Escolar Documentos
Profissional Documentos
Cultura Documentos
S
S!
d
S
u
dt
S
(5)
Rather than solving the control input function u
i
(t) over the
control horizon, the control is parameterized with its S derivatives.
Because u
i
(t) is not required to appear in the state equations in a
linear manner, it can be considered to be a state variable. This
converts determination of a continuous function for control into
determination of a nite and relatively small number of discrete
parameters to determine control. Subsequently this converts the
resulting optimal control problem into a discrete parameter
optimization problem. The parameters to be determined for the
optimal control problem then become
U
i
=
du
i
dt
d
2
u
i
dt
2
d
S
u
i
dt
S
_ _
T
(6)
U = U
1
U
2
U
N
_ _
T
(7)
Assuming that parameters in Eq. (6) dening the control sequence
are known, the control is expressed in Eq. (8).
u
i
(t )
_ _
du
i
dt
d
2
u
i
dt
2
. . .
S1
(S 1)!
d
S
u
i
dt
S
_
d (8)
In practical applications the control is only needed near the current
time ( =0). In this case, only the rst derivative is required to
determine the optimal control input.
u
i
(t
2
)
du
i
dt
(t
2
t
1
) u
i
(t
1
) (9)
To develop compact expressions for the control input, the Taylor
series expansion for the output in Eq. (4) are written in as
y
i
(t ) T
i
Y
i
(10)
T
i
= 1
1!
2
2!
R
i
R
i
!
_ _
(11)
Y
i
= y
i
(t)
dy
i
dt
d
R
i y
i
dt
R
i
_ _
T
(12)
Consider the entire output vector, approximated by a Taylor series
up to order R
i
(possibly different orders for each output).
y(t )
T
1
0 0
0 T
2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0
0 0 T
p
_
_
_
_
_
_
_
_
_
_
Y
1
Y
2
.
.
.
Y
P
_
_
=TY (13)
The matrix
T has dimensions of [P V|, where V =
N
i=1
R
i
. The
desired trajectory is approximated in the same manner.
y
D
(t ) TY
D
(14)
1180 SLEGERS, KYLE, AND COSTELLO
The tracking error can now be formed and expressed in compact
notation
e(t ) =
Y
D
(t )
Y(t ) =
T(
Y
D
Y) (15)
In model predictive control, a quadratic cost function is minimized
over a nite horizon.
J =
1
2
_
T
2
T
1
e(t )
T
Qe(t ) d
=
1
2
_
T
2
T
1
(
Y
D
Y)
T
T
T
Q
T(
Y
D
Y) d (16)
Tracking error cost is weighted with the possibly time-dependent
positive denite matrix Q. Selecting the components of Qto be time
varying allows initial or nal tracking performance to be more
heavily weighted. Because only
T, and possibly Q, depend on , the
cost function can be rewritten in Eq. (17).
J =
1
2
(
Y
D
Y)
T
(
Y
D
Y) (17)
The matrix
i
is given in Eq. (18) for a constant diagonal Q
matrix.
i
=q
i
_
t
2
t
1
T
T
i
T
i
d
=q
i
t
2
t
1
(1)0!0!
t
2
2
t
2
1
(2)0!1!
. . .
t
R
i
1
2
t
R
i
1
1
(R
i
1)0!R
i
!
t
2
2
t
2
1
(2)1!0!
t
3
2
t
3
1
(3)1!1!
t
R
i
2
2
t
R
i
2
1
(R
i
2)1!R
i
!
t
R
i
1
2
t
R
i
1
1
(R
i
1)R
i
!0!
t
R2
2
t
R2
1
(R
i
2)R
i
!1!
t
2R
i
1
2
t
2R
i
1
1
(2R
i
1)R
i
!R
i
!
_
_
_
_
_
_
_
_
(18)
=
1
0
2
.
.
.
0
P
_
_
_
_
_
_
_
(19)
Note that the matrix can be computed in closed formand stored.
For the parameters
U to be an optimal solution, the cost function in
Eqs. (1719) must satisfy the gradient condition
@J
@
U
=(
Y
D
Y)
T
@
Y
@U
=0 (20)
The optimal control input is determined by considering Eq. (20).
Partition
Y,
Y
D
, and @
Y=@
U into upper and lower sections as shown
in Eqs. (2123).
