Study of Stability and Robustness For A Force Augmenting Device

1
Study of Stability and Robustness of a Human

Operating a Force Augmenting Device considering
Time-Delays in the Human Model
Suresh K. Gadi, Ruben A. Garrido, Rogelio Lozano and Antonio Osorio
AbstractIn this paper a force augmenting device (FAD), a

human operator and their interaction are modeled and a stability
analysis of the whole system is presented. It is proved that the
interaction is stable for all the values of the augmenting factor,
if the time-delays in the human model are negligible. A method
based on the Rekasius substitution is applied to find an upper
bound of the time-delays in the human model to preserve stability.
A laboratory prototype is designed for testing the performance
of the system. Numerical simulations are presented using the
estimated values of the experimental prototype.
In order to study the stability of the system in presence of
delays, we first consider that the time-delays in the position and
the velocity measurements for a human operator are equal. An
upper bound for the time-delay of 0.16 s has been computed for
the prototype when using an augmenting factor of 125. Previous
studies show that for a healthy human the time-delays are around
0.04 s, which makes the presented FAD stable for a wide range of
human operators. This paper also presents a method for tuning
the control parameters for obtaining the best performance for
a given FAD. An experiment is performed with the laboratory
prototype, where a human operator lift a load. It is observed that
the human machine interaction is stable and the human operator
is able to move the load to its desired position by experiencing
very little effort. Also, the FAD was experimented with human
operators of four different age groups to verify the robustness of
the control algorithm; the results were found satisfactory.
Index TermsForce augmenting device, time-delay, human
model, human-machine interaction
I. I NTRODUCTION
Recently, there has been a lot of interest in developing
exoskeletons and force augmenting devices[1], i.e. mechanical
systems that are worn by humans in such a way as to increase
their force. There exist many possible applications for these
devices; for instance, in industries where it is commonly
required to move heavy loads. This task is mostly done by
machines, which are usually controlled by humans who do
not necessarily feel the force exerted on the load. This may
lead to unsafe operations in confined spaces with obstacles.
Exoskeletons and force augmenting devices (FAD) represent a
possible solution to this problem [2]. Indeed a FAD amplifies
the human strength and allows the operator handling heavy
S. K. Gadi and R. A. Garrido are with the Department of Automatic
Control, UMI LAFMIA CNRS-CINVESTAV, Av. IPN 2508. Mexico City,
07360, Mexico
R. Lozano is with Heudiasyc UMR-7253, CNRS-UTC, Compiegne, France
A. Osorio is with CGSTIC, CINVESTAV, Av. IPN 2508. Mexico City,
07360, Mexico
Manuscript received March 30, 2013; revised xx xx, xxxx.
loads but still feeling the effort performed to move the load
[3].
There are many papers in the literature dealing with humanmachine interaction [4], [5], [6], [7], [8], [9], [10], however
the stability of the proposed control algorithms has not been
thoroughly studied. In order to study the stability of the
human machine interaction we need to introduce a model
for the human behaviour. There are several possible models
for the human operator [11], [12], [13], [14], [15] and we
have selected the model proposed in [12] because it takes into
account the delays present in the human reflexes.
In this paper we present a simple controller for the FAD
and a stability proof of its interaction with a human operator.
We first study the stability of the interaction considering no
delays in the human model. The human-FAD interaction is of
order 4th and the proof of the stability is carried out using the
Routh-Hurwitz stability criterion. Furthermore, we have also
studied the stability considering delays in the human model.
Indeed, an upper bound for the delay has been found such
that the stability is preserved. The human-FAD interaction is
illustrated through numerical simulations.
The paper is organized as follows: Section II gives the
nomenclature used in this article. Section III introduces a
model for the force augmenting device (FAD), a human model
and the control algorithm proposed in [16]. Section IV presents
the stability proof for the system neglecting delays in the
human model. In Section V the stability is studied considering
that both delays in the human model are equal. A stability
analysis is performed in Section VI, considering that both
delays in the human model are independent. Section VII
focuses on the selection of the control parameters. Simulation
and experimental results are presented in Section VIII and IX.
Concluding remarks are finally given in Section X.
II. N OMENCLATURE
h
vd
v
h
B
d1
d2
E
F
FA
Fe
Human arm position

Virtual desired position
Output of the spinal cords reflex action
External torque acting on the human joint
Viscous friction in the human arm movement
Delay in the position reflex feedback
Delay in the velocity reflex feedback
Physical compliance of the human flesh
Total force exerted on the moving block
Force exerted by the motor on the moving block
Force exerted by the human on the moving block
Kp
Kd
Position sensor
DC Motor
W
_
_
F
+
+
Fm
Ball-screw
mechanism
.
ye
Kf
KA-1
1
Ms
Moving block
Fe
Mechanical
Block
1
s
ye
_
+
yh
la
d2
-1
Force sensor
Fh
la
Human arm
h
.
vd
Fig. 1. PHOTO OF A HUMAN OPERATING A LINEAR FORCE AUGMENTING DEVICE
Fh
g
Gp
Gv
J
K
KA
Kd
Kf
Kp
la
M
s
t
W
yh
Force exerted by the moving block on the human arm

