements in the sagittal plane. In the frontal
plane, small passive movements are possible at the hip joint.
The exoskeleton is adjustable to the patient, which means
it can be applied to patients of different somatotype. The static
height of the hip joint can be adjusted by changing the height of
the exoskeleton pelvis. The width of the hip can be changed by
moving an adjustable tube. The length of leg can be adjusted by
changing the length of the shank. Since there is only a virtual
knee joint, we need not match the human knee joint with the
exoskeleton knee joint mechanism.
Absolute encoders are placed in the exoskeleton, two at
the hip joint and three at the knee joint, which can be seen in
FIG. 6. Lower limb exoskeleton with bionic knee joint.
Figure 7 (the white cylinders in the last figure). Three of the
absolute encoders are used for control, while the other two are
for assist-as-needed overground gait training in a clinical envi- for safety. There is also a force sensor implemented in each
ronment and to be used by patients who are unable to stabilize linear actuator in series, which will be used to measure the
the trunk. driving force for control in the future. Twelve patch type force
The hip and knee joints are driven by linear driving actua- sensors are placed inside the pressure sensing foot.
tors, each of which consists of a DC motor, a toothed belt, and So far, we have presented the structural design of our
a ball screw. The hip linear actuator is placed on the pelvis exoskeleton. Based on the traditional exoskeleton, we pro-
of the lower limb exoskeleton, while the linear knee actuators posed the use of an RPRPR parallel mechanism at the knee
are placed on the thigh of the exoskeleton. There is another joint. Different design strategies were considered. Compar-
negative joint at the ankle, which connects the shank and the ing the workspace, the size of the mechanism and the bi-
pressure sensing foot. The range of motion for the hip joint is onic performance, we chose a serial-parallel mechanism. In
20◦ in flexion and 36◦ in extension, while the range of motion Sec. IV, we will focus on the kinematic analysis of the
for the knee joint is 70◦ in flexion and 0◦ in extension. For exoskeleton.
the ankle, plantar-flexion is shown as extension and dorsi-
flexion as flexion. The maximum flexion and extension values
IV. KINEMATIC ANALYSIS
of the ankle joint are 25◦ and 20◦. These ranges of motion are
decided according to Winter’s study on healthy subjects and After defining the layout of the lower limb exoskeleton
stroke patients.33 The range of joint angles is mechanically mechanics and actuation, we developed a kinematic model
limited to avoid injury to the user. The pressure sensing foot of the exoskeleton. This model was used to both identify
can measure the contact force between the foot and the ground. a set of suitable geometrical parameters for the mechanism
The acquired signal can be used to evaluate the patient’s gait structure and control the robot. The analysis will be sepa-
and also control the robot. Aluminum and stainless steel are rated into two parts: the hip joint and the knee joint. We will
primarily used in the mechanical structure in consideration of then analyze them together and also take the human leg into
mechanical resistance and lightweight. In the primary study, consideration.

FIG. 7. Mechanical parts of the exoskeleton legs assembled.


104301-6 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

B. Knee joint kinematic analysis

For the planar 2-DOF parallel mechanism RPRPR, the


two fixed pivots E and F define the geometry of a fixed base,
and the pivot P defines the geometry of a moving platform.
Two limbs connect the moving platform at point P to the fixed
base at points E and F by revolute joints. Each limb consists
of two links connected by a prismatic joint.
The origin of the fixed coordinate frame is located at point
O1. The y-axis points along the direction of the thigh and the
x-axis is perpendicular to the thigh. Figure 9 shows the lengths
FIG. 8. Hip joint kinematic analysis.
of limbs and joint angles of a typical position. The location of
the moving platform P can be specified in terms of the position
A. Hip joint kinematic analysis of point P (x P , yP ).
For the two cylinders, the following conditions hold:
For the hip joint, there is a rotary degree of freedom. The
pelvis of the exoskeleton can be treated as the fixed link and (x P − x E )2 + ( yP − yE )2 = l 12, (9)
the thigh will be treated as a rotary link. From Figure 8, we can (x P − x F ) + ( yP − yF ) = l 2 ,
2 2 2
(10)
describe the geometric relationship between the linear actuator
length l 0 and the hip joint angle θ 1 where (x E , yE ) and (x F , yF ) are coordinates of two points E
and F, respectively.
θ 1 = π/2 − δ1 − δ2 − ϕ0, For the inverse kinematics, x P and yP are given, and the
d 12 + d 22 − l 02 (4) lengths of limb l 1 and l 2 are to be found. Solving Eqs. (9) and
cos ϕ0 = .
2d 1d 2 (10) for l 1 and l 2 yields
For the direct kinematics of the hip joint, the linear actu-

