c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

AIAA Guidance, Navigation, and

Control Conference and Exhibit A01-37204 AIAA 2001-4393
6-9 August 2001 Montreal, Canada

Guidance Algorithms for Autonomous Rendezvous of Spacecraft with a Target Vehicle in Circular Orbit
Hari B. Hablanr, Myron Tapper*, David Dana-Bashian**
The Boeing Company
5301 Bolsa Avenue, Huntington Beach, CA

Abstract
This paper presents algorithms for autonomous guidance of spacecraft to approach, to fly around, and to depart from a
target vehicle in a circular orbit. The algorithms are based on the closed-form solution of linear Clohessy-Wiltshire equations.
The approach and departure algorithms are adaptations of the glideslope guidance used in the past for rendezvous and
proximity operation of Space Shuttle with other vehicles with astronauts in the guidance loop. The multi-pulse glideslope
algorithms in the paper are general, capable of effecting a translation motion of spacecraft in any direction in space
autonomously, decelerating if approaching the target or a nearby location, and accelerating if receding away from it. The
flyaround algorithm enables the spacecraft to circumnavigate a target spacecraft in any plane, the orbit plane and the local
horizontal plane being two special cases thereof. The circumnavigation is performed in a specified period using a specified
number of pulses; the larger the number of pulses, the smaller the deviation of flyaround from the specified radius of
circumnavigation. The implementation of these algorithms requires estimates of position and velocity of the spacecraft
relative to the target. This relative navigation is performed with an extended Kalman filter using range and angle
measurements of target relative to the spacecraft focal plane and IMU and accelerometer measurements. The corresponding
measurement models and process noise matrix are provided. Several scenarios are simulated to illustrate the guidance
algorithms and relative navigation.
rendezvous. This topic is treated in many textbooks 12-17? as
I. Introduction well as in Ref. 18. Nevertheless, this rendezvous is briefly
This paper presents simple guidance algorithms to described here because it is the cornerstone of the so-called
approach, to fly around, and to depart autonomously from a glideslope and flyaround guidance algorithms developed in
target spacecraft in a circular orbit. Such algorithms have Sec. Ill and Sec. IV. A glideslope is a straight path from the
been developed in the context of Space Shuttle. U 2 current location of the chaser spacecraft to its intended
Pearsonl, for instance, emphasized many practical concerns destination, which may be a target spacecraft center of mass,
which far outweigh the fuel-minimization for rendezvous, a docking port, or a location of interest in space near the
and presented a pragmatic glideslope algorithm for an in- target orbit. The history, motivation and analysis of the
plane rendezvous using canted thrusters. During the Apollo glideslope technique in the context of the Space Shuttle is
era, rendezvous of Lunar Modules with Command and given in Ref. 1. The glideslope analysis therein is limited to
Service Modules were performed^, but such rendezvous the guidance of the chaser in the orbit plane, using canted
were especially designed for lunar liftoff and were also thrusters. Pearson formulated a relationship between the
assisted by astronauts. Of late, detailed studies, experiments glideslope angle, thruster cant angle, range and range rate.
in space, and hardware-in-the-loop laboratory For analysis, the Clohessy-Wiltshire equations in rectilinear
demonstrations of autonomous rendezvous, proximity coordinates were transformed into polar coordinates.
operation, and docking have proliferated both within the Section III generalizes this algorithm. Here we present,
U.S.4-7 and overseas.^-10 Despite these efforts, however, using a matrix formulation instead of the polar formulation
the published literature appears to lack a simple, straight of Ref. 1, a general multi-pulse guidance algorithm to move
forward, mathematical analysis of guiding a chaser a chaser vehicle in the vicinity of a target vehicle,
spacecraft to approach, to fly around, and to depart from a decelerating if approaching the target, and accelerating if
target spacecraft. The present paper attempts to fill this receding away from it. The motion is in any general
need. The guidance algorithms presented here are perhaps direction, not limited to tangential direction (V-bar) or radial
simple generalizations and extensions of those in Refs. 1, 2, direction (R-bar), nor restricted to the target orbit plane.
4 and are based on classical linear Clohessy-Wiltshire Further, it is assumed that, unlike the canted thrusters in
equations. Consistent with the pragmatism of Ref. 1, Ref. 1, six independent thrusters are available to produce an
advanced optimal control techniques, such as those in incremental momentum vector in any direction.
Ref. 11 to minimize fuel consumption using a primer vector, Section IV is concerned with guidance algorithms for
are not called upon here. flying around a target spacecraft autonomously. A chaser
The paper is comprised of five sections. Section II satellite can circumnavigate a target satellite in an in-plane
summarizes the analytic steps of a classical two-impulse elliptic path in one orbit period, with the target at its
f
Technical Fellow, Flight Systems Design & Analysis, Reusable Space Systems
Senior Staff Scientist, Phantom Works Advanced Vehicle Design
** Senior Engineer Scientist, Flight Systems Design & Analysis, Reusable Space Systems

