Aliakbar Alamdari ∗
Graduate Research Assistant, Student Member of ASME
d
Mechanical and Aerospace Engineering
ite
SUNY at Buffalo
Buffalo, NY, 14260
aalamdar@buffalo.edu
ed
py
Venkat Krovi
Professor, ASME Fellow
Mechanical and Aerospace Engineering
Co
SUNY at Buffalo
Buffalo, NY, 14260
vkrovi@buffalo.edu
ot
tN
Assisted motor therapies play a critical role in enhancing 1 INTRODUCTION
functional musculoskeletal recovery and neurological reha-
Several neurological disorders including stroke, spinal
ip
proach is employed to select configuration parameters in- pists. The difficulty and inconsistency in therapy from one
cluding cuff sizes, cuff locations, and the position of fixed session to the next motivated researchers to develop gait
winches. The desired motions for the rehabilitative exer- training treadmills with body weight support to provide con-
pt
cises are prescribed based upon motion patterns from a nor- sistent gait motion [2–4]. Our interest is in rehabilitation sys-
mative subject cohort. We examine the viability of using tems which allow for significant greater flexibility of thera-
ce
two controllers —a joint-space feedback linearized PD con- peutic motions/forces as well as customization of the train-
troller and a task-space force-control strategy —to realize ing regimen. Volpe et al. [5] review the successes noted
Ac
trajectory- and path- tracking of the desired motions within from several groups in reducing impairment and enhanc-
a simulation environment. In particular, we examine per- ing human motor control with task specific exercises de-
formance in terms of (i) coordinated control of the redun- livered by robotic devices. It is worth noting that several
dant system; (ii) reducing internal stresses within the lower- exoskeleton systems such as BLEEX [6], MIT exoskeleton
extremity joints; and (iii) continued satisfaction of the uni- [7], HAL-3 [8] have been developed (in non-rehabilitation
lateral cable-tension constraints throughout the workspace. contexts) to augment human walking. Additionally, many
active/passive orthoses such as gravity-balancing leg ortho-
sis [9], pneumatic-muscle powered ankle-foot orthosis [10],
elastic knee orthosis [11], ALEX [12], and LOPES [13] have
∗ Corresponding author. been developed.
Alamdari 1 JMR-15-1266
d
6 is satisfaction of the tensionability conditions, i.e., assuring
that all cables are always in tension. The manipulation is
ite
Fig. 1. RObotic Physical Exercise and System (ROPES): A Cable- performed via increasing and decreasing the lengths of the
Driven Robotic Rehabilitation System for Lower-Extremity. Motors 1, cables connected to the end-effector. One can increase the
ed
2, 3 and 4 are placed in appropriate positions to generate positive number of redundant cables to satisfy the tensionability of
cable tensions to move lower limbs in the sagittal plane along the the cables. However, this increases the interference of the
desired trajectory, and likewise motors 5, 6 and 7 are placed in frontal cables in the working space. The minimum sufficient number
py
plane to generate positive cable tensions based upon the prescribed of cables in mulibody systems that guarantees all cables are
lateral exercises for lower limbs. in tension have been investigated in [18].
In traditional cable robots, multiple cables are at-
Co
tached to single-payload/platform from multiple points on
Several of these architectures are now being evaluated the ground. In contrast, in the proposed articulated multi-
for rehabilitation application potential to help patients to body cable-robot system, cables are connected to different
ot
overcome muscle-weakness and remedy abnormal motor links of the multibody system from multiple points on the
control etc. The challenge arises both for design of robotic ground as illustrated in Fig. 1. In this work, we assume a
tN
system with adequate degrees of freedom to not hinder natu- simplified model wherein each human lower limb segment
ral gait kinematics/dynamics as well as coordinated control (i.e. foot, shank and thigh) and its corresponding lorthosis-
to achieve the normal physiological gait patterns. Among link are considered as one part. Thus, each human lower
ip
clinically-deployed rehabilitation systems, Lokomat [14] is limbs maybe considered as a serial multibody system driven
an exoskeletal orthosis which is attached to a patient’s legs by the cables attached to them.
r
to assist the person to walk on the treadmill. The Haptic Limited literature has examined multi-body articulated
sc
Walker [15] is a multidegree-of-freedom system intended to cable-drive systems in the past. For example, the wrench
generates foot pedal motions to simulate regular and stair closure workspace of multibody cable-driven mechanisms is
walking. However, in addition to performance limitations, determined based on Lagrange’s approach [18] and recipro-
nu
these devices are expensive and only available in some clini- cal screw theory [19]. The concept of generalized forces and
cal or rehabilitation centers. Lagrange’s method have been employed to eliminate forces
Ma
Hence, in this paper, we examine the feasibility of a and moments from the equilibrium equations. It is notewor-
light-weight and reconfigurable hybrid (articulated multi- thy that the operational workspace of multibody cable-driven
body and cable) robotic system for assisting lower-extremity systems is largely impacted by the choice of cable placement
rehabilitation and analyze its performance. Many of the com- and routing. Bryson and Agrawal [20] identified and ana-
ed
monly prescribed rehabilitation exercises engender closed- lyzed cable configurations for serial robot driven by cables.
loop ankle trajectories which can be realized by an appro- Yan et al. [21] designed a 7-DOF humanoid arm driven with
priate cable-driven rehabilitation device. Such closed-loop 14 cables. They employed the force closure method to evalu-
pt
ankle trajectories can serve to guide the other leg-members ate the workspace of multi-finger grasping. In [22, 23], a hy-
(shank/thigh) via the natural kinematics and dynamics of the brid articulated-cable parallel mechanism named PACER de-
ce
limbs. Cable robots are well-known for their low inertia, rel- veloped by authors for upper limb rehabilitation in 3D space
atively large workspace, low fabrication costs and reconfig- in which with appropriate design, and proper selection of the
Ac
urability. The use of cables allows the relatively heavy mo- position of cable winches and cable attachment points to the
tors and gearboxes to be moved from the joints to the base. multibody system, the positive tension in the cables assured.