ator length l 0 is known and the hip joint angle θ 1 is to be found. l 1 = (x P − x E )2 + ( yP − yE )2,
 (11)
From Eq. (4), we can get l 2 = (x P − x F )2 + ( yP − yF )2.
d 12 + d 22 − l 02 Hence, corresponding to each given location of the mov-
θ 1 = π/2 − δ1 − δ2 − arccos . (5)
2d 1d 2 ing platform, there are generally two possible solutions for
For the inverse kinematics of the hip joint, hip joint angle each limb. However, the negative limb length is physically not
θ 1 is known and the linear actuator length l 0 is to be found. feasible. So Eq. (11) shows the final result.
From Eq. (4), we can get For the direct kinematics, the lengths of limb l 1 and l 2 are
 given, and the position (x P , yP ) of the moving platform are to
be found. (9)–(10) yields
l 0 = d 12 + d 22 − 2d 1d 2 cos(π/2 − δ1 − δ2 − θ 1). (6)
xF − xE (l 12 − l 22) + (x F 2 − x E 2) + ( yF 2 − yE 2)
A velocity equation is obtained by taking the derivative of yP = − xP +
yF − yE 2( yF − yE )
Eq. (5) with respect to time
= sx P + t, (12)
−2l 0
θ̇ 1 =  l˙0. (7) where
[(d 1 + d 2 − l 0 ][l 0 − (d 1 − d 2
)2 2 2
) 2] xF − xE
s=− ,
yF − yE
Writing the above equation in the form of ẋ = J q̇, where
x = θ 1, q = l 0, hence the Jacobian matrix of the hip joint is (l 12 − l 22) + (x F 2 − x E 2) + ( yF 2 − yE 2)
t= .
given by 2( yF − yE )
Equation (12) together with Equation (9) forms a new
−2l 0
J0 =  . (8) system of equations. We may solve Eqs. (9) and (12) for
[(d 1 + d 2)2 − l 02][l 02 − (d 1 − d 2)2] x P , yP .


[x E − s(t − yE )] ± [s(t − yE ) − x E ]2 − (s2 + 1) x E 2 + (t − yE )2 − l 12
 
xP = ,
(s2 + 1)
 (13)
[x E − s(t − yE )] ± [s(t − yE ) − x E ]2 − (s2 + 1) x E 2 + (t − yE )2 − l 12
 
yP = s + t.
(s2 + 1)

Hence, corresponding to each given set of input limb lengths, there are two possible mechanism configurations. One of
the configurations will be rejected for the limitation of the mechanism, which will be described later. The final result can be
104301-7 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

described as

[x E − s(t − yE )] − [s(t − yE ) − x E ]2 − (s2 + 1) x E 2 + (t − yE )2 − l 12
 
xP = ,
(s2 + 1)
 (14)
[x E − s(t − yE )] − [s(t − yE ) − x E ]2 − (s2 + 1) x E 2 + (t − yE )2 − l 12
 
yP = s + t.
(s2 + 1)

Let the actuated joint variables be represented by a vector that is,


q and the location of the moving platform be described by (x − x ) ( y − y )  ẋ  l 0  l˙ 
a vector x. According to the Jacobian analysis of parallel  P F P F   P
= 
1   1 . (16)
manipulators,31 the relationship between the input joint rates (x P − x E ) ( yP − yE )  ẏP   0 l 2 l˙2
  
and the end-effector output velocity is as follows:
Write (16) in the  form of
(x P − x F ) (y P − y F )
l J0xẋ = Jq q̇, where
Jx ẋ = Jq q̇. Jx = (x P − x E ) (yP − yE ) and Jq = 01 l2 .
As l 1 and l 2 do not vanish, matrix Jq is not singular,
Hence the overall Jacobian matrix, J can be written as the overall Jacobian matrix of RPRPR mechanism can be
expressed as follows:
q̇ = J ẋ, where J = Jq −1 Jx .
 x P − x F yP − yF 
 For the mechanism RPRPR, the input vector is q  l l 1 
T T J = Jq−1 Jx =  x −1 x yP − yE  . (17)
= l 1 l 2 and the output vector is x = x P yP .  P E