Copyright © 2001, by H.B. Hablani, M. Tapper, and D. Dana-Bashian. Published by the American Institute of Aeronautics
and Astronautics Inc., with permission.

1
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
center. 19 Under ideal conditions, this elliptic path, once
established with proper initial velocity, persists without any
additional thruster firing. The largest distance of the chaser
from the target is along local horizontal, this distance being
twice the shortest distance along local vertical. This
geometry and the period of circumnavigation, however, may
not always be compatible with the mission requirements.
The mission designer may instead require a faster, circular
flyaround, perhaps also not necessarily in the orbit plane. As
such, Sec. IV presents the formulations for three kinds of
flyarounds: natural in-plane elliptic, circular in-plane, and
circular in any plane.
The algorithms of Sees. Ill and IV can be implemented
only if the target's orbit and relative location of the chaser
are known. Further, real rendezvous, proximity operations,
and docking, though lately commonplace with the advent of
International Space Station, are extraordinarily complex
events, for they include, among other things, sensing of the
target by the sensors onboard the chaser, inertial navigation
of the chaser and the target vehicles, chaser-tar get relative
navigation, Kalman filtering with or without GPS receivers
on the two vehicles, attitude determination of both vehicles,
and more. Reference 2 describes these complexities for
rendezvous and docking of Space Shuttle with Mir with
astronauts in the loop. References 4-6, in contrast, detail
autonomous rendezvous and docking using GPS/INS,
visible/infrared, and video guidance sensors. In this paper,
we assume that the target is in a circular orbit and its
location known exactly, that the initial relative position of
the chaser before initiating any rendezvous operation is
known within some error sphere, and that the two vehicles
are point masses. Under these assumptions, Sec. V is
concerned with the estimation of position and velocity of the
chaser spacecraft relative to the target using range and angle
measurements of the target and extended Kalman filter.
Section VI illustrates the glideslope and circumnavigation
algorithms of Sees III and IV, with or without a Kalman
filter. Section VII concludes the paper.
II. Classical Two-Impulse Rendezvous
Referring to Fig. 1, suppose there is a target vehicle in a
circular orbit of angular velocity 00. Attached to the vehicle
target vehicle
target's orbital
circular
chaser rate CD
orbit vehicle
to Earth's
center of mass
Fig. 1. Local-vertical local-horizontal frame xyzto
describe chaser motion relative to the target
center of mass is a right-handed curvilinear coordinate
frame x y z, with the x-axis along the target velocity vector,
V, or orbit circumference; the z-axis radially downward
along the vector R to the earth's center of mass; and the y-
axis completing the right-handed frame. These notations and
sign conventions follow those in Ref. 19, though,
unfortunately, other notations and conventions are also rife
in the literature (see Ref. 13, for instance). The local vertical
curvilinear (LVC) frame, instead of the rectilinear frame,
might be preferred because the orbital arc distance x of the
chaser from the target then can be arbitrarily large. 12, 14
However, the chaser-target distances in this paper are so
small that the difference between the LVC and the local-
vertical-local horizontal (LVLH) frame is negligible, less
than the sensor noise. Hence the two frames will not be
distinguished in the paper. In the LVLH coordinate system,
the motion of a chaser spacecraft located at a station (x, y,
z), where y and z are much smaller than the target orbit
radius and x along the orbit circumference is not necessarily
small, is governed by the following Clohessy-Wiltshire
equations:
tangential forward: x - 2 c o z =a x
cross-track: y + CO2 y =a y (1)
radial (down): z +2 cox -3 co2 z = a7
x = y, y = z, z = x
where ax, ay, az are the acceleration components acting on
the chaser in the xyz frame. These linear equations are
derived and their properties investigated in several
textbooks 12-17, 19 and in Ref. 18. As is well known, the
cross-track motion y normal to the orbit plane is not coupled
with the in-plane motion along x and z. The closed-form
solution of Eq. (1) is readily available from Refs. 12-19. For
convenient manipulations, this analytical solution is
arranged in a vector-matrix form, and to do so, the position
and velocity vectors of the chaser vehicle are defined thus:
r = [x y z]T (2a)
r =[x y z]T (2b)
The vector r is not to be confused with the radial
downward direction R mentioned earlier. Let the initial
values of these vectors be, respectively, rQ and TQ . Using
the Clohessy-Wiltshire equations, the position and velocity
vectors at a later time t are given by
r(t) = * f f (t)r 0 + *ri(t)ro (3a)
r(t) = f c f r W r o + *s (t)r 0 (3b)
where the four 3x3 O matrices are given by 1
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