This reduces the mass and inertia of the moving bodies and The key contribution of this paper is the development
can allow the robot to be designed with smaller, less costly and feasibility/performance evaluation of a reconfigurable
motors. hybrid cable-articulated architecture that will work closely
Several recent exoskeleton designs have emerged to take with the human musculoskeletal system to provide motor
advantage of the cable-based architecture. In [16] a cable- therapies. This system is an extension of traditional single-
driven robotic gait training system called CaLT was designed body cable-driven to multibody cable-driven system. The
for gait training of spinal cord injury patients, and it deliv- combined system features multiple holonomic cable-loop-
ered a promising and acceptable experimental results. In this closure constraints acting on a tree-structured multibody
Alamdari 2 JMR-15-1266
d
ically, the focus of this paper is to assist the performance C1 C2
of repetitive therapy of human lower limbs in sagittal plane. d 3y
ite
The desired motion is prescribed based upon normative sub- y2
jects’ motion patterns. The appropriate coordination of ca- T1 t1 ,l1 T2 t 2 ,l 2
ed
ble forces to each segment of the lower limbs is realized by
p
an impedance/force-field and a feedback linearized PD con- d1y d 2y x2 2
+ qa
py
increasing the internal forces at the lower extremity joints.
The remainder of the paper is organized as follows: In
Fig. 2. A cable-driven robotic rehabilitation system, in which Ti , ti
Section 2, the kinematic analysis of ROPES is presented.
Co
are cable tension and cable unit vector, respectively; diy and dix are
The dynamic equations of motion of ROPES using the
cuff size and its position in the local frame; Kti is tensional spring for
Newton-Euler, and Lagrangian approach are derived in both
increasing the wrench feasible workspace.
joint-space and task-space in Section 3. In Section 4, the
ot
optimal cable configuration analysis of ROPES is identified,
and then the tension distribution in the resulting optimal con- gles as shown in Fig. 2), and JT = tTi ∂∂uqi is a Jacobian matrix
tN
figuration is examined in Section 5, and Section 6 and 7 are which maps joint angular velocities to the cable velocities.
devoted to designing feedback-linearized PD controller and The angular velocity and acceleration of each segment
impedance/force-field controller for ROPES, and finally Sec- i.e., thigh, shank and foot in sagittal plane can be found such
ip
tion 8 concludes the paper. that for thigh is, ωt = q̇1 , ω̇t = q̈1 , and linear acceleration
of the mass center of thigh is act = ω̇t × lct + ωt × (ωt × lct ),
2 KINEMATIC ANALYSIS OF ROPES
r
shank can be expressed by, ωs = q̇1 + q̇2 , ω̇s = q̈1 + q̈2 , and
In these relations wi denotes the winch position respect to the
linear acceleration of the mass center of shank is acs = at +
fixed frame {F}, and ui denotes the position vector of cable
ω̇s × lcs + ωs × (ωs × lcs ), where lcs = F R2 2 lcs is the link (shank)
attachment point to orthosis respect to the fixed frame {F}.
mass center vector expressed in fixed frame; 2 lcs is the shank
ed
(3) Finally for the foot, the angular velocity and accelera-
tion can be written as, ω f = q̇1 + q̇2 + q̇3 , ω̇ f = q̈1 + q̈2 + q̈3 ,
and linear acceleration is acf = as + ω̇ f × lcf + ω f × (ω f × lcf ),
Ac
Alamdari 3 JMR-15-1266
d
is placed on a circle with center of C0 , of radius R, and angle their devices based on the average kinetic and kinematic
θi , i = {1, 2, 3, 4} respect to the fixed coordinate frame. data of human [26, 27]. The compliance of lower extrem-
ite
ity joints during locomotion can be investigated by the con-
cept of quasi-stiffness. This term can be distinguished from
Table 1. Design parameter ranges of independent parameters of
ed
the passive and active stiffness of a joint typically used to
ROPES. describe the local tangent to the moment-angle curve shown
Parameter Range Units Parameter Range Units for given joint at a specific angle [28]. As illustrated in Fig.
py
3, the quasi-stiffness of a joint is defined more globally, as
d1y [45, 85] mm d2y [50, 85] mm
the slope of the best linear fit on the moment-angle curve of
d3y [55, 95] mm d4y [75, 125] mm a joint over a whole stride or specific phase of a stride [29].
Co
d1x [70, 220] mm d2x [70, 220] mm A series of empirical studies [29–31] have been conducted to
characterize the quasi-stiffness and linear behavior of lower
d3x [110, 440] mm d4x [150, 450] mm extremity joints during walking for adult humans as function
ot
θ1 [170, 195] deg θ2 [260, 330] deg of body size (height and weight). This forms the basis of
a normative anthropometric statistical model which we will
θ3 [−10, 80] deg θ4 [110, 165] deg
tN
employ to predict the hip, knee and ankle quasi-stiffness for
adults walking on level ground. The mean value of quasi-
stiffness of each joint is enumerated in Table 2.