Differentiating both sides of (9) and (10), we can get  l2 l 2 
the relationship between input speed and output speed of the
RPRPR mechanism
C. Whole leg kinematic analysis
(x − x F ) ẋ P + ( yP − yF ) ẏP = l 1 · l˙1 For the whole leg, the exoskeleton is a serial-parallel
 P

 (x P − x E ) ẋ P + ( yP − yE ) ẏP = l 2 · l˙2 ,

(15)
 mechanism. We have analyzed the serial part and parallel part,
respectively. Now we will combine them together.
For the direct kinematics, l 0, l 1, and l 2 are given, and the
position P(x, y) of the platform is to be found. From the anal-
ysis above, we can obtain θ 1 from l 0 by the Eq. (5). The position
of P(x P , y P ) in the coordinate O1 x 1 y1 can also be determined
from l 1 and l 2 by the Eq. (14). Then the final position of P(x, y)
in the coordinate Ox y is given by

P(x, y) = 0R1 P(x P , y P ), (18)


 cos θ sin θ

where 0 R1 = − sin θ11 cos θ11 .
For the inverse kinematics, the position P(x, y) of the
platform is given, and the actuators lengths l 0, l 1, and l 2 are
to be found. From Eq. (18), we can find there are two equa-
tions; however, three unknown quantities, namely θ 1, x P , y P .
The number of unknown quantities is more than the num-
ber of equations. Therefore it cannot be solved and there are
numerous solutions. From the above analysis, we know θ 1 is
related with l 0 by Eq. (15) and x P , y P are related with l 1 and l 2
by Eq. (14). So there are numerous solutions of l 0, l 1, and l 2,
too.

D. Leg and exoskeleton kinematic analysis together


If we want the previous problem to be solved, another
condition should be given. The condition can be obtained from
the human lower limb motion. From the analysis above, θ 1
equals to the angle of human hip joint, which we need not
analyze, and P(x P , y P ) relates to the motion of human knee
joint. In this part, we will develop the kinematic model of the
FIG. 9. Knee joint kinematics analysis. exoskeleton coupled with the human leg.
104301-8 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

V. WORKSPACE ANALYSIS AND OPTIMIZATION


A. Workspace analysis
One of the main advantages of our robot design is that it
has a large workspace which enables it to produce the complex
coupled knee joint motion in a broad range of patients. The
workspace of the robot is defined as the set of points that
can be reached by its end-effector, namely point P. In this
section, we will analyze the workspace characteristics and
shape to evaluate the efficiency and kinematic performance of
the exoskeleton.
The Monte Carlo method is a numerical approach to solve
mathematical problems by means of random sampling. Imple-
menting the kinematic model in MATLAB, we can simulate
the workspace of the robot with the Monte Carlo method.
Figure 11(a) shows the workspace of the robot knee joint,
FIG. 10. Whole leg and exoskeleton kinematics analysis together. namely the parallel mechanism PRRPR, and Figure 11(b)
shows the workspace of the whole leg.
As shown in Figure 10, for the inverse kinematics, the We will now focus on the knee joint and study the
human hip joint θ 1 and knee joint θ 2 traces can be gained workspace of the parallel mechanism by graphical method. As
from healthy subjects. From the previous human lower limb shown in Figure 12, the area within the red line is the effective
kinematic model in Section II, if we know patient’s knee joint workspace. This is obtained by changing the length of linear
configuration k x and k y , we can obtain P(x P , y P ) in O1 x 1 y1 actuators l 1 and l 2.
coordinate system In Figure 12, the bold black line represents a simplified
lower limb exoskeleton, and the blue line represents three
x P = −k x − l k s sin θ 2 − c2 cos θ 2, linear actuators. The yellow arc shows the trajectory of the
(19)
yP = l t + k y + l k s cos θ 2 − c2 sin θ 2. end effector with a simplified pin joint, while the green arc
shows the trajectory of the end effector of an actual knee
where l k s is the length of K S and c2 is the length of PS
joint. It is apparent that there is a difference between the
(Fig. 10).
trajectories. When the rotation angle is large, the difference
Then, the lengths of limb l 1 and l 2 can be obtained from
cannot be ignored. Note that the area within the red line,
Eq. (11).
namely the workspace of the designed mechanism, covers both
For the direct kinematics, l 1 and l 2 are given, and k x and
of the previous two trajectories. This indicates that we can
k y of human knee joint are to be found. From Eq. (14), we can
use this mechanism to simulate the biological trajectory to
obtain P(x P , y P ). From Eq. (19), we can get
produce the complex coupled motion of the knee joint. Since
k x = −x P − l k s sin θ 2 − c2 cos θ 2, the workspace of the mechanism is large, it is possible to
(20) customize different working trajectories according to each pa-
k y = yP − l t − l k s cos θ 2 + c2 sin θ 2.
tient’s physiological structure. As the planned trajectory is fit
If we can measure the knee joint angle θ 2 at different to patient’s physiological structure, the robot will not put extra
position P(x P , y P ), we can calculate k x and k y from the above force on patient’s knee joint and will enable the patient to walk
equation. This process can be used to measure different pa- naturally. The large workspace also necessitates that a range
tients’ knee joint configuration. limiting mechanism is fixed to the exoskeleton, which can be