Y
i
=
_
Y
U
i
Y
L
i
_
=
_
i
0
(x)
.
.
.
i
M
i
1
(x)
i
M
i
(x; u
1N
)
i
M
i
1
(x; u
1N
)
i1
(x) _ u
1
iN
(x) _ u
N
.
.
.
i
R
i
(x; u
1N
; . . . ; d
(S1)
u
1N
=dt
(S1)
)
i1
(x)
d
S
u
1
dt
S
iN
(x)
d
S
u
N
dt
S
_
_
(21)
@Y
i
@U
j
=
_
_
0 0 0
0 0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0
i1
0 0 0
@
i
M
i
2
@u
1
j
i1
0 0
.
.
.
.
.
.
0
@
i
R
i
@u
1
j
@
i
R
i
@u
2
j
i1
_
_
=
_
0
+
i1
_
(22)
@
Y
@
U
=
0 0 0
+
11
+
12
+
1N
0 0 0
+
21
+
22
+
2N
.
.
.
0 0 0
+
P1
+
P2
+
PN
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
(23)
In a similar manner partition to be conformal with the vectors
Y
and
Y
D
.
=
1
11
1
12
0
1
21
1
22
.
.
.
P
11
P
12
0
P
21
P
22
_
_
_
_
_
_
_
_
_
_
_
_
_
(24)
The format of @
Y=@
U and allows for their multiplication to be
written in compact form.
@
Y
@
U
=
1
12
0 0
. . .
1
22
0 0
0
2
12
0
2
22
.
.
.
.
.
.
0 0
P
12
. . .
0 0
P
22
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
+
11
+
12
+
1N
+
21
+
22
+
2N
.
.
.
.
.
.
.
.
.
.
.
.
+
P1
+
P2
+
PN
_
_
_
_
_
_
_
=
R
dY
(25)
With the selection of R
i
such that the number of Taylor expansion
terms in the control approximation for each control is equal and the
SLEGERS, KYLE, AND COSTELLO 1181
system is square, i.e., the number of inputs and outputs are equal,
dY is a square S P matrix and
R
has dimensions of
(V P) (S P). The equivalent optimal condition is then
expressed as follows.
(
Y
D
Y)
T
R
dY =0 (26)
Where dY
1
exists, we have the following condition.
(
Y
D
Y)
T
R
=0 (27)
Because of the block structure
R
can be expanded as in Eq. (28).
The resulting P conditions for optimality are shown in Eq. (29).
__
Y
U
Di
Y
L
Di
_
_
Y
U
i
Y
L
i
__
T
_
i
12
i
22
_
=[0| (28)
Y
L
i
=(
i
22
)
1
(
i
12
)
T
(Y
U
Di
Y
U
i
) Y
L
Di
(29)
If we are interested only in the current control input, then _ u is the
only derivative required which is the rst component of Y
L
i
. Denoting
the rst row of (
i
22
)
1
(
i
12
)
T
as K
i
and evaluating the rst
component of Y
L
i
the optimal solution for the rst time derivative of
the control u is given in Eqs. (3037).
B =
11
11
.
.
.
.
.
.
.
.
.
P1
PN
_
_
_
_
_ (30)
A =
1
M
1
1
N
M
P
1
_ _
T
(31)
Y
L1
D
= Y
L
D1
(1) Y
L
DP
(1)
_ _
T
(32)
Y
U
D
=
Y
U
D1
.
.
.
Y
U
DP
_
_
_
_
_ (33)
Y
U
=
Y
U
1
.
.
.
Y
U
P
_
_
_
_
_ (34)
U
C
= _ u
1
(t) _ u
N
(t)
_ _
T
(35)
K =
K
1
0 0
0 K
2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0
0 0 K
P
_
_
_
_
_
_
_
_
(36)
Using this notation, the MIMO control law is
U
C
=B
1
[K
T
(Y
U
D
Y
U
) Y
L1
D
A| (37)
In the case of a SISO system, Eq. (37) reduces to the following
scalar result for the optimal control derivative (13).
_ u =
1
M1
_
K(Y
U
D
Y
U
)
d
M1
y
D
dt
M1
M1
_
(38)
III. Application to a Rigid Air Vehicle Model
Many air vehicles can be modeled as a rigid body possessing six
degrees of freedom (DOF) including three inertial position
components of the system mass center as well as the three Euler
orientation angles. Kinematic equations of motion for the general
system are provided in Eqs. (39) and (40).