The cable configuration parameters, as described in Fig.
ip
allowed to vary from the 75mm to 125mm in increments of sional springs placed on the orthosis with units of Nm/rad [29–31].
sc
5mm, and its position, d4x , is allowed to range from 150mm Parameter Value
to 450mm in increments of 5mm.
nu
3 DYNAMIC EQUATIONS OF MOTION OF LOWER Knee quasi-stiffness in extension phase KKe 263
LIMBS DRIVEN WITH CABLES
Knee quasi-stiffness in flexion phase KK f 304
Guaranteeing positive cable tensions for the cable-
driven articulated orthosis (within an adequately large work- Ankle quasi-stiffness in plantar-flexion phase KAp 202
ed
region) remains a critical issue. Hence the dynamic equa- Ankle quasi-stiffness in dorsi-flexion phase KAd 246
tions of human lower limb (orthosis + human leg) were de-
veloped to facilitate model-based control of the tension dur- Hip stiffness on the orthosis KOH 327
pt
ing the rehabilitation exercises. The exercises include normal Knee stiffness on the orthosis KOK 283
walking in the sagittal plane and lateral leg-lifting in frontal
ce
Alamdari 4 JMR-15-1266
10 -20
K He
60
K Ap 0 -40 T3
40
K Ad -10 K Kf -60 Fs f
20
t3 −ts f tts 0 0
Fts (6)
-20 -80
us3 × t3 −us2 × ts f us1 × tts −ms f mts Ms f
0 -30 -100
0 0.4 0.8 1.2 0 0.4 0.8 1.2 -0.3 -0.1 0.1 0.3 0.5
Ankle Angle (rad) Knee Angle (rad) Hip Angle (rad)
Mts
d
Fig. 3. Ankle, knee, and hip moment vs. angle curve for repre- ms as − ms g
=
ite
sentative subject walking at 1.25 m/s. Quasi-stiffness is claculated Is ω̇s + [ωs ×]Is ωs + Jk − Ja
based on the slope of the best-line fit to the moment-angle curve for
ankle plantar-flexion (KAp ), ankle dorsi-flexion (KAd ), knee flexion
ed
where Jk = (KKe/K f + KOK )q2 − st2 × Kt2 lt2 , Is =
(KK f ), knee extension (KKe ), and hip extension (KHe ) and flexion
F R 2 I F RT ; t is the unit vector along the cable with
(KH f ) [29–31]. 2 s 2 3
magnitude T3 ; tts and mts are the unit vectors of force Fts
py
and moment Ms f exerted from thigh to the shank; ms is the
very useful in a design/simulation setting in helping analyze mass of the shank; usi , i = {1, 2, 3} are the position vector
Co
force-profiles (both internal forces/moments within as well from the mass center of the shank to the point of application
as the external forces needed to drive) the cable-articulated of the applied force expressed in the fixed frame; KKe/K f is
orthosis. Each segment’s dynamic equations, subjected to quasi-stiffness of knee which depends on the stance phase
cable-and constraint-forces, is derived separately. Note that during walking, for example, KKe/K f = KK f in flexion
ot
each segment is subjected to cable forces as well as con- phase, KKe/K f = KKe in extension phase; both inertia tensor
straint forces and moments. Since the weight and moment of Is and the angular velocity ωs of the foot are expressed in
tN
inertia of the orthosis segments and cuffs are much smaller fixed frame of reference, and finally Kt2 and lt2 are zero-free
than lower limbs segments, their weight and moment of in- length tensional spring stiffness and vector along the spring,
ertia are ignored in this simulation . The recursive Newton- respectively; st2 is the position vector from the mass center
ip
Euler formulation in the sagittal plane is written for the foot of the shank to the spring attachment point on the orthosis;
segment in matrix form as follows (notice that boldface sym- the generated force by the linear axial springs is equal to
r
bols have been used for vectors and matrices), Kt2 lt2 which lt2 is the vector of zero-free length spring. This
sc
u f 1 × t1 u f 2 × t2 u f 3 × ts f ms f Fs f
Ms f t4 −tts tbt 0 0
Jt = (7)
ut4 × t4 −ut2 × tts ut3 × tbt −mts mbt
m f af − Fext − m f g
= T
If ω̇ f + [ω f ×]If ω f − Mext − u f 4 × Fext + Ja Tt = T4 Fts Fbt Mts Mbt
ed
mt at − mt g
Ft =
where Ja = KAP/Ad (q3 − 3π It ω̇t + [ωt ×]It ωt − Jk + Jh
2 )+KOA (q3 −Φ3 ); Φ3 is the initial
pt
along the cable expressed in fixed frame and Ti is the mag- where Jh = KHe/H f (q1 − 3π 2 ) + KOH (q1 − Φ1 ) − st1 × Kt1 lt1 ,
nitude of cable forces; ts f and ms f are unit vectors of force It = F R1 1 It F RT1 ; mt is the mass of the thigh; t4 is the unit
Ac
Fs f and moment Ms f exerted from shank to the foot in sagit- vector along the cable with magnitude T4 ; tbt and mbt are
tal plane; Fext and Mext are given external forces and mo- the unit vectors of force Fbt and moment Mbt exerted from
ments from the ground to the foot; m f is the mass of the the human body to the thigh; uti , i = {2, 3, 4} is the position
foot; u f i , i = {1, 2, 3} is the position vector from the mass vector from the mass center of the thigh to the point of ap-
center of the foot to the point of application of the applied plication of the applied force expressed in the fixed frame;
forces expressed in the fixed frame. Notice that the iner- KHe/H f is quasi-stiffness of hip which depends on the stance
tia tensor If of the foot is expressed in fixed frame of refer- phase during walking, for example, KHe/H f = KH f in flexion