Acceleration due to gravity
Control parameter of the spinal chord
Control parameter of the spinal chord
Human arm moment of inertia
Muscle stiffness
Force augmenting gain
Derivative gain
Viscous friction coefficient of FAD
Proportional gain
Human arm length
Mass of the moving block
Laplace transform complex variable
Time
Weight of the moving block
Human arm displacement at the end of the arm
III. M ATHEMATICAL M ODEL
The force augmenting device (FAD) considered in this paper

is shown in Fig. 1 and Fig. 2 depicts its block diagram. This
figure also shows the interaction between a human operator
and the FAD. The FAD has a moving block of 24 kg connected
to a ball-screw mechanism, and acts as the load to be lifted by
the human operator. The ball-screw mechanism is driven by a
DC servo motor. Force and position sensors are attached to the
FAD to capture the force exerted by the human on the moving
block and to measure its position. The total force exerted on
the moving block (F ) can be decomposed as
F
FA + Fe W
(1)
where W = M g. Considering zero initial conditions, the

dynamics of the moving block can be written as
ye (s)
F (s)
1
M s2 + Kf s
_
+
(2)
_
+
Gv
1
s
K
+
vd
+
_
1
Js
Musculoskeletal system
Gp
d1
Spinal cord
Fig. 2. BLOCK DIAGRAM OF THE HUMAN-FAD INTERACTION
The human model used for describing the human operator

arm is based on a servo hypothesis proposed in [12], where the
spinal cord actuates the muscles of the musculoskeletal system
by computing the signals received from the brain, and the
position and velocity feedback signals produced by the sensory
organs of the musculoskeletal system. The position reached by
the human arm is not exactly equal to the position commanded
by the spinal cord due to the arm inertia, so the command
signal given by the spinal cord to the musculoskeletal system
is termed as the virtual position [12], [17]. The human brain
generates a desired trajectory through which the arm must
move; based on the desired trajectory the brain computes a
signal and sends it to the spinal cord [13], which is the virtual
desired position [12]. The musculoskeletal system is modeled
as a second order linear transfer function.
The dynamics of the musculoskeletal system shown in Fig. 2
can be written as
h (s)
= J h (s) + B h (s) + K(h (s) v (s))
(3)
The term v represents the virtual arm position. The following equation represents the spinal cord reflex action:
v (s)
= vd (s) + Gp (vd (s) h (s)esd1 ) + sGv (vd (s)

h (s)esd2 )
(4)
The force exerted by the human arm Fe on the moving

block can be expressed as
Fe (t)
= Fh (t) = (yh (t) ye (t))E
(5)
yh (t)
= h (t)la
(6)
The torque exerted by the moving block on the human arm

can be given as
= Fh (t)la = (ye (t) yh (t))Ela
h (t)
(7)
The following control algorithm [16] is applied to the FAD:

FA (t)
(KA 1)Fe (t) Kd y e (t) Kp ye (t)
(8)
where KA is the augmenting factor which allows amplifying

the force exerted by the human operator and the term Kd y e (t)
introduces the required damping into the system.
Since the input terms W , yvd and syvd do not effect the
stability, these terms can be neglected for the stability study.
The characteristic equation for the closed loop system can be
written as
P (s)
= C4 s4 + C3 s3 + C2 s2 + C1 s + C0
(9)
V. S TABILITY CONSIDERING IDENTICAL DELAYS

Assume that both delays are the same, i.e. d1 = d2 = d.
The substitution proposed by Rekasius is [18], [19]
1 Ts
T <, d <+
(24)
1 + Ts
which is defined when s = j, <. This substitution allows
to replace the exponential transcendental term associated to
the time-delay (i.e. esd ) with a rational expression of the
variables s and T . The relation between d , T and can be
given as [18], [19]
esd
d
l
P (s)
= JM
C3
= M (B + Gv Kesd2 ) + J(Kd + Kf )
C2
+ K(Gp e
sd1
(12)
(B + Gv Kesd2 )(Kp + EKA ) + (Kd
+Kf )(Ela2 + K(Gp esd1 + 1))

C0
(11)
+ 1)) + (Kd + Kf )(B
+Gv Kesd2 ) + J(Kp + EKA )

C1
B5
JM T
(27)
(28)
B4
JM + BM T + J(Kd + Kf )T Gv KM T
B3
EM T la2 + BM + J(Kd + Kf ) + Gv KM
+B(Kd + Kf )T + JKp T + KM T + EJKA T
Gv K(Kd + Kf )T Gp KM T
(14)
(26)
where
(13)
E 2 KA la2
(25)
B5 s5 + B4 s4 + B3 s3 + B2 s2 + B1 s
+B0
(Kp + EKA )(Ela2 + K(Gp esd1 + 1))
(10)
M (Ela2
2
arctan (T ) + l
:= , ... 1, 0, 1, ...,
=
where for a fixed each T maps to infinitely many values of

d. Substituting (24) into (9), we get
where
C4
B2
(29)
B(Kd + Kf ) + JKp + KM + EJKA