FIG. 11. Workspace analysis.


104301-9 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

labelled S1 and S2. FG = F H = l 2, F J = F I = FK = l 2 + x 2,


and EH = EI = l 1, where l 1 and l 2 are the shortest lengths
of the two linear actuators, respectively, and x 1 and x 2 are
the stroke lengths of the two linear actuators, respectively.
According to the sector area and geometric relationship, we
can determine S1 and S2.
From Figure 13, we know that S1 can be obtained by
S1 = SsecFKI − S∆F H I − (SsecEHI − S∆E H I ), (21)
where SsecFKI is the area of sector FK I, S∆F H I is the area of
triangle F H I, SsecEHI is the area of sector EH I, and S∆E H I is
the area of triangle EH I.
S2 can be obtained by subtracting the sector area FGH
from the sector area FGK
S2 = SsecFJK − SsecFGH. (22)
Then the workspace S can be obtained by adding S1 and
S2 together

FIG. 12. Workspace of the parallel knee joint mechanism. S = S1 + S2. (23)

seen in Figure 12. This ensures that excessive movement of B. The optimization of structural parameters
knee joint during walking is avoided, providing user safety. After choosing the linear actuators according to the torque
The workspace can be displayed graphically. As shown and length requirements, a simulation of the planar 2-DOF
in Figure 13, GHIKJ is the outline of the workspace S, which parallel mechanism was implemented in MATLAB in order
can be divided into two parts. The areas of the two parts are to identify a set of suitable values to place the two linear
actuators. For this purpose, we iteratively modified a, b, c, d
in Figure 13 until we achieved a maximum workspace area.
Considering the human leg length and the need for a compact
structure in order to avoid interference, the ranges of a, b, c, d
are chosen as follows:
300 mm ≤ a ≤ 390 mm, 200 mm ≤ b ≤ 300 mm,
60 mm ≤ c ≤ 120 mm, 30 mm ≤ d ≤ 50 mm.
The simulation revealed the following design parameters:
a = 360 mm, b = 210 mm, c = 90 mm, and d = 50 mm. The
maximum area of the workspace is Smax = 44 962 mm2. These
parameters were used to place the two linear actuators.

VI. THE CONTROL SYSTEM OF THE LOWER


LIMB EXOSKELETON
The exoskeleton control system consists of two parts: the
host computer and the target computer. The therapist controls
the exoskeleton via a user interface running on the host com-
puter. It consists of a graphic user interface (GUI) and a data-
base. The host computer also generates the required trajectory
and transfers it to the target computer. The target computer
consists of three parts: the power management module, a digital
signal processor (DSP), and a field-programmable gate array
(FPGA). The DSP is used to control the motors and commu-
nicate with the host computer by a Controller Area Network
(CAN) field-bus, while the FPGA is used to process the force
and position messages from the force sensors and absolute en-
coders. Different training modules can be selected via the GUI.
The control strategy of the exoskeleton is depicted in
Figure 14. The controller is partitioned into two parts: the
FIG. 13. The calculation method of workspace area. position controller and the velocity controller. Both use a
104301-10 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

FIG. 14. The control strategy of the exoskeleton.