_
_
_
_ x
I
_ y
I
_ z
I
_
_
_
=R
T
IB
_
_
_
u
b
v
b
w
b
_
_
_
(39)
_
_
_
_
_
_
_
=
1 s
0 c
0 s
=c
=c
_
_
_
_
_
_
_
p
q
r
_
_
_
=K
RM
! (40)
The matrix R
IB
represents the transformation matrix from an
inertial reference frame to the body reference frame.
R
IB
=
c
s c
_
_
_
_
(41)
The common shorthand notation for trigonometric functions is
employed where sin() s
, cos() c
, and tan() t
. The
dynamic equations of motion are provided in Eqs. (42) and (43).
_
_
_
_ u
b
_ v
b
_ w
b
_
_
_
=
1
m
_
_
_
X
Y
Z
_
_
_
g
_
_
_
s
_
_
_
0 r q
r 0 p
q p 0
_
_
_
_
_
_
_
u
b
v
b
w
b
_
_
_
(42)
_
_
_
_ p
_ q
_ r
_
_
_
=I
1
_
_
_
L
M
N
_
_
_
0 r q
r 0 p
q p 0
_
_
_
_
I
T
_
_
_
p
q
r
_
_
_
_
_
_
_
(43)
Where
_
_
_
X
Y
Z
_
_
_
=
1
2
S
R
V
2
A
(F
A
F
C
u) (44)
_
_
_
L
M
N
_
_
_
=
1
2
S
R
V
2
A
(M
A
M
C
u) (45)
_
_
_
_
L
_
M
_
N
_
_
_
=S
R
V
A
_
V
A
(M
A
M
C
u)
1
2
S
R
V
2
A
(
_
M
A
_
M
C
u M
C
_ u)
(46)
A common goal of air vehicles is to track desired roll, pitch, and
yaw angles. The output for an angle tracking controller is given in
Eq. (47).
y =
_ _
T
(47)
The relative degree of each output is 2; therefore three derivatives
of the output equations are required. The rst derivative of the output
is Eq. (40), the second derivative is not shown, and the third
derivative follows:
y
:::
=2
_
K
RM
_ !K
RM
!
K
RM
! (48)
The matrix B in Eq. (30) and the vector A in Eq. (31) are found to
be
A =2
_
K
RM
_ !
K
RM
!K
RM
I
1
_
S
R
V
A
_
V
A
(M
A
M
C
u)
1
2
S
R
V
2
A
(
_
M
A
_
M
C
u) S
_ !
I! S
!
I _ !
_
(49)
B =
1
2
S
R
V
2
A
K
RM
I
1
M
C
(50)
Using Eqs. (48) and (49) we can solve for the desired control
derivatives by substitution into Eq. (37).
1182 SLEGERS, KYLE, AND COSTELLO
U
C
=
2
S
R
V
2
A
M
1
C
_
IK
1
RM
[Y
L1
D
K(Y
U
D
Y
U
) 2
_
K
RM
_ !
K
RM
!|
S
R
V
A
_
V
A
(M
A
M
C
u)
1
2
S
R
V
2
A
(
_
M
A
_
M
C
u)
S
_ !
I! S
!
I _ !
_
(51)
The matrices S
!
and S
_ !
are the skew-symmetric cross product
operators on !and _ !, respectively. The quantities F
A
and F
C
which
include aerodynamic force coefcients do not directly appear in
Eq. (51) but will enter in through
_
V
A
. Two matrix inversions are
required in the control lawand fortunately one is easily determined in
closed form:
K
1
RM
=
s
0 1
s
0
c
0
_
_
_
_
(52)
It is clear fromEq. (51) that the optimal control derivatives U
C
are
valid for a general rigid air vehicle where only the force and moment
equations depend on the specic system. The optimal control
derivatives U
C
can be found for a specic system provided that
aerodynamic expressions for F
A
, F
C
, M
A
, and M
C
are known.