+Gv K(Kd + Kf ) + Gp KM + BKp T + K(Kd
In the following section we will study the stability of the

above polynomial.
+Kf )T + EM la2 + E(Kd + Kf )T la2

+BEKA T Gp K(Kd + Kf )T Gv KKp T
EGv KKA T
IV. S TABILITY ANALYSIS WHEN THE DELAYS ARE ZERO

Considering that d1 = d2 = 0, equation (9) can be rewritten
B1
P (s)
A4 s4 + A3 s3 + A2 s2 + A1 s + A0
+EKKA T Gp KKp T EGp KKA T

B0
A4
= JM
(16)
A3
= M (B + Gv K) + J(Kd + Kf )
(17)
M (Ela2
+ K(Gp + 1)) + (Kd + Kf )(B + Gv K)
+J(Kp + EKA )
=
(B + Gv K)(Kp + EKA ) + (Kd +

+K(Gp + 1))
A0
+Kf )la2 + EKp T la2 + EGv KKA
(15)
where
A1
BKp + K(Kd + Kf ) + BEKA + Gp K(Kd

+Kf ) + Gv KKp + KKp T + E(Kd
as
A2
(30)
EKp la2
(18)
Kf )(Ela2
(19)
+ (Gp + 1)KKp + E(Gp + 1)KKA (20)
As per the Routh-Hurwitz stability criterion, the closed loop

system is stable if and only if
A0 > 0; A1 > 0; A2 > 0; A3 > 0; A4 > 0; A5 > 0 (21)

(22)
b1 := A3 A2 A4 A1 A3 > 0

c1 := b1 A1 A3 A0 b1 > 0
(23)
The remaining of the stability proof is given in the Appendix.
EKp la2
(31)
+ KKp + EKKA + Gp KKp
+EGp KKA
(32)
The estimated values for the laboratory setup shown in

Fig. 1 are M = 23.4 kg and Kf = 0 N m1 s. The authors in [12] use the following parameters for a human
arm: J = 0.1 N m rad1 s2 , B = 0.89 N m rad1 s1 ,
K = 4 N m rad1 , Gp = 2 and Gv = 0.3 s. The value
E = 920 N m1 for human flesh is given in [20]. Taking la =
0.35 m and the control parameters KA = 125, Kd = 65 kg s1
and Kp = 45 kg s2 , permits rewriting (26) as
P (s)
2.3 105 T s5 + (2.3 105 7.5 104 T )s4

+(1.4 109 T + 5.5 106 )s3 + (1.5 109
2.9 109 T )s2 + (2.5 1010 4.6 1010 T )s
+1.4 1011
(33)
Since the system is stable when the delays are zero, all the
poles are in the left hand side of the complex plane. New poles
s5
s4
s3
s2
s1
s0
2.3 105 T
(2.3 0.8T ) 105
B31
B21
B11
1.4 1011
(140.3T + 0.6) 106

(1.5 2.9T ) 109
B32
1.4 1011
0
0
B53
B43
0
0
0
0
where
=
(2.5 4.6T ) 1010
B43
B31
1.4 1011
.

1 108 563.5T 2 12.8T + 1.3
23.4 7.5T

.

= 5 1010 6.8T 2 89.9T + 11.6
23.4 7.5T

= 1 108 16113.9T 3 8528.6T 2 + 114.2T
.

5.3
563.5T 2 12.8T + 1.3

= 1 1010 14.7T 4 37.7T 3 + 4.5T 2 0.07T
.

+2416.6
16.1T 3 8.5T 2 + 0.1T 0.005
B32
B21
B11
As per the Routh-Hurwitz stability criterion, when the poles

are on the imaginary axis, a row of the Routh array becomes
zero and its previous row is called the auxiliary polynomial.
The auxiliary polynomial gives the location of the poles on
the imaginary axis. Solving B11 = 0 and considering only real
values for T , the solution is T = 2.44372 or T = 0.11131.
The auxiliary polynomial (PA ) is given by the next expression
PA
175
Unstable region
150
125
100
Stable region
75
50
0.15
0.16
0.17
dc (s)
0.18
B21 s + 1.4 10
11
VI. S TABILITY CONSIDERING INDEPENDENT DELAYS

In this section let us consider different delays d1 and d2 . A
Rekasius substitution similar to (24) for the two time-delays
is
1 T1 s
esd1 =
T1 <, d1 <+
(35)
1 + T1 s
1 T2 s
T2 <, d2 <+
(36)
esd2 =
1 + T2 s
and the relation between T1 , 1 , and d1 and T2 , 2 , and d2
is given as
d1
d2
l
2
arctan (T1 ) + l
2
=
arctan (T2 ) + l
:= , ... 1, 0, 1, ...,
=
P (s)
(38)
= D6 s6 + D5 s5 + D4 s4 + D3 s3 + D2 s2
+D1 s + D0
(39)
where
D6
= JM T1 T2
D5
= JM T1 + JM T2 + BM T1 T2 + JKd T1 T2
(40)
+JKf T1 T2 Gv KM T1 T2
D4
EM T1 T2 la2
(41)
+ JM + BM T1 + BM T2
+JKd T1 + JKd T2 + JKf T1 + JKf T2
d|l=0
0.1599
+Gv KM T1 Gv KM T2 + BKd T1 T2
d|l=1
3.3015
+BKf T1 T2 + JKp T1 T2 + KM T1 T2
= 2.9817
The smallest positive delay at which the poles cross the

