Anda di halaman 1dari 8

950 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO.

3, MAY 2013
Throwing Motion Control of the Springed Pendubot
Takuya Shoji, Shunsuke Katsumata, Shigeki Nakaura, and Mitsuji Sampei
AbstractThis brief describes a control strategy for the
throwing motion of an underactuated two-link planar robot
called the Pendubot. The springed Pendubot is built based on the
concept of unstable zero dynamics, and our investigation uses it
as a dynamic model of superior limbs to imitate human throwing
motion. In the proposed control strategy, the zero dynamics is
intentionally destabilized when a ball held by the end-effector
is constrained on a geometric path in a vertical plane, using
output zeroing control for the deviation between the ball and
geometric path. The unstable zero dynamics drives the ball along
the geometric path to achieve fast and accurate throw in a desired
direction. The unstable zero dynamics is analytically derived
to guarantee the dynamic acceleration of the ball along the
geometric path. Numerical simulations and experimental results
conrm the effectiveness of the proposed control strategy.
Index TermsInputoutput linearization, nonlinear feedback
control, output zeroing control, pendubot, throwing motion,
underactuated mechanical system, zero dynamics.
I. INTRODUCTION
D
EXTEROUS throwing motion is naturally achieved by
humans without any difculty despite the complexity of
muscle and joint activities. Because of the limitations of their
physical abilities, humans are less inclined to use extension
torque at the cubital joint, instead they use dexterous actions
generated by transferring energy from the trunk to the hand to
produce maximum speed at the end of a distal segment in a
kinematic chain, as shown in Fig. 1. Because throwing motion
consists of a variety of interesting aspects of both dynamics
and kinematics, it has attracted a number of researchers in the
eld of biomechanics [1]. Feltner and Dapena [2] analyzed
the dynamics of the throwing arm during baseball pitches, and
showed that rapid extension of the cubital joint was generated
by motion-dependent torque of the upper arm and not by
extension torque at the cubital joint. Sakurai [3] suggested
that whip-like motion, which transfers energy from the trunk
to the hand, should efciently accelerate the throwing arm.
Manuscript received December 26, 2009; revised September 24, 2010 and
July 7, 2011; accepted March 20, 2012. Manuscript received in nal form
March 21, 2012. Date of publication April 26, 2012; date of current version
April 17, 2013. Recommended by Associate Editor M. Zefran.
T. Shoji is with the Brake System Development Department, Chassis
Development Division, Toyota Motor Corporation, Aichi 471-8572, Japan
(e-mail: shoji@mail.tec.toyota.co.jp).
S. Katsumata is with the Business Development Division, Business
Incubation Unit, Sony Corporation, Minato-ku 108-0075, Japan (e-mail:
hunsukeA.Katsumata@jp.sony.com).
S. Nakaura is with the Department of Mechanical Engineering, Sasebo
National College of Technology, Nagasaki 857-1193, Japan (e-mail:
nakaura@post.cc.sasebo.ac.jp).
M. Sampei is with the Department of Mechanical and Control
Engineering, Tokyo Institute of Technology, Tokyo 152-8550, Japan (e-mail:
sampei@ctrl.titech.ac.jp).
Color versions of one or more of the gures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TCST.2012.2192121
Fig. 1. Dexterous throwing motion by transferring energy from the trunk to
the hand to achieve dynamic acceleration of the ball.
After throwing motion is investigated as a mechanical control
problem, however, study of the aspects of nonlinear dynamics
[4][6] and structural underactuation [7][9] appears to be a
fascinating challenge.
The purpose of this brief is to propose a control strat-
egy for throwing motion of the springed Pendubot. For our
investigation, we use an underactuated robot in the shape of
the Pendubot [10] as a dynamic model of superior limbs to
imitate human throwing motion. The Pendubot, addressed in
this brief, is an underactuated two-link planar robot with an