ence i.e., If = F R3 3 If F RT3 ; KAP/Ad is quasi-stiffness of ankle phase, KHe/H f = KHe in extension phase; Φ1 is the initial an-
which depends on the stance phase during walking, for ex- gular position of the torsion spring installed at the hip joint;
ample, KAP/Ad = KAP in plantar-flexion phase, KAP/Ad = KAd st1 is the position vector from the mass center of the thigh to
Alamdari 5 JMR-15-1266
d
where where
ite
1 2
J f J f 0 0 J3f 0 0 J4f 0 0
T ∂ u1 T ∂ u2 T ∂ u3 T ∂ u4
t1 ∂ q t2 ∂ q t3 ∂ q t4 ∂ q
J = 0 0 J1s 0 J2s J3s 0 J4s J5s 0
ed
T T ∂ u11 T ∂ u21 T ∂ u13 T ∂ u41
0 0 0 Jt1 0 Jt2 Jt3 0 Jt4 Jt5 JT = t1 ∂ q t2 ∂ q t3 ∂ q t4 ∂ q
2 2 2 2
T tT1 ∂∂ qu1 tT2 ∂∂ uq2 tT3 ∂∂ uq3 tT4 ∂∂ uq4
T = T1 T2 T3 T4 Fs f Fts Fbt Ms f Mts Mbt
py
3 3 3 3
T
F = FTf FTs FtT
where JTT T is the part of generalized force related to cable
Co
forces and JTe Fe includes generalized external forces and mo-
where Jif , Jis and Jti are the ith column of the J f , Js and Jt , ments (Fe = [Fext , Mext ]T ).
respectively, defined in Eqns. (5), (6) and (7). Similarly, This form is well suited for trajectory tracking control
dynamic equations of each segment of human leg can be applications where the desired end-effector trajectory is pre-
ot
derived separately for the orthosis in the frontal plane (but sented in terms of joint angles, velocities, and accelerations
omitted here for brevity [32]). for which a feedback linearization controller is developed.
tN
3.2 Closed-form dynamic formulation in joint-space 3.3 Closed-form dynamic formulation in task-space
ip
The Lagrangian formulation is developed to reduce the However, in other circumstances (e.g. development of
size of structure matrix J in Eqn. (8). The general form of an impedance controller) we may wish to move the ankle
r
dynamic equations for the cable-driven system in joint-space on a trajectory in Cartesian space without converting task-
sc
can be written as space variables. In such cases, the joint-space dynamics can
be projected into the task-space to realize the (often lower-
M(q)q̈ + V(q̇, q) + G(q) = Q (9) order) task-space dynamics equations as follows
nu
olis forces), and G = [G1 , G2 , G3 ]T is combination of gravi- It is noteworthy that although the equations are ex-
tational forces and spring forces. pressed in task-space, still some terms such as Ṽ, G̃, and Q̃
are written as function of joint variables q. Due to nonlinear-
pt
Alamdari 6 JMR-15-1266
d
vector, T, yields the following Eqn. (13);
ite
20 1 m/s
1.5 m/s
40
10
ed
20
T
−10
−20 0
0 1 2 3 0 1 2 3
where (JTT )# is the Moore-Penrose pseudo-inverse of matrix
py
Time (sec) Time (sec)
Co
T 40 1.5 m/s
1.5 m/s 1
For this case, since JTT is 3 × 4, then ηJT is 4 × 1 vector.
T 20
The λ = 0 case of Eqn. (13) is the minimum norm solution 0.5
0
but the tension vector components are not guaranteed to be
ot
non-negative. −20
0 1 2 3
0
0 1 2 3
If all elements of the null vector, ηJT , have the same Time (sec) Time (sec)
T
tN
sign, then it can be seen that regardless of minimum norm
solution values a λ can be chosen such that the ηJT term Fig. 4. Hip, knee, ankle and normalized ground reaction forces of
T
be more than particular solution and gives a tension solution healthy subject during walking with different speeds. These values
ip
which all components are positive. Here, we calculate the are considered as desired angles and forces, in trajectory tracking
wrench-closure workspace by exploiting this characteristic problem [38].
r
(i.e., if all elements of the null vector are the same sign, then
sc
that cable tensions are greater than some prescribed mini- Configurations which satisfy the wrench feasibility con-
mum and less than some prescribed maximum values. This straints are accumulated in database of desired configura-
workspace depends not only on geometric parameters, but tions.
also on the allowable tension ranges, gravitational effects, To visualize the trends of these high-dimensional data-
ed
and the required wrench set. set, each possible parameter value is shown in separate x-axis
The choice of cable attachment both on fixed frame and on the plot, in a range defined for parameter values in Table 1,
and the number of wrench feasible configurations including
pt
Alamdari 7 JMR-15-1266
Number of configurations
lf 243.3 mm mt 11.8 kg 280 d4y
220
(I f )z 30.4 kg.cm2 uf3 117 mm
200
us1 183 mm us2 283 mm
180
ut2 262 mm ut3 210 mm
d
160
40 50 60 70 80 90 100 110 120 130
ite
diy (mm)
to 260◦ . As illustrated in Fig. 6, the trend for d1y indicates Fig. 6. The trend of each cuff radius.
an evidence that the cuff radius is better to be chosen with
ed
smaller size, here 56mm is selected for d1y . The trend for d2y
suggest bigger size for second cuff radius. Accordingly, the 600
py
θ1
trends for cuffs radius, d3y and d4y , suggests that there are θ2
500
better to be selected around 65mm and 115mm, respectively. 400
θ3
Similarly, the preferred values for winch positions are shown θ4
Number of configurations
Co
200
400
in Fig. 7.