imaginary axis for the first time is called the critical delay (dc ).
At l = 0, d = dc = 0.1599, the critical delay for this system
is observed. Repeating the above calculations for the different
values of KA , we can obtain the critical delay corresponding
to KA . Fig. 3 shows the change in critical delay as a function
of the augmenting factor KA .
(37)
Substituting (35) and (36) into (9), we get
(34)
Solving (34) at T = 2.44372 and T = 0.11131, yields respectively s = j = j(j5.0048) and s = j = j11.1765.
Since <, T = 0.11131, = 11.1765 is the required
value. A given < and T < produce infinite timedelays d satisfying (25), at which a pair of poles are transferred
from one side to another. Equation (25) is used to obtain the
following values:
d|l=1
0.19
Fig. 3. CRITICAL DELAY dc AT 50 KA 200 IN EQUAL DELAY

CASE (i.e. d1 = D2 )
B53
200
KA
are introduced from the left hand side of the complex plane
as a consequence of the delays [21]. The position of the poles
in the complex plane as a function of the delay is continuous
[21], [22]. As the delays increase, the poles move towards the
right hand side of the plane and finally cross the j axis. The
substitution (24) is valid at the moment when the poles are on
the imaginary axis just before crossing it.
A Routh array can be constructed for (33) as
+EJKA T1 T2 Gv KKd T1 T2
Gv KKf T1 T2 Gp KM T1 T2
D3
= BM + JKd + JKf + Gv KM + BKd T1

+BKd T2 + BKf T1 + BKf T2 + JKp T1
+JKp T2 + KM T1 + KM T2 + EM T1 la2
+EM T2 la2 + EJKA T1 + EJKA T2
+Gv KKd T1 Gv KKd T2 + Gv KKf T1
(42)
Gv KKf T2 Gp KM T1 + Gp KM T2
0.4
+BKp T1 T2 + KKd T1 T2 + KKf T1 T2

+EKd T1 T2 la2
(43)
= BKd + BKf + JKp + KM + EJKA
0.25
d2 (s)
Gv KKp T1 T2 EGv KKA T1 T2

+Gv KKd + Gv KKf + Gp KM + BKp T1
Stable region
0.2
p2 p0
0.15 p1
+BKp T2 + KKd T1 + KKd T2 + KKf T1
p3
p4
p5
0.1
+KKf T2 + EM la2 + EKd T1 la2
p6
0.05
+EKd T2 la2 + EKf T1 la2
0
0
+EKf T2 la2 + BEKA T1 + BEKA T2
0.1
0.2
0.3
Gp KKd T1 + Gp KKd T2 Gp KKf T1

+Gp KKf T2 + Gv KKp T1 Gv KKp T2
+KKp T1 T2 +
Unstable region
0.3
+BEKA T1 T2 Gp KKd T1 T2 Gp KKf T1 T2

D2
p8
0.35
EKf T1 T2 la2
0.4
d1 (s)
0.5
0.6
0.7
Fig. 4. CRITICAL DELAYS FOR d1 AND d2 AT KA = 125
EKp T1 T2 la2
+EGv KKA T1 EGv KKA T2

+EKKA T1 T2 Gp KKp T1 T2
200
EGp KKA T1 T2
D1
(44)
= BKp + KKd + KKf + BEKA + Gp KKd
Unstable region
150
KA
+Gp KKf + Gv KKp + KKp T1 + KKp T2

+EKd la2 + EKf la2 + EKp T1 la2
100
+EKp T2 la2 + EGv KKA + EKKA T1

50
0.4
+EKKA T2 Gp KKp T1 + Gp KKp T2

EGp KKA T1 + EGp KKA T2
D0
EKp la2
(45)
(46)
Using the values for the parameters B, E, Gp , Gv , J, K,

KA , Kd , Kf , Kp , la , M given in the previous section, we
can rewrite (39) as
=
0.2
+ KKp + EKKA + Gp KKp
+EGp KKA
P (s)
Stable region
0.8
d1 (s)
+KA Fe (t)
(48)
0.3T2 + 6 10 T1 T2 + 1)s + (6 10 T1
+6 103 T2 1 104 T1 T2 + 23.7)s3
When the operator removes his hand from the mechanical

block, the system dynamics can be given as
+(1.1 105 T1 1.2 104 T2 2 105 T1 T2

M ye (t)
+6.2 103 )s2 + (6 105 T2 2 105 T1

+1.1 105 )s + 6 105
(47)
Substituting T1 by a numerical value into (47), we get

(47) with only one variable T2 , which is similar to (33). The
computation for solving T given in the previous section can be
used to obtain T2 , therefore the critical delays corresponding
to T1 and T2 can be computed using (37) and (38).
Varying T1 in the region [0.001 10] and solving for T2 and
hence finding critical delays for d1 and d2 , see Fig. 4. Also,
varying KA from 50 to 200 we obtain Fig. 5 which shows the
critical delays of d1 and d2 as a function of KA .
VII. S ELECTION OF THE CONTROLLER PARAMETERS
Substituting (1) and the control algorithm (8) into (2), we
get
=
0.2
0.6
Fig. 5. CRITICAL DELAYS FOR d1 AND d2 AT 50 KA 200
T1 T2 s6 + (T1 + T2 0.3T1 T2 )s5 + (23.7T1