actuator at the shoulder joint but no actuator at the elbow joint.
A torsion spring, instead, is mounted on the passive joint of the
Pendubot representing the exibility of the cubital joint. This
replacement of an actuator by a spring reduces the weight and
load of mechanical systems with fewer inputs than the number
of degrees-of-freedom. This brief aims to design an energy-
efcient controller based on the structural underactuation
of mechanical systems to investigate the springed Pendubot
as a dynamic model of human throwing motion, which is
fundamentally underactuated in nature.
Because the control objective in this brief is to generate the
motion of underactuated mechanical systems, which cannot be
achieved by approximate linearization or trajectory tracking
control if the systems are not fully feedback linearizable [11],
the proposed control strategy is based on the concept of unsta-
ble zero dynamics using inputoutput linearization and output
zeroing control [12][13]. The zero dynamics describes the
internal behavior of the system when the output is constrained
to remain identically zero [14], [15]. The zero dynamics
is intentionally destabilized when a ball is constrained on
a geometric path in a vertical plane, using output zeroing
control for the deviation between the ball and geometric
path. The unstable zero dynamics [16][18] drives the ball
along the geometric path to achieve dynamic acceleration of
the ball. The unstable zero dynamics is analytically derived
to guarantee the dynamic acceleration of the ball along a
1063-6536/$31.00 2012 IEEE
SHOJI et al.: THROWING MOTION CONTROL OF THE SPRINGED PENDUBOT 951
Fig. 2. Underactuated two-link planar robot called the Pendubot with a
torsion spring mounted on the passive joint.
geometric path by phase portrait analysis. Numerical simu-
lations verify the validity of the proposed control strategy.
For the experimental validation, the experimental setup is well
designed to fulll the assumptions in the numerical simulations
and to follow the dynamic model in the theoretical analysis.
Experimental results conrm the effectiveness of the proposed
control strategy.
This brief is organized as follows. In Section II, a dynamic
model of the throwing motion is derived with the springed
Pendubot. In Section III, a control strategy is proposed for the
throwing motion of the springed Pendubot on the basis of the
concept of unstable zero dynamics. In Section IV, instability
of the zero dynamics is discussed based on phase portrait
analysis. In Section V, numerical simulations are performed
to verify the validity of the designed controller. In Section VI,
experimental results conrm the effectiveness of the proposed
control strategy. Section VII concludes this brief.
II. DYNAMIC MODEL AND STATE EQUATION
This section introduces a dynamic model for controlling the
throwing motion of the springed Pendubot as shown in Fig. 2.
The Pendubot is an underactuated two-link planar robot with
an actuator at joint 1 but no actuator at joint 2. Instead, a
torsion spring is mounted on the passive joint representing
the exibility of the cubital joint of the human arm. Links 1
and 2 represent the upper arm and forearm of a human body,
respectively. The end-effector holds a ball and releases it
instantaneously when it enters a release point in the rst
quadrant for the rst time, while its center of mass rotates in a
positive direction around joint 1. The notations are as follows:
1) for i = 1, 2, m
i
denotes the mass of link i ; 2) l
i
denotes the
length of link i ; 3) l
gi
denotes the distance from the previous
joint to the center of mass of link i ; 4) J
i
denotes the moment
of inertia of link i about an axis coming out of the page,
passing through the center of mass of link i ; 5) m
b
denotes the
mass of the ball; and 6) k denotes the spring constant at joint 2.
The dynamics of the springed Pendubot can be described
by a system of two differential equations, originating from
Lagrangian mechanics
M(q) q + C(q, q) q + G(q) + K(q) = Bu (1)
with the vector of generalized coordinates
q :=
_