0
These initial trends indicate that there is an opportunity 300 165 190
to further optimize the performance of the ROPES system.
ot
For example, the d2y trend indicates there might be value in 200
red arrows).
sc
700
some function of the cable tensions. In [34], it was shown
d1x
600 100
d2x
that while the L∞ -norm provides optimal solutions, they may
Ma
500
50 d4x tion can, however, be approximated using a p-norm (1 < p <
400 ∞), and the resulting minimum-norm solution is proven to be
0
180 220 260 300 340 380 420 unique and continuous except at singular configurations.
ed
300
In [40], a discussion on the potential limitations of the
200 more traditional L1 and L2 norms is provided in the context
of tension distribution. Use of the L1 -norm results in a linear
pt
100
programming problem which seeks to minimize (or maxi-
mize) the sum of the tensions. The primary drawback of
ce
0
60 100 140 180 220 260 300 340 380 420 460
d (mm)
ix
this approach is that it is susceptible to discontinuities, as
the optimal operating point may jump from one vertex of the
Ac
Fig. 5. The distance of each cuff from the local frame origin as feasible polyhedron to another between successive computa-
shown in Fig. 2. tions. This can potentially excite high-frequency modes and
degrade the stability of the system. The L2 -norm, which re-
sults in a quadratic programming problem and seeks to min-
imize (or maximize) the sum of the squared tensions, im-
5 TENSION DISTRIBUTION IN ROPES proves upon this limitation by providing a smooth objective
A variety of tension distribution algorithms have been function.
proposed for resolving the actuation redundancy present in Both methods, however, result in tensions that fre-
most cable-driven robots. Each of these approaches have quently lie on the lower or upper tension limits. Opera-
different characteristics and varying computational cost [39]. tion at the lower tension limits increases the risk of a cable
Alamdari 8 JMR-15-1266
d
In this controller, the desired trajectory in terms of gen-
This equation at each instant of time is a set of linear equa- eralized coordinates is defined as a function of time qd =
ite
tions in terms of T. The size of Jacobian matrix JTT is 3 × 4 qd (t). The block diagram of the proposed control law for
which indicates that the number of unknowns is one more trajectory tracking system is shown in Fig. 8. Hence, from
ed
than the the number of equations. If the number of unknowns closed-form dynamic equations (Eqn. 9) one can write the
was equal to the number of equations, there would be one and virtual control law as
only one solution for each cable.
py
While a variety of redundancy resolution techniques
have been proposed, one approach commonly used is based τ1 = M(q) (q̈d + Kd q̇e + K p qe ) + V(q̇, q) + G(q) (15)
on minimizing the norm of the cable tensions. This has
Co
the beneficial effect of minimizing the energy and torque re- This law linearizes the equations to an exponentially
quirements of the system. The optimization problem can be stable system, q̈ + K q̇ + K q = 0, where q = q − q,
e d e p e e d
formulated as: K = diag{K , K , K }, and K = diag{K , K , K }
p p1 p2 p3 d d1 d2 d3
ot
are positive matrices. In this simulation the controller gain
minkTk p (14) coefficients are selected to be K pi = 125 and Kdi = 15, and
tN
user-determined allowable minimum and maximum cable
tensions are chosen to be 2 and 80N, respectively.
subject to: JTT T = Q − JTe Fe and Tmin ≤ Ti ≤ Tmax , i = As previously noted, there are more cables than the DOF
ip
{1, 2, 3, 4}. Thus, the optimized cable tensions must satisfy of orthosis, therefore there are many solutions for cable ten-
the dynamic equilibrium equations and remain within some sions. As we discussed in cable tension distribution section,
r
specified upper and lower bounds. In general, the lower limit it is desirable to find the set of cable tensions with smaller
sc
corresponds to the amount of tension required to keep the positive values. This can be solved by quadratic program-
cables taut, while the upper bound depends on the torque ca- ming approach. Then, the cable force distribution can be
pacity of the motors and/or the failure point of the cables.
nu
t1 t2 + Orthosis
velopment of rehabilitative robots and motor therapy. In - +
[41], control strategies have been categorized in four groups. V(q,q),G(q) IMU
Compensation sensors
The first group is assisting controllers which move the pa-
pt
q
tients injured limbs in a predefined trajectory to stretch the
limb muscles and rebuild the human motor control system. Fig. 8. Block diagram of trajectory tracking controller for human user
ce
Hip angle(deg)
20
L L L L shown their capabilities for providing compliant interaction 1 2 3 4
0
1.4 with the human limbs. Here, we present the development of
−20
0 0.5 1 1.5 2 2.5 3 1.2
impedance and force-field control for human lower limbs.