3
M ye (t)
d2 (s)
0.4
W (Kd + Kf )y e (t) Kp ye (t)
= W (Kd + Kf )y e (t) Kp ye (t) (49)
a) Selection of Kp :: At the equilibrium position, we can

write (48) as
Fe (t)
W + Kp ye (t)
KA
(50)
The above equation shows that the force exerted by the

human depends on the weight of the mechanical block and its
position. In an ideal condition, the force exerted by the human
arm (Fei ) is proportional to the weight of the mechanical
block, i.e.
Fei (t)
W
KA
(51)
Let the error be

e :=
Fe (t) Fei (t)

100%
Fei
(52)
0.4
0.4
Moving block position (ye)
0.1
0
0
0.2
Position (m)
0.2
0.1
10
0
0
15
0.3
0.2
0.1
Time (s)
10
0
0
15
0.8
Moving block position (y )
0.4
0.2
10
0
0
15
0
0
15
0.6
0.6
0.2
10
15
0.8
0.6
0.4
0
0
0.4
0.2
10
0
0
15
Time (s)
(g) POSITION VS TIME WITH DELAYS p =

p6
15
(f) POSITION VS TIME WITH DELAYS p =

p5
0.2
Time (s)
10
Time (s)
Position (m)
0.8
Position (m)
Position (m)
10
(e) POSITION VS TIME WITH DELAYS p =

p4
0.8
0.4
Time (s)
(d) POSITION VS TIME WITH DELAYS p =

p3
0
0
0.6
0.2
Time (s)
0.4
Position (m)
Position (m)
Position (m)
0.1
Moving block position (y )
0.8
0.6
15
(c) POSITION VS TIME WITH DELAYS p =

p2
0.3
10
Time (s)
(b) POSITION VS TIME WITH DELAYS p =

p1
0.4
0
0
Time (s)
(a) POSITION VS TIME WITH DELAYS p =

p0
0.2
0.4
0.3
Position (m)
Position (m)
0.3
0.5
10
15
Time (s)
(h) POSITION VS TIME WITH DELAYS p =

p7
(i) POSITION VS TIME WITH DELAYS p =

p8
Fig. 6. SIMULATION RESULTS
Assuming that this error may not exceed 10%, we get

0.1W
Kp min
(53)
ye (t)
0.1W
Kp
(54)
Yem
where Yem = max (ye (t)).
b) Selection of Kd :: When the human operator removes
his hand from the mechanical block, the mechanical block
should not oscillate. In other words, the damping factor for
the system (49) should not be less than 1, i.e.
Kd + Kf
p
2 Kp M
(55)
Kd
p
Kp M Kf
(56)
c) Selection of KA :: From (50) it is observed that the

force exerted by the human operator is amplified when KA >
1 and attenuated when 0 < KA < 1. The human operator can
select KA depending on the requirement.
VIII. S IMULATION RESULTS
A simulation has been performed on the system shown
in Fig. 2 using MATLAB Simulink. A Runge-Kutta solver
of 4th order with a step size of 1 ms is used for the numerical simulation. Simulation is performed with the system
parameters given in the previous sections and KA is taken as
125. p = (d1 , d2 ) be a point on Fig. 4 representing delays.
0.1
0.05
2
30
40
Time (s)
50
0
0
60
(a) Human operator of age 21 operating FAD

with KA = 50 and M = 24 kg
Fe
Position (m)
0.3
10
0.1
10
20
30
40
Time (s)
50
60
10
20
30
40
Time (s)
50
0.4
60
0
0
10
20
30
40
Time (s)
50
0.5
0
0
Fe
(d) Human operator of age 21 operating FAD

10
20
30
40
Time (s)
50
60
ye
Fe
0.2
60
(c) Human operator of age 21 operating FAD

10
ye
15
0.2
0
0
(b) Human operator of age 21 operating FAD

20
ye
Force (N)
0.4
0.2
Force (N)
Force (N)
0.15
20
Position (m)
10
0.2
10
Fe
Position (m)
0.25
0
0
ye
12
Force (N)
Position (m)
14
0.5
10
ye
Force (N)
Fe
0.3
Position (m)
0.35
0.4
16
Position (m)
ye
Force (N)
0.4
0
0
(e) Human operator of age 21 operating FAD

10
20
30
Time (s)
40
50
(f) Human operator of age 21 operating FAD

Fig. 7. Experimental results for the human operator of age 21
30
40
Time (s)
50
60
0
0

Fe
Position (m)
0.3
20
0.1
10
10
20
30
40
Time (s)
50
30
40
Time (s)
50
0.4
0
60

0
0
20
30
40
Time (s)
50
0
60

30
40
Time (s)
0
60
50
ye
Fe
0.3
20
10
20
0.4
Fe
0.2
10

40
ye
Force (N)
Position (m)
0
0
60
30
0.2
0
0
20

40
ye
Force (N)
0.4
10
0.2
0.1
0
0
10
20
30
40
Time (s)
50
Force (N)
20
Position (m)
10
Position (m)
Fe
Force (N)
0.1
0.2
ye
Fe
Force (N)
10
0.5
10
ye
15
0.2
0
0
0.4
Position (m)
Fe
0.3
Position (m)
20
ye
Force (N)
0.4
0
60