1

2
_
T
R
2
(2)
the inertia matrix
M(q) =
_
p
1
+ p
3
+ 2p
2
cos
2
p
3
+ p
2
cos
2
p
3
+ p
2
cos
2
p
3
_
(3)
the matrix corresponding to Coriolis and centrifugal forces
C(q, q) =
_
p
2

2
sin
2
p
2
(

1
+

2
) sin
2
p
2

1
sin
2
0
_
(4)
the gravity vector
G(q) =
_
gp
4
cos
1
+ gp
5
cos (
1
+
2
)
gp
5
cos(
1
+
2
)
_
(5)
the spring torque vector
K(q) =
_
0 k
2
_
T
(6)
the input matrix mapping the applied control torque to gener-
alized forces
B =
_
1 0
_
T
(7)
and the generalized forces
u :=
1
R (8)
where
1
is the absolute angle of link 1,
2
is the relative angle
between the two links, and
1
is the controlled torque applied
to link 1. The physical parameters taken from the experimental
setup are combined as follows:
p
1
= J
1
+l
g1
2
m
1
+l
1
2
(m
2
+ m
b
) (9)
p
2
= l
1
(l
g2
m
2
+l
2
m
b
) (10)
p
3
= J
2
+l
g2
2
m
2
+l
2
2
m
b
(11)
p
4
= l
g1
m
1
+l
1
(m
2
+ m
b
) (12)
p
5
= l
g2
m
2
+l
2
m
b
. (13)
The dynamic model is written in state space form by dening
x =
_
q
M(q)
1
[C(q, q) qG(q)K(q)+Bu]
_
(14)
=: f (x) + g(x)u (15)
with the state vector
x :=
_
q
T
q
T
_
T
R
4
. (16)
III. CONTROLLER DESIGN
This section proposes a strategy for controlling the throwing
motion of the springed Pendubot. Because the dynamic model
in this brief is an underactuated mechanical system, which
is not input-state linearizable and is non-minimum phase, the
dynamics is partially linearized using inputoutput lineariza-
tion. When the partially linearized dynamics converges to zero
using output zeroing control, the system is represented by a
nonlinear unobservable dynamics called the zero dynamics.
952 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 3, MAY 2013
Fig. 3. Output zeroing control for the deviation between the ball and the
geometric path in a vertical plane.
The zero dynamics is intentionally destabilized to achieve the
dynamic acceleration of the ball held by the end-effector.
A throwing motion by humans consists of the following
steps according to biomechanical analysis. At the early stage
of acceleration, rotation of the trunk is transformed to exion
of the cubital joint, building up energy at the superior limb.
At the middle stage of acceleration, extension of the cubital
joint transfers the energy from the trunk to the hand to produce
maximum speed at the end of a distal segment in a kinematic
chain. Accordingly, the ball moves on a curved line at the rst
two stages of acceleration. At the last stage of acceleration,
the geometric trajectory of the ball changes from the curved
line to a straight line improving the accuracy of the direction
of throw. The control objective in this brief, therefore, is to
drive a ball held by the end-effector on a geometric path
that consists of approximately linear and curvilinear sections,
thus throwing the ball in a desired direction at maximum
speed.
In accordance with the previous discussion about the bio-
mechanics of human throwing motion, the ball is constrained
on a geometric path that consists of approximately linear and
curvilinear sections
Y = (X) (17)
which are specied by the Cartesian coordinates in a
vertical plane, as shown in Fig. 3. For the underactuation
of the dynamic model, which is not input-state linearizable,
inputoutput linearization is considered with an output
y = h(x) := y
b
(x
b
) R (18)
where (x
b
, y
b
) denotes the Cartesian coordinates of the ball in
a vertical plane as
x
b
= l
1
cos
1
l
2
cos(
1
+
2
) (19)
y
b
= l
1
sin
1
+l
2
sin(
1
+
2
). (20)
Note that y = 0 implies y
b
= (x
b
). Since the designed
output depends only on the conguration variables, then,
because of the second-order nature of the robot model, the
rst-time derivative of the output along the solutions of the
Fig. 4. Zero dynamics governs the acceleration of the ball along the geometric
path in a vertical plane.
state equation
y =
dy
dt
=
h(x)
x
x = L
f
h(x) (21)
does not depend directly on the input. Therefore, the relative
degree of the output is at least two. The second-time derivative
of the output along the solutions of the state equation is
derived as
y =
d
2
y
dt
2
=
L
f
h(x)
x
x = L
2
f
h(x) + L
g
L
f
h(x)u (22)
which indicates that the relative degree of the output is two.
The preliminary feedback
u = L
g
L
f
h(x)
1
_
v L
2
f
h(x)
_
(23)
yields the output dynamics y = v when v is a new input. The
dynamics of the springed Pendubot is then partially linearized
via inputoutput linearization as
d
dt
_
y
y
_
=
_
0 1
0 0
_ _
y
y
_
+
_
0
1
_
v. (24)
This differential equation called the external dynamics, gives
a linear relationship between the input and the output. For the
linearity of the external dynamics of dimension two, the state
variable feedback
v = K
P
y K
D
y (25)
results in
y + K
D
y + K
P
y = 0 (26)
that is
d
dt
_
y
y
_
=
_
0 1
K
P
K
D
_ _
y
y
_
. (27)
For K
P
, K
D
> 0, the external dynamics converges exponen-
tially to zero.
For y 0, that is y
b
(x
b
), the system is represented by
a reduced-order dynamics called the zero dynamics, provided
that initial conditions are chosen to satisfy the constraint as
shown in Fig. 4. The state variable of the zero dynamics
chosen as
z := x
b
R (28)
SHOJI et al.: THROWING MOTION CONTROL OF THE SPRINGED PENDUBOT 953
denes a mapping
(x) :=
_
_
_
_
y
y
z
z
_