IV. Parafoil and Payload Aircraft Application
A parafoil and payload controlled by left and right parafoil brake
deection can be modeled as a SISO system where heading angle is
the output and asymmetric brake deection
a
is the control. The
parafoil and payload shown in Fig. 1 can be represented as a 6-DOF
system with aerodynamic forces acting at the total system mass
center and aerodynamic moments about the systemmass center. The
aerodynamic loads are given in Eqs. (53) and (54). Apparent mass
effects which become negligible in near steady-state conditions have
been neglected [17].
_
_
X
Y
Z
_
_
=
1
2
S
R
V
A
(C
L0
C
L
)
_
_
w
a
0
u
a
_
1
2
S
R
V
A
(C
D0
C
D2
2
C
Da
a
)
_
_
u
a
v
a
w
a
_
_
(53)
_
_
_
L
M
N
_
_
_
=
1
2
S
R
V
2
A
_
_
C
l
b
C
lp
b
2
p
2V
A
C
m0
c C
m
c
C
mq
c
2
q
2V
A
C
nr
b
2
r
2V
A
_
_
C
la
b
d
0
C
na
b
d
_
a
_
_
_
_
_
_
(54)
Because of the symmetry of the parafoil system the inertia matrix
takes the form
I =
I
XX
0 I
XZ
0 I
YY
0
I
XZ
0 I
ZZ
_
_
_
_
(55)
I
1
=
I
XXI
0 I
XZI
0 I
YYI
0
I
XZI
0 I
ZZI
_
_
_
_
(56)
For a SISO system the control derivatives are given in Eq. (38).
The parafoil system has relative degree of 2, requiring three
derivatives of the output equation to nd
3
and
3
. Taking the three
derivatives of the output and writing
_
L,
_
M, and
_
N in compact form, y
:::
can be written in the desired form of Eq. (3) with
3
and
3
given in
Eqs. (60) and (61).
_
_
_
_
L
_
M
_
N
_
_
_
=
_
_
_
_
L
_
M
_
N
_
_
_
_
_
_
_
L
_
M
_
N
_
_
_
_ u (57)
_
_
_
_
L
_
M
_
N
_
_
_
=
1
2
S
R
_
_
2V
A
_
V
A
C
l
b V
2
A
C
l
b
_
_
V
A
C
lp
b
2
p
2
V
A
C
lp
b
2
_ p
2
2V
A
_
V
A
C
la
b
a
d
2V
A
_
V
A
C
m0
c 2V
A
_
V
A
cC
m
V
2
A
cC
m
_
_
V
A
C
mq
c
2
q
2
V
A
C
mq
c
2
_ q
2
_
V
A
C
nr
b
2
r
2
V
A
C
nr
b
2
_ r
2
2V
A
_
V
A
C
na
b
a
d
_
_
(58)
_
_
_
L
_
M
_
N
_
=
S
R
b
2d
_
_
_
V
2
A
C
l
a
0
V
2
A
C
n
a
_
_
_
(59)
3
=
_
1
c
__
S
2
q c
q 2c
_
_ q
_
2
S
q
c
2
S
q t
_
c
_
q t
_
S
_ q c
2
r S
r 2S
_
q_ r
_
2
c
r
c
2
c
r t
_
S
_
r
t
_
c
_ r
_
_
t
_
(c
_
q S
_ q t
_
S
q S
_
r c
_ r t
_
c
r)
_
c
_
{I
XZI
[
_
L
(I
YY
I
ZZ
)( _ qr q_ r) I
XZ
( _ pq p_ q)|
I
ZZI
[
_
N
(I
XX
I
YY
)( _ pq p_ q) I
XZ
( _ qr q_ r)|]
_
S
_
I
YYI
[
_
M
(I
ZZ
I
XX
)( _ pr p_ r) 2I
XZ
(p _ p r_ r)| (60)
3
=
_
c
_
(I
XZI
_
L
I
ZZI
_
N
)
_
S
_
I
YYI
_
M
(61)
The optimal solution for the rst time derivative of the control
a
given in Eq. (38) is now found for the parafoil and payload system.
For
_
a
to be nite it must be veried that
3
is not zero throughout the
expected ight envelope, i.e., the roll angle must not be =2 and
_
L
,
_
M
, and
_
N
are not all zero. It is clear from Eq. (59) that two
conditions could make
_
L
and
_
N
zero: V
A
being equal to zero and
the aerodynamic control coefcients C
l
and C
n
_
_
b
2
__
C
L
bC
Lp
p
2V
A
bC
Lr
r
2V
A
_
c
_
C
M0
C
M
cC
Mq
q
V
A
_
_
b
2
__
C
N
bC
Np
p
2V
A
bC
Nr
r
2V
A
_
_
_
(65)
[M
C
| =
dC
La
2
0 0
0 cC
M
f
cC
M
e
bC
Na
2
0 0
_
_
_
_
_ (66)
Fig. 8 Desired heading angle geometry for path tracking.