4
2
0
0
60

ye
Fe
Position (m)
0.3
0.2
20
40
Time (s)
15
50
ye
Fe
0.15
0.1
0.05
2
30
40
Time (s)
50
50
0.4
60
ye
10
Fe
0.2
0
0
60
30
40
Time (s)
12
10
20
20
14
0.2
10
10

16
0.25
0
0

0
0
60
0.3
60
30
40
Time (s)
0.4
0
0
20
0.35
10
0.1
10

20
Force (N)
0.4
Force (N)
0.1
0.05
0.2

10
20
30
40
Time (s)
50
Force (N)
50
Position (m)
0.15
30
40
Time (s)
Fe
Position (m)
10
20
ye
12
0.2
10
0.5
10
Force (N)
Position (m)
14
0.25
0
0
ye
Force (N)
Fe
0.3
Position (m)
0.35
0.4
16
Position (m)
ye
Force (N)
0.4
60

0.1
0.05

ye
0.35
Fe
Position (m)
0.3
10
8
0.15
0.1
0.05
2
20
30
40
Time (s)
50
60
20
30
40
Time (s)
50
60
0
0
0.4
ye
10
20
30
40
Time (s)
50
60

10
0.4
ye
Fe
10
Fe
12
0.2
10
10
14
0.25
0
0

16
Force (N)
0.4
0.1
0
0
60

0.2
0
0
10
20
30
40
Time (s)
50
60

0.2
0
0
10
20
30
40
Time (s)
50
60
Force (N)
40
Time (s)
0.2
Fe
Position (m)
20
ye
Force (N)
0.15
0.5
Position (m)
10
0.2
Fe
0.3
12
Force (N)
Position (m)
14
0.25
0
0
ye
Force (N)
Fe
0.3
Position (m)
0.35
0.4
16
Position (m)
ye
Force (N)
0.4

Simulation is performed at p0 = (0.1599, 0.1599), p1 =

(0, 0.15), p2 = (0.1, 0.15), p3 = (0.2, 0.15), p4 = (0.6, 0.15),
p5 = (0.7, 0.15), p6 = (0.6, 0.05), p7 = (0.6, 0.35), p8 =
(0.6, 0.375). Simulation results are shown in Fig. 6.
c1s
:= bh1 A1 A3 A0 > 0
=
B 3 EKA (Kd + Kf )M + B 3 (Kd + Kf )Kp M
+B 2 E 2 KA M 2 la2 + 3B 2 EGv KKA (Kd + Kf )M
IX. E XPERIMENTAL RESULTS

The prototype designed at the laboratory uses a computer
with Windows XP operating system, which runs MATLABSIMULINK along with the WinCon software. This setup
allows performing a real time experiment with the FAD. We
have used a sampling time of 1 ms. The control parameters
used for the experiments are Kd = 65 kg s1 and Kp =
45 kg s2 .
Four human operators of age 21, 28, 40 and 52 are selected
to perform a task of lifting the moving block to a height
of 100 mm, 200 mm, 250 mm, and 300 mm and to maintain
at each position for 10 s and to release their hand from the
moving block to make it reach the ground position. This task
is performed with KA = 50, KA = 100 and KA = 200 with
the moving block weight of 24 kg and 28 kg. Experimental
results are shown Fig. 7 through Fig. 10.
+B 2 EJKA (Kd + Kf )2 + B 2 E(Kd + Kf )2 M la2

+B 2 (Gp + 1)K(Kd + Kf )2 M + 3B 2 Gv K(Kd
+Kf )Kp M + B 2 J(Kd + Kf )2 Kp
+2BE 2 Gv KKA M 2 la2 + 3BEG2v K 2 KA (Kd
+Kf )M + 2BEGv JKKA (Kd + Kf )2
+2BEGv K(Kd + Kf )2 M la2 + BEJ(Kd
+Kf )3 la2 + 2B(Gp + 1)Gv K 2 (Kd
+Kf )2 M + B(Gp + 1)JK(Kd + Kf )3
+3BG2v K 2 (Kd + Kf )Kp M + 2BGv JK(Kd
+Kf )2 Kp + E 2 G2v K 2 KA M 2 la2
+E 2 J 2 KA (Kd + Kf )2 la2 + EG3v K 3 KA (Kd
+Kf )M + EG2v JK 2 KA (Kd + Kf )2
+EG2v K 2 (Kd + Kf )2 M la2 + EGv JK(Kd
+Kf )3 la2 + (Gp + 1)G2v K 3 (Kd + Kf )2 M
X. C ONCLUSION
In this paper we have proved the stability of the interaction
of a human and a FAD. We have considered a general human
operator model proposed in [12]. We have proved the stability
in the case of zero delays using the Routh-Hurwitz criterion.
When the human model includes time-delays, we have also
found an upper bound for the delays such that the stability
is preserved. It is observed that the delays in the human
model are around 40 ms [12], which are smaller than the
upper bound found for this system i.e. 159.9 ms, which makes
this system robust to the time-delays in the human model.
Numerical simulations have shown the performance of the
closed loop system when there are time-delays in the human
model. Robustness is tested in realtime by performing an
experiment where human operators of 4 different age groups
are allowed to interact with the FAD. It was observed that the
human machine interaction is stable and also it is observed
that damping is enough to maintain the system without any
oscillation. Furthermore, it is observed that the human operator
is actually exerting a small fraction of the total effort needed
to lift the weight.
+(Gp + 1)Gv JK 2 (Kd + Kf )3 + G3v K 3 (Kd