_
=
_
_
_
_
y
b
(x
b
)
y
b

d(x
b
)
dx
b
x
b
x
b
x
b
_

_
(29)
whose Jacobian matrix is nonsingular for all x as a local
coordinates transformation. Because the output (18) consists of
nonlinear functions with respect to the conguration variables,
deriving the zero dynamics is a difcult task. However, with
the focus on the mathematical properties of the dynamic model
and output, the derivation is nally achieved to analyze the
instability of the zero dynamics. The transformation from
the Cartesian coordinates to the polar coordinates under the
constraint y = 0, that is y
b
= (x
b
), is written as

1
= cos
1
z
r(z)
+ cos
1
l
2
1
+r(z)
2
l
2
2
2l
1
r(z)
=:
1
(z) (30)

2
= + cos
1
l
2
1
+l
2
2
r(z)
2
2l
1
l
2
=:
2
(z) (31)
where
r(z) :=
_
x
2
b
+ y
2
b
=
_
z
2
+ (z)
2
. (32)
Regions considered in the throwing motion control can be
covered by appropriately switching intervals of inverse func-
tions. The time derivatives of the new coordinates previous
mentioned can be calculated as

1
=
d
1
(z)
dz
z (33)

2
=
d
2
(z)
dz
z (34)

1
=
d
2

1
(z)
dz
2
z
2
+
d
1
(z)
dz
z (35)

2
=
d
2

2
(z)
dz
2
z
2
+
d
2
(z)
dz
z. (36)
The nonholonomic constraint of the dynamic model is then
reduced as
(z) z + (z) z
2
+ (z) = 0 (37)
where
(z) = ( p
3
+ p
2
cos
2
(z))
d
1
(z)
dz
+ p
3
d
2
(z)
dz
(38)
(z) = ( p
3
+ p
2
cos
2
(z))
d
2

1
(z)
dz
2
+ p
3
d
2

2
(z)
dz
2
+ p
2
_
d
1
(z)
dz
_
2
sin
2
(z) (39)
(z) = gp
5
cos(
1
(z) +
2
(z)) + k
2
(z) (40)
using the new coordinates and the time derivatives. A differ-
ential equation of the zero dynamics is nally derived with
the form
d
dt
_
z
z
_
=
_
z

(z)
(z)
z
2

(z)
(z)
_
. (41)
This differential equation is well known as the Bernoulli
equation whose general solution is always given by
z
2
= (z
0
, z)
_
z
2
0