Knee angle(deg)
60
40
target path. Let P be the current position of ankle in a Carte-
20
0.4 sian reference frame attached to subject’s trunk. The more
0
distance between the current position and desired position,
d
−20 0.2
0 0.5 1 1.5 2 2.5 3
Time(sec)
0 0.5 1 1.5
Time (sec)
2 2.5 3
the more forces will be applied to thigh and shank to bring
ite
back ankle point toward the target path in sagittal plane. The
Fig. 9. Hip, knee, ankle, and cable length variations during a gait impedance controller tries to control the lower limb such that
cycle. against an external force it acts as a mass-spring-damper sys-
ed
tem. So,
80
py
T1 T2 T3 T4 Fd = Ka P̈d − P̈ + Kv Ṗd − Ṗ + Kx (Pd − P) (16)
70
Cable Tension Forces (N)
60
where Ka = diag{Ka1 , Ka2 }, Kv = diag{Kv1 , Kv2 }, and Kx =
Co
50 diag{Kx1 , Kx2 } are impedance gains matrices, P is the cur-
40
rent position of the ankle, and Pd represents the desired po-
sition.
ot
30 Similarly, if foot angle deviates from the desired foot
20 angle trajectory q3d , a torque τd at the ankle joint will bring
tN
the foot towards the desired trajectory, so one can define τd
10
as
0
0 0.5 1 1.5 2 2.5 3
ip
Time (sec)
τd = Kfα (q̈3d − q̈3 ) + Kfω (q̇3d − q̇3 ) + Kfθ (q3d − q3 )
(17)
r
rent foot angle, and q3d represents the desired foot angle. In
Reaction forces/moments at the hip, knee and ankle joints this simulation, the parameters are defined in Table 4.
nu
Forces at the Ankle (N)
40
Anterior and posterior forces 60 As illustrated in the block diagram Fig. 12, the haptic in-
Superior and inferior forces
40 terface in impedance controller design consists of two major
20 20
components, hardware and software. The hardware consists
Ma
0
0 −20 of motors, load cells, orthosis and IMU sensors, and is simu-
−40
0 1 2 3 0 1 2 3 lated for purposes of the current study. The software compo-
200 nent consists of forward kinematics, gravity compensation,
Forces at the Hip (N)
100
0
Matlab/Simulink. The ankle point position (or foot angle)
0
−20 sent from Simulink to virtual reality will be compared to
pt
−100
0 1 2 3 −40 desired target Pd (or desired angle q3d ), and then the de-
0 1 2 3
Time (sec) sired force Fd (or torque τd ) will be generated. Impedance
ce
Alamdari 10 JMR-15-1266
d
to Kt inside the tunnel.
ite
Fig. 12. A block diagram for impedance control of human user lower Magnitude of force−field around the desired path
limbs by creating a virtual force to the ankle point to move it along the −0.75
target path.
ed
−0.8
py
Parameter Value Parameter Value Parameter Value −0.85
Y(m)
Kai 1 Kvi 10 Kxi 25
Co
−0.9
Kfα 0.5 Kfω 4 Kfθ 8
δn 8 δt 3 rn 8 −0.95
rt 2 Kn 20 Kt 10
ot
ζn 20 ζt 10 −1
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
X(m)
tN
point target path, and along the desired foot angle trajectory. Fig. 13. The magnitude of force-field (black arrows) around the an-
If the ankle point (or foot angle) deviate from the target path kle path (red line), and the stream of forces around the path (blue
ip
(or foot desired angle trajectory), the controller acts as spring curve lines).
and brings them back to target.
r
sc
The high level force-field controller generates: (i) a Foot angle variation along the traget path
force Fd at the ankle point that has a normal (Fn ) and tan-
gential (Ft ) components, i.e., Fd = Fn + Ft , (ii) a torque vec-
nu
Target path
Foot angle (deg)
20
target path (or desired joint angle trajectory), and tangential
10
components i.e. Ft and τt are tangential to target path/desired
0
joint angle trajectory, and assist in tracking the target path (or 0.2
−10 0.1
ed
2q
2q
−( min )2 −( min )2
kτn k = ζn 1 − e δn , kτt k = ζt e δt (19) Fig. 14. The magnitude of foot angle (cyan line) on the target path
(red line), and upper and lower bound around the foot angle (blue
Ac
curve lines).
where Kn and ζn are gain vectors for normal force-field, and
Kt and ζt are gain vectors for tangential force-field, respec-
tively; dmin is minimum distance from the ankle point to a 8 Discussion
point on the target path, and qmin is the distance between the In this paper we examined various aspects of the mod-
current foot angle and desired joint angle. eling, analysis and simulation of an articulated-multibody
Two virtual force-field tunnels with diameters rn and rt cable-driven system for rehabilitative exercises on lower ex-
are created around the target path in task-space, and sim- tremity. The sagittal-plane dynamic model of lower limbs
ilarly two force-field tunnels with diameters δn and δt are with 3-DOF was formulated based on the both Newton-Euler
created around the desired joint angle trajectory in the joint and Lagrangian formulations to support the design/control
Alamdari 11 JMR-15-1266
d
frame. This analysis helped to identify not only the most- hal-3 for gait disorder person”. Computers helping people
sensitive configurations of ROPES but also the most-robust with special needs, pp. 19–29.
ite
(with respect to assembly- and set-up errors). We highlight [9] Banala, S. K., Agrawal, S. K., Fattah, A., Krishnamoor-
the opportunity and intend to pursue this as one aspect of thy, V., Hsu, W.-L., Scholz, J., and Rudolph, K., 2006.
“Gravity-balancing leg orthosis and its performance evalua-
ed
future work.
tion”. Robotics, IEEE Transactions on, 22(6), pp. 1228–1239.