+Kf )Kp M + G2v JK 2 (Kd + Kf )2 Kp + Gv K(Kd

+Kf ) EM la2 + JKp + EJKA (Gp
2
+1)KM + 2E 2 Gv JKKA (Kd + Kf )M la2

+B(Kd + Kf ) EM la2 + JKp + EJKA
2
(Gp + 1)KM + 2BE 2 JKA (Kd
,
i h
2
+Kf )M la
BM + J(Kd + Kf )
i
+Gv KM > 0
(58)
Conditions (21), (57) and (58) are satisfied for all positive
coefficients, so the polynomial is always stable considering no
delays.
ACKNOWLEDGMENT
The authors would like to thank Gerardo Castro, and Jesus
Meza, for their help in implementing the electrical setup and
Roberto Lagunes for his help in building the mechanical setup.
A PPENDIX A
R EMAINING PART OF THE STABILITY PROOF
Simplifying b1 from (22), we get
h
b1 = (BJ + Gv JK)(Kd + Kf )2 + (B 2 M
R EFERENCES
+J 2 Kp + EJ 2 KA + G2v K 2 M
+2BGv KM )(Kd + Kf ) + (Gp
+1)Gv K 2 M 2 + EGv KM 2 la2 + B(Gp
,
i h
2
2 2
+1)KM + BEM la
(B
i
+Gv K)M + J(Kd + Kf ) > 0
Condition (23) can be simplified as
(57)
[1] E. Guizzo and H. Goldstein, The rise of the body bots [robotic
exoskeletons], Spectrum, IEEE, vol. 42, no. 10, pp. 5056, 2005.
[2] W. Kim, S. Lee, H. Lee, S. Yu, J. Han, and C. Han, Development of
the heavy load transferring task oriented exoskeleton adapted by lower
extremity using qausi-active joints, in ICCAS-SICE, 2009. IEEE, 2009,
pp. 13531358.
[3] T. Snyder and H. Kazerooni, A novel material handling system, in
1996 IEEE International Conference on Robotics and Automation, 1996.
Proceedings., vol. 2. IEEE, 1996, pp. 11471152.
[4] S. M. H. P. Operation, Hardiman 1 prototype project, General Electric
Company, Schenectady, New York 12305, Tech. Rep., December 1969.
10
[5] H. Kazerooni, Human machine interaction via the transfer of power

and information signals, ASME Winter Annual Meeting, DECEMBER
1988.
[6] , Human-robot interaction via the transfer of power and information signals, IEEE Transactions on Systems, Man and Cybernetics,
vol. 20, no. 2, pp. 450463, 1990.
[7] S. Lee and Y. Sankai, Power assist control for leg with hal-3 based on
virtual torque and impedance adjustment, in 2002 IEEE International
Conference on Systems, Man and Cybernetics, vol. 4. IEEE, 2002, pp.
6pp.
[8] K. Yamamoto, K. Hyodo, M. Ishii, and T. Matsuo, Development
of power assisting suit for assisting nurse labor, JSME International
Journal Series C, vol. 45, no. 3, pp. 703711, 2002.
[9] H. Kazerooni, The Berkeley lower extremity exoskeleton, in Field and
Service Robotics. Springer, 2006, pp. 915.
[10] K. Kong and M. Tomizuka, Control of exoskeletons inspired by fictitious gain in human model, IEEE/ASME Transactions on Mechatronics,
vol. 14, no. 6, pp. 689698, 2009.
[11] P. Merton, Speculations on the servo-control of movement, in Ciba
Foundation Symposium-The Spinal Cord. Wiley Online Library, 1953,
pp. 247260.
[12] J. McIntyre and E. Bizzi, Servo hypotheses for the biological control
of movement, Journal of Motor Behavior, vol. 25, no. 3, pp. 193202,
1993.
[13] N. Schweighofer, M. Arbib, and M. Kawato, Role of the cerebellum in
reaching movements in humans. i. distributed inverse dynamics control,
European Journal of Neuroscience, vol. 10, no. 1, pp. 8694, 1998.
[14] N. Schweighofer, J. Spoelstra, M. Arbib, and M. Kawato, Role of the
cerebellum in reaching movements in humans. ii. a neural model of the
intermediate cerebellum, European Journal of Neuroscience, vol. 10,
no. 1, pp. 95105, 1998.
[15] T. Oshima, T. Fujikawa, O. Kameyama, and M. Kumamoto, Robotic
analyses of output force distribution developed by human limbs, in
Robot and Human Interactive Communication, 2000. RO-MAN 2000.
Proceedings. 9th IEEE International Workshop on. IEEE, 2000, pp.
229234.
[16] S. K. Gadi, R. Lozano, R. Garrido, and A. Osorio, Stability analysis and
experiments for a force augmenting device, in 9th International Conference on Electrical Engineering, Computing Science and Automatic
Control (CCE), 2012, sept. 2012, pp. 16.
[17] M. Latash and G. Gottlieb, Reconstruction of shifting elbow joint compliant characteristics during fast and slow movements, Neuroscience,
vol. 43, no. 2-3, pp. 697712, 1991.
[18] N. Olgac and R. Sipahi, The direct method for stability analysis of
time delayed LTI systems, in American Control Conference, 2003.
Proceedings of the 2003, vol. 1. IEEE, 2003, pp. 869874.
[19] Z. Rekasius, A stability test for systems with delays, in Proceedings
of the Joint Automatic Control Conference, San Francisco, CA, paper
TP9-A, 1980.
[20] T. Pataky, M. Latash, and V. Zatsiorsky, Viscoelastic response of
the finger pad to incremental tangential displacements, Journal of
biomechanics, vol. 38, no. 7, pp. 14411449, 2005.
[21] W. Michiels and S. Niculescu, Stability and Stabilization of Time-Delay
Systems: An Eigenvalue-Based Approach, ser. Advances in Design and
Control. Society for Industrial and Applied Mathematics, 2008. [Online]. Available: http://books.google.com.mx/books?id=O7F0Vu7xeXcC
[22] J. Neimark, D-subdivisions and spaces of quasipolynomials, Prikladnaya Matematika i Mekhanika, vol. 13, pp. 349380, 1949.