FIG. 15. The simulation of linear interpolation.

FIG. 16. The simulation of human walking step.

proportional–integral–derivative (PID) control strategy. This the final results. The actual trajectory can trace the desired
controller can achieve a high temporal resolution. In the trajectory accurately. The second trajectory is the human step.
controller, the reference trajectory is generated by the host The hip and knee joint trajectories of a human step are obtained
computer. The closed-loop controller is implemented on the from Winter’s study on healthy subjects.33 We use the hip and
DSP. knee trajectories to derive the desired trajectory and use the
absolute encoders as feedback to derive the actual trajectory.
The results of the exoskeleton simulating human walking are
VII. MATLAB - ADAMS CO-SIMULATION shown in Figure 16. The simulation results demonstrate that
OF THE KINEMATICS the robot can be controlled well and has the potential to help
The motion of the lower limb exoskeleton is simulated patients regain the ability to walk by producing a standard
with MATLAB/SIMULINK and MSC Adams. We coded the walking trajectory.
control system in MATLAB and mechanism system in Adams
and linked them together. In this co-simulation, the control
VIII. EXPERIMENTAL INVESTIGATION
system in MATLAB sends commands to Adams, while the
AND DISCUSSION
mechanism system in Adams processes the commands and
sends feedback to the control system.34 The objectives of the experimental study are to vali-
Two types of trajectories are tested. The first one is a date the kinematic model of the exoskeleton and to evaluate
linear trajectory. We make the exoskeleton end effector (Point the performance of the exoskeleton. Figure 17 (Multimedia
P) move from (−60, 750) to (−300, 600). Figure 15 shows view) shows the prototype of the exoskeleton. There are three
104301-11 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

FIG. 19. Position tracking results when the exoskeleton working like
Lokomat.

this process is 0.032 rad (1.8◦). Results indicate that the robot
can generally follow the desired trajectory to an end point
and return. The deviation between the desired and measured
trajectory is mostly due to machining and installation errors.
We performed another experiment to validate that the
exoskeleton could move in a similar manner to the Lokomat,
which treats the knee as a rotary joint for simplification. In
this experiment, we suppose the length of thigh is 470 mm.
The ends of the thigh are two rotary joints, namely the hip
FIG. 17. The prototype of the exoskeleton. (Multimedia view) [URL: http://
dx.doi.org/10.1063/1.4964136.1]
and knee joints. This requires that we make an end effector of
the parallel mechanism at the knee joint to produce a special
trajectory, which uses the knee joint as the center to form a
absolute encoders used for feedback, one placed at the hip circle. The planned trajectory was the clockwise rotation of
joint and the other two placed at the upper ends of the linear the hip joint from 0◦ to −15◦ and then a return to 0◦. At the
actuators. same time, the knee joint rotates anticlockwise from 0◦ to
In the first experiment, we will make the exoskeleton 25◦ and then returns to 0◦. We measured the hip joint and
end effector (point P) move from (−100, 730) to (−200, 650) knee joint angles, which can be seen in Figure 19. The final
along a line and then return to the start point. During the trajectory of point P can also be seen in Figure 20. The position
motion, the hip joint rotates clockwise from 0◦ to −10◦ and tracking error of end point P was calculated and displayed in
then returns to 0◦. We will test the exoskeleton’s tracing ability Figure 21. The maximum error during this period is less than
by measuring changes in joint angle. The results are shown 6.5 mm, which is negligible compared to the human leg length.
in Figure 18, in which the solid lines represent a series of This experiment verifies that our exoskeleton can work in a
desired joint angles and the dashed lines represent joint angles similar manner to the Lokomat. The purpose of this compar-
measured by encoders. The maximum joint angle error during ison is to compare our exoskeleton to the industry standard