_
z
z
0
(s, z
0
)
2 (s)
(s)
ds
_
(42)
where
(z
0
, z) = exp
__
z
z
0

2(s)
(s)
ds
_
(43)
with initial conditions z
0
, z
0
. Because the calculation of the
inner integral (43) is very complicated, a general solution of
the differential equation can never be found. The trajectories of
the zero dynamics are discussed on the basis of phase portrait
analysis in the next section.
IV. ZERO DYNAMICS ANALYSIS
This section discusses the instability of the zero dynamics
given by (41) and determines the design parameter d
a
. Because
a general solution of the zero dynamics cannot be derived for
complexity of the form, phase portrait analysis is performed
for the second-order ordinary differential equation, which is
now independent of the original system of dimension four with
output zeroing control. Note that phase portrait analysis and
parameter determination can never be achieved without ana-
lytical derivation of the zero dynamics, because the behavior
of a fully dimensional system depends on not only design
parameters but also the feedback gains for output zeroing
control. Therefore, the exact analysis of the instability of the
zero dynamics is nally achieved with the analytically derived
differential equation of the zero dynamics.
A. Geometric Path Candidates
The control objective in this brief is to drive a ball held by
the end-effector on a generalized geometric path Y = (X),
which consists of approximately linear and curvilinear
sections. Three candidates for the generalized geometric path
are proposed as follows.
1) Sigmoidal path
Y = (X) := d
c
tanh(d
a
(X + d
b
)). (44)
2) Exponential path
Y = (X) := d
c
exp(d
a
(X + d
b
)) + d
c
. (45)
3) Arctangential path
Y = (X) :=
2

d
c
tan
1
(d
a
(X + d
b
)) (46)
where d
a
is the curvature of the geometric path, d
b
is the
horizontal position of the geometric path, and d
c
is the
vertical position of the geometric path. Initial conguration
of the springed Pendubot prespecies
d
b
=
_
l
2
1
+l
2
2
2l
1
l
2
cos( +
20
) (47)
where
20
is the initial angle of joint 2, provided that the
ball is placed on a horizontal axis. Parameter d
c
is also
prespecied as d
c
= 0.4, clarifying the effects that the
shaping of the geometric paths have on speed and accuracy
at the instant of ball release. Hence, d
a
remains as a design
parameter. The physical parameters of the springed Pendubot
are taken from the experimental setup and organized in
Table I. See [8] and [9] for details of the experiment.
954 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 3, MAY 2013
(a) (b) (c)
Fig. 5. Phase portrait of the zero dynamics represented by nonlinear 2-D dynamics with (a) d
a
= 3.5, (b) d
a
= 5.0, and (c) d
a
= 15.0.
(a) (b) (c)
Fig. 6. Geometric trajectory of the ball driven by the unstable zero dynamics with (a) d
a
= 3.5, (b) d
a
= 5.0, and (c) d
a
= 15.0.
TABLE I
PARAMETERS OF THE SPRINGED PENDUBOT
Par. Units Value Par. Units Value
m
1
kg 1.972 m
2
kg 1.537
l
1
m 0.250 l
2
m 0.300
l
g1
m 0.092 l
g2
m 0.074
J
1
kgm
2
0.055 J
2
kgm
2
0.021
m
b
kg 0.057 k Nm/rad 1.604
B. Phase Portrait and Equilibrium Point
The phase portraits of the zero dynamics with values of
the physical parameters listed in Table I are shown in Fig. 5.
The initial conditions of the zero dynamics are now chosen
as z
0
= d
b
, z
0
= 0, which place the ball held by the
end-effector on a horizontal axis for initial conguration.
The second state variable of the zero dynamics increases in
value as the rst state variable of the zero dynamics moves
in a positive direction until the instant of ball release at
z = 0.3. This implies a divergence of the zero dynamics
between the initial and terminal states. Fig. 6 shows geometric
trajectories of the ball along the geometric path. Dynamic
acceleration of the ball is achieved by the designed controller
using the unstable zero dynamics along the geometric path
when the input is devoted to constrain the end-effector on
the geometric path. The terminal velocity of the zero dynam-
ics depends on the choice of design parameter and initial
conditions.
For the complexity of the differential equation given
by (41), the zero dynamics is not analytically solvable.
The equilibrium point of the zero dynamics can numerically
be calculated as z