For the subsequent control analysis, we down-selected
[10] Ferris, D. P., Czerniecki, J. M., Hannaford, B., of Washington,
one particular configuration (from among the numerous U., and System, V. P. S. H., 2005. “An ankle-foot orthosis
py
high-performance configurations). Two types of controllers powered by artificial pneumatic muscles”. Journal of Applied
were implemented for Lagrangian model, trajectory tracking Biomechanics, 21(2), p. 189.
PD controller in joint-space, and force-control strategies in [11] Cherry, M. S., Choi, D. J., Deng, K. J., Kota, S., and Ferris,
Co
task-space. Simulation results for both controllers were pre- D. P., 2006. “Design and fabrication of an elastic knee ortho-
sented to show how model-based controller can apply forces sis: preliminary results”. In ASME 2006 International De-
through cable-driven system. Finally, using the cable tension sign Engineering Technical Conferences and Computers and
ot
results, and splitting the Jacobian matrix derived in Newton- Information in Engineering Conference, American Society of
Euler formulation, we were able to calculate the internal Mechanical Engineers, pp. 565–573.
[12] Banala, S. K., Kim, S. H., Agrawal, S. K., and Scholz, J. P.,
tN
forces and moments at the hip, knee and ankle joints due
2009. “Robot assisted gait training with active leg exoskele-
to forces through cable-driven systems to avoid increasing of
ton (alex)”. Neural Systems and Rehabilitation Engineering,
internal stresses at lower extremity joints. IEEE Transactions on, 17(1), pp. 2–8.
ip
For future work, there is an opportunity to include mo- [13] Veneman, J., Ekkelenkamp, R., Kruidhof, R., Van der Helm,
bility in fixed bases and add springs in series with cables to F., and Van der Kooij, H., 2005. “Design of a series elastic-
r
(i) generate appropriate pretension in cables without becom- and bowden cable-based actuation system for use as torque-
sc
ing slack, (ii) be able to change the ROPES stiffness, and actuator in exoskeleton-type training”. In Rehabilitation
(iii) increase the safety of mechanism as well. Last but not Robotics, 2005. ICORR 2005. 9th International Conference
least is the need for experimental validation of this overall on, IEEE, pp. 496–499.
nu
framework. [14] Jezernik, S., Colombo, G., Keller, T., Frueh, H., and Morari,
M., 2003. “Robotic orthosis lokomat: A rehabilitation and
research tool”. Neuromodulation: Technology at the neural
Ma
[1] Alamdari, A., Jun, S., Ramsey, D., and Krovi, V., 2016. “A re-
view of home-based robotic rehabilitation”. In Encyclopedia [17] Jin, X., Cui, X., and Agrawal, S. K., 2015. “Design of a
of Medical Robotics. World Scientific. (In press). cable-driven active leg exoskeleton (c-alex) and gait training
ce
[2] Hornby, T. G., Zemon, D. H., and Campbell, D., 2005. experiments with human subjects”. In Robotics and Automa-
“Robotic-assisted, body-weight–supported treadmill training tion (ICRA), 2015 IEEE International Conference on, IEEE,
pp. 5578–5583.
Ac
Alamdari 12 JMR-15-1266
d
neering Technical Conferences and Computers in Engineer- over-constrained cable-driven parallel robots”. In Computa-
ing Conference, August 2-5, American Society of Mechanical tional Kinematics. Springer, pp. 139–146.
ite
Engineers. [40] Borgstrom, P. H., Jordan, B. L., Sukhatme, G. S., Batalin, M.,
[23] Alamdari, A., and Krovi, V., 2015. “Modeling and control Kaiser, W. J., et al., 2009. “Rapid computation of optimally
of a novel home-based cable-driven parallel platform robot: safe tension distributions for parallel cable-driven robots”.
ed
Pacer”. In Intelligent Robots and Systems (IROS), 2015 Robotics, IEEE Transactions on, 25(6), pp. 1271–1281.
IEEE/RSJ International Conference on, IEEE. [41] Marchal-Crespo, L., and Reinkensmeyer, D. J., 2009. “Re-
[24] Bosscher, P., Riechel, A. T., and Ebert-Uphoff, I., 2006. view of control strategies for robotic movement training after
py
“Wrench-feasible workspace generation for cable-driven neurologic injury”. Journal of neuroengineering and rehabil-
robots”. Robotics, IEEE Transactions on, 22(5), pp. 890–902. itation, 6(1), p. 20.
Co
[25] Taghavi, A., Behzadipour, S., Khalilinasab, N., and Zohoor, [42] Perez, M. A., Lungholt, B. K., Nyborg, K., and Nielsen, J. B.,
H., 2013. “Workspace improvement of two-link cable-driven 2004. “Motor skill training induces changes in the excitabil-
mechanisms with spring cable”. In Cable-Driven Parallel ity of the leg cortical area in healthy humans”. Experimental
Robots. Springer, pp. 201–213. Brain Research, 159(2), pp. 197–205.
ot
[26] Collins, S. H., and Kuo, A. D., 2010. “Recycling energy to [43] Reinkensmeyer, D. J., Kahn, L. E., Averbuch, M., McKenna-
restore impaired ankle function during human walking”. PLoS Cole, A., Schmit, B. D., and Rymer, W. Z., 2000. “Under-
one, 5(2), p. e9307. standing and treating arm movement impairment after chronic
tN
[27] Sup, F., Varol, H. A., Mitchell, J., Withrow, T., and Goldfarb, brain injury: progress with the arm guide”. Journal of reha-
M., 2008. “Design and control of an active electrical knee and bilitation research and development, 37(6), pp. 653–662.