FIG. 18. The experiment results of line trajectory tracking. FIG. 20. Position tracking results of end point P.
104301-12 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

and inverse kinematics are analyzed. The workspace of the


exoskeleton and optimization of the structural parameters are
also presented. A 3D model of the lower limb exoskeleton is
also presented in the paper. Co-simulation and experimental
results demonstrate the function of the exoskeleton. We show
that the exoskeleton can perform human step trajectories to
assist walking. We also provide evidence that the exoskeleton
has good trajectory tracking ability within its workspace. Since
the exoskeleton knee joint’s workspace is at least that of bio-
logical knee joint, we can conclude that a customized trajec-
tory within the workspace can align with the movement of a pa-
tient’s knee joint, which is beneficial for patient rehabilitation.

ACKNOWLEDGMENTS
FIG. 21. Position tracking error of end point P. The work reported in this paper is supported by the Na-
tional Nature Science Foundation of China (under Grant No.
Lokomat which is an extraordinary device with many clinical 61573047 and No. 51675018). M.L. would like to thank Chao
applications. Given that our exoskeleton can produce the same Liao for technical assistance and Daniel Woolley for proof-
trajectory, it highlights the practical utility of our exoskeleton. reading the paper.
A video of the experiments can be viewed (Multimedia view).
Furthermore, our simulation results also demonstrate that the 1N. M. Crewe and J. S. Krause, Medical, Psychosocial and Vocational
robot can closely replicate human walking. Aspects of Disability (Elliott and Fitzpatrick, Athens, 2009), pp. 289–304.
2S. J. Olney and C. Richards, Gait Posture 4, 136 (1996).
From the two experiments, we suggest that the designed 3E. Taub, G. Uswatte, and T. Elbert, Nat. Rev. Neurosci. 3, 228 (2002).
exoskeleton can work well to perform the desired task. It can 4C. D. Takahashi, L. Der-Yeghiaian, V. Le, R. R. Motiwala, and S. C. Cramer,

track any planned trajectory in the workspace. It can also work Brain 131, 425 (2008).
5M. Knikou and C. K. Mummidisetty, J. Neurophysiol. 111, 2264 (2014).
in a similar manner to previous designs which treat the knee 6A. M. Dollar and H. Herr, IEEE Trans. Rob. 24, 144 (2008).
as having a fixed center of rotation. However, since there is 7G. Colombo, M. Joerg, R. Schreier, and V. Dietz, J. Rehabil. Res. Dev. 37,
another active degree of freedom at the knee joint, it makes 693 (2000).
8S. Jezernik, G. Colombo, and M. Morari, IEEE Trans. Rob. Autom. 20, 574
it possible to customize the trajectory for individual patients
according to their physiological structure. Furthermore, the (2004).
9R. Riener, L. Lünenburger, I. C. Maier, and G. Colombo, J. Healthcare Eng.
large workspace allows the design of different trajectories for 1, 197 (2010).
different patients. 10J. F. Veneman, R. Ekkelenkamp, R. Kruidhof, F. C. T. Van Der Helm, and

In the negative control strategy, we can also use the robot H. Van Der Kooij, Int. J. Rob. Res. 25, 261 (2006).
11H. Vallery, J. Veneman, E. Van Asseldonk, R. Ekkelenkamp, M. Buss, and
to measure the physiological structure of the patient’s knee,
H. Van Der Kooij, IEEE Rob. Autom. Mag. 15, 60 (2008).
which will be implemented for future clinical applications. 12S. K. Banala, S. H. Kim, S. K. Agrawal, and J. P. Scholz, IEEE Trans. Neural
Different patients have different sliding characteristic param- Syst. Rehabil. Eng. 17, 2 (2009).
13D. Zanotto, P. Stegall, and S. K. Agrawal, in Proceedings of the IEEE
eters (k x and k y ), which will affect the trajectory of the end
effector P. As presented in Section IV D, by measuring the International Conference on Robotics and Automation (ICRA) (IEEE,
2014), pp. 724–729.
knee joint angle θ 2 at different positions P(x P , y P ), the sliding 14M. Bouri, Y. Stauffer, C. Schmitt, Y. Allemand, S. Gnemmi, and R.
parameters in an individual can be obtained from Eq. (20). Clavel, in Proceedings of the IEEE International Conference on Robotics
Position P(x P , y P ) can be easily determined by detecting the and Biomimetics (ROBIO) (IEEE, 2006), pp. 1616–1621.
15Y. Stauffer, Y. Allemand, M. Bouri, J. Fournier, R. Clavel, P. Metrailler, R.
linear actuator length l 1 and l 2. Regarding the patient’s knee
Brodard, and F. Reynard, IEEE Trans. Neural Syst. Rehabil. Eng. 17, 38
joint angle θ 2, it will be measured with a goniometer (Bio- (2009).
metrics Ltd). By putting one segment on the thigh and the 16I. Díaz, J. J. Gil, and E. Sánchez, J. Robotics 2011, Article ID 759764