= 0.34, z

= 0 in the region considered,


however, which is a saddle point, as found by the numerical
computation of the approximately linearized system with
values given in Table I. Therefore, if the equilibrium point of
the approximately linearized state equation is a saddle point,
then in the vicinity of the equilibrium point, the trajectories
of the original nonlinear state equation will also behave like
a saddle point [15]. Note that the values of the equilibrium
point depend on the choice of design parameters, which
govern the shape and position of the geometric path.
C. Optimization for Geometric Paths
In this section, for optimization, we investigate the design
parameter d
a
, which is the curvature of the geometric path.
The signicance of the parameter determination is traced to
the fact that the release speed and angle of the ball can be
represented by the state variables of the zero dynamics under
the constraint y = 0, that is y
b
= (x
b
), for output zeroing
control. There are performance trade-offs in the evaluation
criteria, given as follows.
1) Maximization of release speed
v
r
:=
_
x
2
b
+ y
2
b
= | z|
_
1 +
_
d(z)
dz
_
2
. (48)
SHOJI et al.: THROWING MOTION CONTROL OF THE SPRINGED PENDUBOT 955
TABLE II
RESULTS OF OPTIMIZATION FOR GEOMETRIC PATHS WHERE
VALUES INSIDE BRACKETS CORRESPOND TO d
a
a
r
= 0.01 a
r
= 0.03 a
r
= 0.05
sigmoidal v
r
= 11.9 v
r
= 12.9 v
r
= 14.3
path (4.810) (3.845) (3.375)
exponential v
r
= 12.0 v
r
= 12.9 v
r
= 14.0
path (8.430) (6.450) (5.470)
arctangential v
r
= 10.3 v
r
= 13.6 v
r
= 16.2
path (53.40) (17.70) (10.50)
2) Minimization of release angle
a
r
:= tan
1

y
b
x
b

= tan
1

d(z)
dz

. (49)
Both of the speed and angle at ball release can be written as
a function of the zero dynamics under constraint of output
zeroing control given by y = 0, that is y
b
= (x
b
).
The results of numerical optimization for the geometric path
based on d
a
are organized in Table II. For all geometric paths
proposed earlier, there exists a duality relationship between the
speed and accuracy of ball release. The release angle decreases
as d
a
increases, while the release speed increases as d
a
decreases. The opposite is equally true. The effects of shaping
geometric paths can be seen in Table II. The arctangential path
is the best choice among the three candidates to achieve fast
throw in a desired direction. The exponential path, on the other
hand, is the best choice among the three candidates to achieve
accurate throw in a desired direction.
V. NUMERICAL SIMULATIONS
This section examines numerical simulations to verify the
validity of the designed controller. The design parameters are
chosen as d
a
= 3.65, d
b
= 0.39, and d
c
= 0.40 for the
sigmoidal path (44) when the initial condition is chosen as
x
0
=
_
q
0
q
0
_
T
=
_
cos
1
d
2
b
+l
2
1
l
2
2
2d
b
l
1