ankle prosthesis”. In Biomedical Robotics and Biomechatron- [44] Yano, H., Kasai, K., Saitou, H., and Iwata, H., 2003. “De-
ip
ics, 2008. BioRob 2008. 2nd IEEE RAS & EMBS Interna- velopment of a gait rehabilitation system using a locomotion
tional Conference on, IEEE, pp. 523–528. interface”. The Journal of Visualization and Computer Ani-
r
[28] Frigo, C., Crenna, P., and Jensen, L., 1996. “Moment-angle mation, 14(5), pp. 243–252.
sc
relationship at lower limb joints during human walking at dif- [45] Hesse, S., Schmidt, H., and Werner, C., 2006. “Machines to
ferent velocities”. Journal of Electromyography and Kinesi- support motor rehabilitation after stroke: 10 years of experi-
ology, 6(3), pp. 177–190. ence in berlin”. Journal of rehabilitation research and devel-
nu
[29] Shamaei, K., Sawicki, G. S., and Dollar, A. M., 2013. “Es- opment, 43(5), p. 671.
timation of quasi-stiffness and propulsive work of the hu- [46] Ekkelenkamp, R., Veltink, P., Stramigioli, S., and van der
man ankle in the stance phase of walking”. PloS one, 8(3), Kooij, H., 2007. “Evaluation of a virtual model control for
Ma
[31] Shamaei, K., Sawicki, G. S., and Dollar, A. M., 2013. “Es- ankle rehabilitation robot for various exercises”. Journal of
timation of quasi-stiffness of the human knee in the stance Robotic Systems, 22(S1), pp. S15–S33.
phase of walking”. PloS one, 8(3), p. e59993. [48] Lam, T., Wirz, M., Lünenburger, L., and Dietz, V., 2008.
pt
[32] Alamdari, A., and Krovi, V., 2015. “Robotic physical exercise “Swing phase resistance enhances flexor muscle activity dur-
and system (ROPES): A cable-driven robotic rehabilitation ing treadmill locomotion in incomplete spinal cord injury”.
ce
system for lower-extremity motor therapy”. In Proceedings Neurorehabilitation and Neural Repair, 22(5), pp. 438–446.
of the ASME 2015 International Design Engineering Techni- [49] Boian, R. F., Deutsch, J. E., Lee, C. S., Burdea, G. C., and
cal Conferences and Computers in Engineering Conference, Lewis, J., 2003. “Haptic effects for virtual reality-based post-
Ac
August 2-5, American Society of Mechanical Engineers. stroke rehabilitation”. In Haptic Interfaces for Virtual Envi-
[33] Gouttefarde, M., and Gosselin, C. M., 2006. “Analysis ronment and Teleoperator Systems, 2003. HAPTICS 2003.
of the wrench-closure workspace of planar parallel cable- Proceedings. 11th Symposium on, IEEE, pp. 247–253.
driven mechanisms”. Robotics, IEEE Transactions on, 22(3), [50] Huang, V. S., Shadmehr, R., and Diedrichsen, J., 2008. “Ac-
pp. 434–445. tive learning: learning a motor skill without a coach”. Journal
[34] Verhoeven, R., and Hiller, M., 2002. “Tension distribution in of neurophysiology, 100(2), pp. 879–887.
tendon-based stewart platforms”. In Advances in Robot Kine- [51] Agarwal, P., and Deshpande, A. D., 2015. “Impedance
matics. Springer, pp. 117–124. and force-field control of the index finger module of a hand
[35] Lim, W. B., Yang, G., Yeo, S. H., and Mustafa, S. K., 2011. “A exoskeleton for rehabilitation”. In Rehabilitation Robotics
generic force-closure analysis algorithm for cable-driven par- (ICORR), 2015 IEEE International Conference on, IEEE,
Alamdari 13 JMR-15-1266
d
ite
ed
py
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
Alamdari 14 JMR-15-1266
d
5 The distance of each cuff from the local frame origin as shown in Fig. 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
6 The trend of each cuff radius. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
ite
7 The trend of cable placements on the fixed frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
8 Block diagram of trajectory tracking controller for human user lower limbs. . . . . . . . . . . . . . . . . . . . . . . . . . 9
9 Hip, knee, ankle, and cable length variations during a gait cycle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
ed
10 Cable tension forces in ROPES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
11 Internal forces/moments at the lower extremity joints due to cable tensions. . . . . . . . . . . . . . . . . . . . . . . . . . 10
12 A block diagram for impedance control of human user lower limbs by creating a virtual force to the ankle point to move it
py
along the target path. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
13 The magnitude of force-field (black arrows) around the ankle path (red line), and the stream of forces around the path (blue
Co
curve lines). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
14 The magnitude of foot angle (cyan line) on the target path (red line), and upper and lower bound around the foot angle
(blue curve lines). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
ot
List of Tables
tN
1 Design parameter ranges of independent parameters of ROPES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Quasi-stiffness of lower extremity joints and stiffness of torsional springs placed on the orthosis with units of Nm/rad [29–31]. 4
3 Anthropometry and mass distribution for human body [37] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 Parameters/gains of impedance and force-field controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
r ip
sc
nu
Ma
ed
pt
ce
Ac
Alamdari 15 JMR-15-1266