Fig. 9 Tracking a straight path and 27 deg turn.
Fig. 10 Yaw and desired yaw angle for path tracking.
Fig. 11 Yaw and desired yaw angle for path tracking.
Fig. 12 Glider schematic.
1186 SLEGERS, KYLE, AND COSTELLO
The matrix M
c
contains the systems control moment coefcients
and may in general be singular. In the glider considered, the elevator
and ap controls are redundant with respect to control moments,
therefore the coefcient matrix is singular. In this case the
pseudoinverse may be used in the optimal control solution to nd the
best t solution.
The preceding system of equations describing the rigid glider are
numerically integrated using a fourth-order RungeKutta algorithm
to generate trajectories of the system. One example scenario is
examined using the physical parameters provided in Table 3, and the
aerodynamic properties provided in Table 4. In the following results,
the control derivative is updated every 0.01 s.
Figures 1316 show simulation results for the glider tracking a
constant heading angle of 60 deg, roll angle of 0.34 times the heading
angle error, and a pitch angle of 22:5 sin(t) 6 deg. The values
1 2 1
_ _
are used in the error weighting matrix, the control is approximated
with an eighth-order Taylor series expansion, and the prediction
horizon is 1.06.0 s.
As in the parafoil case, the prediction horizon and Taylor series
expansion orders are used to balance control magnitude and tracking
error. Heading angle and pitch angle converge to their desired values
within 7 s. To allow a more aggressive time varying pitch trajectory
to be easily tracked, a constant weighing factor of 2 is applied to the
pitch angle error compared to unity for both the heading and roll
angle errors.
VI. Conclusions
A nonlinear model predictive control strategy for tracking a
desired orientation trajectory was developed for a general rigid air
vehicle. The control strategy was simulated for an autonomous
parafoil and payload system and an autonomous glider. The
performance of the controller was evaluated by varying prediction
horizons and number of Taylor expansion terms used for the
approximation of the output equation and the control sequence. It
was observed that a penalty for control can be implemented through
the selection of the number of Taylor series expansion terms even
with control being absent in the cost function. The selection of the
number of terms in the Taylor series expansion was limited only by
the conditioning of the matrix
22
. Also, the limitation on the size of
S can be circumvented because a nearly equivalent controller to that
for a large S and large prediction horizon can be found for a small S
with the appropriate prediction horizon. This observation leads to
two possible methods for selecting S and the prediction horizon. The
rst method: select a desired prediction horizon specic to the plant
and then choose S to achieve an appropriate penalty on the control
sequence. The second method: if S cannot be chosen large enough in
the rst method, then a nearly equivalent controller can be designed
by selecting S as large as reasonable so that the matrix
22
is well
Table 3 Glider physical properties
Variable Value Units
0.00238 slug ft
3
Weight 1.99 lbf
S 4.35 ft
2
b 5.50 ft
c 0.557 ft
I
XX
0.0548 slug ft
2
I
YY
0.0288 slug ft
2
I
ZZ
0.0813 slug ft
2
I
XZ
5:78e 005 slug ft
2
I
XY
5:05e 006 slug ft
2
I
YZ
1:93e 006 slug ft
2
Table 4 Glider airfoil parameters
Main wing airfoil RG-15
Main wing aps Trailing edge individual control
T-tail airfoil NACA 0009
T-tail aps Trailing edge elevator only
Fig. 14 Glider pitch angle time history.
Fig. 13 Glider heading angle time history. Fig. 15 Glider bank angle time history.
SLEGERS, KYLE, AND COSTELLO 1187
conditioned and then choosing an appropriate prediction horizon so
as to achieve a suitable control penalty. Another more complex
example is reported for an autonomous glider system in which
elevator and ap controls were redundant. The redundancy made the
control matrix M
C
singular. It was successfully demonstrated that a
pseudoinverse can be used.