other on the shank, the goniometer can detect the knee joint (2011).
17H. Kawamoto and Y. Sankai, Computers Helping People with Special Needs
angle in real time. Obtaining the parameters that describe an (Springer Berlin Heidelberg, 2002), pp. 196–203.
individual’s sliding characteristics at the knee joint allows for 18T. Hayashi, H. Kawamoto, and Y. Sankai, in Proceedings of the IEEE/RSJ

customized trajectories. The goal of customized trajectories is International Conference on Intelligent Robots and Systems (IROS) (IEEE,
to improve patient rehabilitation. In future work, we will test 2005), pp. 3063–3068.
19K. H. Low, in Proceedings of the IEEE International Conference on
the exoskeleton on patients. Defense Science Research Conference and Expo (DSR) (IEEE, 2011),
pp. 1–10.
20E. Strickland, “Good-bye, wheelchair,” IEEE Spectrum 49(1), 30–32

IX. CONCLUSIONS (2012).


21R. J. Farris, H. A. Quintero, S. A. Murray, K. H. Ha, C. Hartigan, and M.

This paper presents a novel design of a lower limb Goldfarb, IEEE Trans. Neural Syst. Rehabil. Eng. 22, 482 (2014).
22C. Schmitt and P. Métrailler, in 9th Annual Conference of the International
exoskeleton, which is fully compatible with the human knee FES Society. No. LSRO2-CONF-2006-010 (2004).
joint. The linkage type selection is justified. An RPRPR 23G. Aguirre-Ollinger, J. E. Colgate, M. A. Peshkin, and A. Goswami, IEEE

planar 2-DOF parallel mechanism is used. Forward kinematics Trans. Neural Syst. Rehabil. Eng. 20, 68 (2012).
104301-13 Lyu et al. Rev. Sci. Instrum. 87, 104301 (2016)

24H. Iwaki, V. Pinskerova, and M. A. R. Freeman, J. Bone Jt. Surg. 82, 1189 30K. M. Lee and J. Guo, in IEEE/ASME International Conference on
(2000). Advanced Intelligent Mechatronics (AIM) (IEEE, 2009), pp. 30–35.
25K.-M. Lee and J. Guo, J. Biomech. 43, 1231 (2010). 31L. W. Tsai, Robot Analysis: The Mechanics of Serial and Parallel Manipu-
26D. Wang, K.-M. Lee, J. Guo, and C.-J. Yang, IEEE/ASME Trans. Mecha- lators (John Wiley & Sons, 1999), pp. 223–259.
tronics 19, 1268 (2014). 32J. Apkarian, S. Naumann, and B. Cairns, J. Biomech. 22, 143
27M. A. Ergin and V. Patoglu, in Proceedings of the IEEE/RSJ International (1989).
Conference on Intelligent Robots and Systems (IROS) (IEEE, 2011), 33D. A. Winter, Biomechanics and Motor Control of Human Gait: Normal,

pp. 4917–4922. Elderly and Pathological (University of Waterloo Press, 1991).


28D. Jin, R. Zhang, H. O. Dimo, R. Wang, and J. Zhang, J. Rehabil. Res. Dev. 34C. Liao, J. Zhang, W. Chen, and M. Lv, in IEEE 10th Conference

40, 39 (2003). on Industrial Electronics and Applications (ICIEA) (IEEE, 2015),


29J. B. Morrison, J. Biomech. 3, 51 (1970). pp. 557–562.

S-ar putea să vă placă și