Study of Stability and Robustness For A Force Augmenting Device

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Study of Stability and Robustness For A Force Augmenting Device

Uploaded by

Copyright:

Available Formats

1

Study of Stability and Robustness of a Human

AbstractIn this paper a force augmenting device (FAD), a

Human arm position

Fig. 1. PHOTO OF A HUMAN OPERATING A LINEAR FORCE AUGMENTING DEVICE

Force exerted by the moving block on the human arm

The force augmenting device (FAD) considered in this paper

where W = M g. Considering zero initial conditions, the

The human model used for describing the human operator

= J h (s) + B h (s) + K(h (s) v (s))

= vd (s) + Gp (vd (s) h (s)esd1 ) + sGv (vd (s)

The force exerted by the human arm Fe on the moving

= Fh (t) = (yh (t) ye (t))E

The torque exerted by the moving block on the human arm

The following control algorithm [16] is applied to the FAD:

(KA 1)Fe (t) Kd y e (t) Kp ye (t)

where KA is the augmenting factor which allows amplifying

V. S TABILITY CONSIDERING IDENTICAL DELAYS

(B + Gv Kesd2 )(Kp + EKA ) + (Kd

+Kf )(Ela2 + K(Gp esd1 + 1))

+ 1)) + (Kd + Kf )(B

+Gv Kesd2 ) + J(Kp + EKA )

(Kp + EKA )(Ela2 + K(Gp esd1 + 1))

where for a fixed each T maps to infinitely many values of

B(Kd + Kf ) + JKp + KM + EJKA

In the following section we will study the stability of the

+Kf )T + EM la2 + E(Kd + Kf )T la2

IV. S TABILITY ANALYSIS WHEN THE DELAYS ARE ZERO

+EKKA T Gp KKp T EGp KKA T

+ K(Gp + 1)) + (Kd + Kf )(B + Gv K)

(B + Gv K)(Kp + EKA ) + (Kd +

+Kf )la2 + EKp T la2 + EGv KKA

BKp + K(Kd + Kf ) + BEKA + Gp K(Kd

+ (Gp + 1)KKp + E(Gp + 1)KKA (20)

As per the Routh-Hurwitz stability criterion, the closed loop

+ KKp + EKKA + Gp KKp

The estimated values for the laboratory setup shown in

2.3 105 T s5 + (2.3 105 7.5 104 T )s4

(140.3T + 0.6) 106

(2.5 4.6T ) 1010

As per the Routh-Hurwitz stability criterion, when the poles

VI. S TABILITY CONSIDERING INDEPENDENT DELAYS

+JKd T1 + JKd T2 + JKf T1 + JKf T2

The smallest positive delay at which the poles cross the

Substituting (35) and (36) into (9), we get

Fig. 3. CRITICAL DELAY dc AT 50 KA 200 IN EQUAL DELAY

= BM + JKd + JKf + Gv KM + BKd T1

+BKp T1 T2 + KKd T1 T2 + KKf T1 T2

= BKd + BKf + JKp + KM + EJKA

Gv KKp T1 T2 EGv KKA T1 T2

+BKp T2 + KKd T1 + KKd T2 + KKf T1

+KKf T2 + EM la2 + EKd T1 la2

+EKd T2 la2 + EKf T1 la2

+EKf T2 la2 + BEKA T1 + BEKA T2

Gp KKd T1 + Gp KKd T2 Gp KKf T1

+BEKA T1 T2 Gp KKd T1 T2 Gp KKf T1 T2

Fig. 4. CRITICAL DELAYS FOR d1 AND d2 AT KA = 125

+EGv KKA T1 EGv KKA T2

= BKp + KKd + KKf + BEKA + Gp KKd

+Gp KKf + Gv KKp + KKp T1 + KKp T2

+EKp T2 la2 + EGv KKA + EKKA T1

+EKKA T2 Gp KKp T1 + Gp KKp T2

Using the values for the parameters B, E, Gp , Gv , J, K,