References
[1] Dogan, A., and Venkataramanan, S., Nonlinear Control for
Reconguration of Unmanned-Aerial-Vehicle Formation, Journal of
Guidance, Control, and Dynamics, Vol. 28, No. 4, 2005, pp. 667678.
[2] Chaudhuri, A., and Seetharama Bhat, M., Output Feedback-Based
Discrete-Time Sliding-Mode Controller Design for Model Aircraft,
Journal of Guidance, Control, and Dynamics, Vol. 28, No. 1, 2005,
pp. 177181.
[3] Innocenti, M., Pollini, L., and Turra, D., Guidance of Unmanned Air
Vehicles Based on Fuzzy Sets and Fixed Waypoints, Journal of
Guidance, Control, and Dynamics, Vol. 27, No. 4, 2004, pp. 715720.
[4] Ikonen, E., and Najim, K., Advanced Process Identication and
Control, Marcel Dekker, New York, 2002, pp. 181197.
[5] Chen, H., and Allgwer, F., Quasi-Innite Horizon Nonlinear Model
Predictive Control Scheme with Guaranteed Stability, Automatica: the
Journal of IFAC, the International Federation of Automatic Control,
Vol. 34, No. 10, 1998, pp. 12051217.
[6] Cloutier, J. R., State-Dependent Riccati Equation Techniques: An
Overview, Proceedings of the American Controls Conference, Vol. 2,
IEEE, Piscataway, NJ, 1997, pp. 932936.
[7] Sznaier, M., Cloutier, J., Hull, R., Jacques, D., and Mracek, M., A
Receding Horizon State Dependent Riccati Equation Approach to
Suboptimal Regulation of Nonlinear Systems, Proceedings of the 37th
IEEE Conference on Decision & Control, Vol. 2, IEEE, Piscataway,
NJ, 1998, pp. 17921797.
[8] Kouvaritakis, B., Cannon, M., and Rossiter, J. A., Non-Linear Model
Based Predictive Control, International Journal of Control, Vol. 72,
No. 10, 1999, pp. 919928.
[9] Brooms, A. C., and Kouvaritakis, B., Successive Constrained
Optimization and Interpolation in Non-Linear Model Based Predictive
Control, International Journal of Control, Vol. 73, No. 4, 2000,
pp. 312316.
[10] Cannon, M., Kouvaritakis, B., Lee, Y. I., and Brooms, A. C., Efcient
Non-Linear Model Based Predictive Control, International Journal of
Control, Vol. 74, No. 4, 2001, pp. 361372.
[11] Magni, L., De Nicolao, G., Scattolini, R., and Allgwer, F., Robust
Model Predictive Control for Nonlinear Discrete-Time Systems,
International Journal of Robust and Nonlinear Control, Vol. 13,
Nos. 34, MarchApril 2003, pp. 229246.
[12] Xin, M., and Balakrishnan, S. N., A New Method for Suboptimal
Control of a Class of Nonlinear Systems, Proceedings of IEEE
Conference on Decision and Control, Vol. 3, IEEE Control System
Society, IEEE, Piscataway, NJ, 2002, pp. 27562761.
[13] Patwardhan, A. A., Rawlings, J. B., and Edgar, T. F., Nonlinear Model
Predictive Control, Chemical Engineering Communications, Vol. 87,
1990, pp. 123141.
[14] Patwardhan, A. A., and Madhavan, K. F., Nonlinear Model Predictive
Control Using Second-Order Model Approximation, Industrial and
Engineering Chemistry Research, Vol. 32, No. 2, Feb. 1993, pp. 334
344.
[15] Mutha, R. K., Cluett, W. R., and Penlidis, A., Nonlinear Model-Based
Predictive Control of Control Nonafne Systems, Automatica: the
Journal of IFAC, the International Federation of Automatic Control,
Vol. 33, No. 5, 1997, pp. 907913.
[16] Chen, W. H., Predictive Control of a General Nonlinear SystemUsing
Approximation, IEE Proceedings-Control Theory and Applications,
Vol. 151, No. 2, March 2004, pp. 233239.
[17] Lissaman, P. B. S., and Brown, G. J., Apparent Mass Effects on
Parafoil Dynamics, AIAA Paper 93-1236, 1993.
Fig. 16 Glider control deection time history.
1188 SLEGERS, KYLE, AND COSTELLO