1
2
0 0
_
T
(50)
which places the end-effector on a horizontal axis when the
angle of joint 2 is initialized as
20
= /2. The values of
the feedback gains for output zeroing control are chosen as
K
P
= 200.0 and K
D
= 30.0.
The results of the numerical simulations are shown in Fig. 7.
Because the control objective of this brief is to constrain the
end-effector on the geometric path, Fig. 7(a) shows its achieve-
ment with the geometric trajectory of the ball. The solid line
in Fig. 7(a) represents the geometric trajectory of the ball held
by the end-effector, while the dashed line shows the geometric
trajectory of the ball thrown off the end-effector. Fig. 7(b) and
(c) shows that rotation of joint 1 induces extension of joint
2 when the unstable zero dynamics drives the ball held by
the end-effector along the geometric path in a vertical plane.
This implies that the throwing motion control of the springed
Pendubot is successfully achieved by the designed controller,
according to the biomechanics of human throwing motion. The
ball is thrown off the end-effector in a desired direction at the
release speed of 14.39 m/s (51.80 km/h) in 0.120 s with the
release angle of 0.049 rad.
(a)
(b)
(c)
Fig. 7. Simulation results for the sigmoidal path with design parameters
d
a
= 3.65, d
b
= 0.39, and d
c
= 0.40. (a) Geometric trajectory of the ball.
(b) Time behavior of the angular velocities. (c) Time behavior of the angular
velocities.
VI. EXPERIMENTS
This section describes experimental verication for the
throwing motion control of the springed Pendubot to
demonstrate the effectiveness of the proposed control strategy.
The controller designed in Section III was implemented
to set up the experiment, which had to be well designed to
follow the dynamic model used in the theoretical analysis.
The mechanism of the experimental setup, which implements
the assumptions in the numerical simulations is also introduced
in this section. See [8] and [9] for details of the experiment.
A. Experimental Setup
The overview and details of the experimental setup are
shown in Fig. 8. The whole body of the springed Pendubot
represents the superior limb of a human body where links 1
and 2 represent the upper arm and the forearm, respectively.
The angles of joints 1 and 2 are measured by rotary encoders,
and the angular velocities are calculated by differentiating
956 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 21, NO. 3, MAY 2013
(a)
(b)
Fig. 8. Experimental setup of the Pendubot. (a) Experimental setup with the
links, joints, and ball. (b) Experimental setup with the motor, encoder, and
spring.
the measured angles. An AC servo motor having the input
saturation at 79.2 Nm is mounted on joint 1 to generate the
controlled torque u. A torsion spring is mounted on the passive
joint of the Pendubot representing the exibility of the cubital
joint.
Because the dynamic model is classied into underactuated
mechanical systems, it is not possible to maintain the initial
conguration assumed in the numerical simulations. An
electric magnet, therefore, is mounted on the passive joint
of the Pendubot to constrain the initial angle by applying
electromagnetic force, as shown in Fig. 9. The assumption
of the instantaneous release of the ball is implemented with
an electric magnet, and a compression spring as shown in
Fig. 10. In this mechanism, as the electric magnet is turned
off, the compression spring helps the ngers release the ball
when the release condition is satised. Design parameters
d
a
, d
b
, and d
c
, and feedback gains for output zeroing control
K
P
and K
D
are given the same values used in the numerical
simulations.
(a)
(b)
Fig. 9. Mechanism of the passive joint using an electric magnet to maintain
the initial conguration of the springed Pendubot. (a) Before. (b) After.
(a)
(b)
Fig. 10. Mechanism of the end-effector using an electric magnet and a
compression spring to realize instantaneous release of the ball. (a) Before.
(b) After.
B. Experimental Results
Fig. 11 shows photographs of the throwing motion of
the springed Pendubot with the end-effector driven by the
unstable zero dynamics along the geometric path. Fig. 12(a)
and (b) shows that the rotation of joint 1 induces the extension
of joint 2 when the unstable zero dynamics generates the
dynamic acceleration of the ball. This result implies that the
control objective of this brief was successfully achieved by
the designed controller based on the concept of unstable zero
dynamics. The ball is thrown off the end-effector in a desired
direction at the release speed of 14.02 m/s (50.47 km/h) in
0.121 s with the release angle of 0.022 rad. The causes for the
SHOJI et al.: THROWING MOTION CONTROL OF THE SPRINGED PENDUBOT 957
Fig. 11. Series of the throwing motion control of the springed Pendubot.
(a)
(b)
Fig. 12. Experimental results for throwing motion control of the Pendubot.
(a) Time behavior of joint angles. (b) Time behavior of angular velocities.
differences between numerical simulations and experimental
verication are the viscous friction at the joints of the experi-
mental setup, and the input saturation of the AC servo motor
at 79.2 Nm.
VII. CONCLUSION
This brief investigated throwing motion control of the
springed Pendubot based on the concept of unstable zero
dynamics. The Pendubot with a torsion spring mounted on
the passive joint was investigated, as a dynamic model of
the superior limb to imitate human throwing motion. The
observable dynamics of the springed Pendubot from the
chosen output was partially linearized via inputoutput lin-
earization. The unobservable nonlinear dynamics, on the other
hand, was intentionally destabilized to generate the dynamic
acceleration of the ball held by the end-effector. Numerical
simulations and experimental results conrmed the effec-
tiveness of the designed controller based on zero dynamics
analysis.
REFERENCES
[1] Y. Hong, International Research in Sports Biomechanics. Evanston, IL:
Routledge, 2002.
[2] M. Feltner and J. Dapena, Dynamics of the shoulder and elbow joints
of the throwing arm during a baseball pitch, Int. J. Sports Biomech.,
vol. 2, no. 4, pp. 235259, 1986.
[3] S. Sakurai, Biomechanics of overhand throwing motion: Past, present,
and future research trend, in International Research in Sports Biome-
chanics. Evanston, IL: Routledge, 2002.
[4] A. Shiriaev, J. W. Perram, and C. Canudas-de-Wit, Constructive tool
for orbital stabilization of underactuated nonlinear systems: Virtual
constraints approach, IEEE Trans. Autom. Control, vol. 50, no. 8,
pp. 11641176, Aug. 2005.
[5] L. Freidovich, A. Robertsson, A. Shiriaev, and R. Johansson, Peri-
odic motions of the pendubot via virtual holonomic constraints: The-
ory and experiments, Automatica, vol. 44, no. 3, pp. 785791,
2008.
[6] U. Mettin, P. X. L. Hera, L. B. Freidovich, and A. S. Shiriaev, How
springs can help to stabilize motions of underactuated systems with weak
actuators, in Proc. IEEE 47th Conf. Decision Control, Cancun, Mexico,
Dec. 2008, pp. 49634968.
[7] T. Shoji, S. Nakaura, and M. Sampei, Throwing motion control of
the springed pendubot via unstable zero dynamics, in Proc. IEEE Int.
Conf. Control Appl. Part Multi-Conf. Syst. Control, Yokohama, Japan,
Sep. 2010, pp. 16021607.
[8] S. Katsumata, S. Ichinose, T. Shoji, S. Nakaura, and M. Sampei,
Throwing motion control based on output zeroing utilizing 2-link
underactuated arm, in Proc. Amer. Control Conf., Jun. 2009, pp. 3057
3064.
[9] S. Ichinose, S. Katsumata, S. Nakaura, and M. Sampei, Throw-
ing motion control experiment utilizing 2-link arm with passive
joint, in Proc. SICE Annu. Conf., Tokyo, Japan, Aug. 2008, pp.
32563261.
[10] M. W. Spong and D. J. Block, The pendubot: A mechatronic system for
control research and education, in Proc. IEEE Conf. Decision Control,
Dec. 1995, pp. 555556.
[11] E. R. Westervelt, J. W. Grizzle, C. Chevallereau, J. H. Choi, and
B. Morris, Feedback Control of Dynamic Bipedal Robot Locomotion,
1st ed. Boca Raton, FL: CRC Press, 2007.
[12] E. R. Westervelt, J. W. Grizzle, and D. E. Koditschek, Hybrid zero
dynamics of planar biped walkers, IEEE Trans. Autom. Control, vol. 48,
no. 1, pp. 4256, Jan. 2003.
[13] R. Anami, S. Nakaura, and M. Sampei, Swing up control for the acrobot
considering compliance of high bar and energy interaction with each
component, in Proc. IEEE 46th Conf. Decision Control, Dec. 2007,
pp. 19291936.
[14] A. Isidori, Nonlinear Control System, 3rd ed. New York: Springer-
Verlag, 1995.
[15] H. K. Khalil, Nonlinear Systems, 2nd ed. Englewood Cliffs, NJ: Prentice
Hall, 1996.
[16] M. W. Spong, Swing up control of the acrobot, in Proc. IEEE Int.
Conf. Robot. Autom., San Diego, CA, May. 1994, pp. 23562361.
[17] D. B. Dacic and P. V. Kokotovic, Path-following for linear systems with
unstable zero dynamics, Automatica, vol. 42, no. 10, pp. 16731683,
2006.
[18] D. B. Dacic, M. V. Subbotin, and P. V. Kokotovic, Path-following
for a class of nonlinear systems with unstable zero dynamics,
in Proc. IEEE 43rd Conf. Decision Control, Dec. 2004, pp.
49154920.

Anda mungkin juga menyukai