"l 0 6(cot-s)~ If r j (T) = 0, that is, if the chaser is commanded to arrive at

0 c 0 , the target station at t = T, the arrival velocity r^ (T) must be
0 0 4 -3c annihilated by imparting a pulse AV j equal to
4 2
— s-3t 0 -(1-c) (4) (8)
CO CO

—rr v 0 -1CO o so that, to effect docking, the net relative velocity at r_i (T) is
zero.
2 c
The two-impulse scheme just described is highly
—— (1-c) 0 -
CO CO idealized and unfit for a real rendezvous, but it is useful for
analyzing the transfer of a chaser vehicle in space near a
target circular orbit. Indeed, this scheme enables us to
0 0 6(0(1- implement the glideslope and circumnavigation guidance
0 -cos 0 algorithms in Sees. Ill and IV.
0 0 3cos III. General Multi-Pulse Glideslope Transfer
(4)
-3 + 4c 0 2s Common directions of approaching a target or
0 c 0 retracting from it are along the orbital motion in front of the
-2s 0 c target or from behind, popularly known as the V -approach,
and radial, from below or above the target, known as the
and s = sin cot, c = cos cot. The velocity rj required at r R-approach. But this relative motion could be in any
general direction, either in or out of the orbit plane,
(0) at time t = 0 to arrive at a specific location, r_j, in time T including normal to the orbit plane. See Pearson* for the
is obtained easily from Eq. (3a) 15; examples of in-plane glideslopes. Further, the travel does
not have to be directly to the center of mass of the target;
ij=*^ 1 (ri-* r r ro) (5) instead, the travel may be to some other location of interest.
Klumpp20 considers trajectory shaping by a sequence of
The initial velocity at TQ , denoted as TQ ^ TQ , is changed velocity increments, but necessary mathematical details are
not provided in the paper.
instantaneously to rj , as depicted in Fig. 2, by imparting an When a chaser vehicle is required to approach a target
vehicle, an inbound glideslope guidance is invoked. Like-
incremental velocity equal to 13 wise, for receding away from the target, an outbound
glideslope is called for. In both scenarios, thruster activity
(6) near the target is to be minimized so as to avoid plume
impingement on the target vehicle and contamination of its
target vehicle surfaces^. in addition, as a chaser approaches the target, its
relative velocity must diminish to certain safe limits. These
target requirements are fulfilled by designing a guidance trajectory
orbit wherein the range rate is proportional to the range 1.
chaser vehicle Reference 1 shows that, in a glideslope with continuous
path thrusting, this relationship, while linear for the most part, is
nonlinear near the end. In this paper, for ease of analysis, a
linear relationship between the range and range rate is
postulated to be the mission design goal, whether the motion
is in-plane or out-of-plane. Such guidance trajectories are
formulated below for both inbound and outbound
glideslopes.
Inbound Decelerating Glideslope
chaser Fig. 3 illustrates a target in a circular orbit and the
orbit associated LVC xyz frame at its center of mass. Relative to
Fig. 2. Two-Impulse Transfer of Chaser Vehicle this frame, at t = 0, the chaser satellite is located at TQ, with
from r0 to r/
its relative velocity equal to TQ . The chaser vehicle is
The arrival velocity at rj(T), denoted r^T), is then required to arrive at r = rj in a transfer time T with a
velocity specified below. A straight line from TQ to r/p,
furnished by Eq. (3b):
denoted the vector p in Fig. 3, is the most natural
(7) commanded path for this transfer. Let rc (t), measured from
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

target The glideslope guidance specifies the distance-to-go, p, as a

function of time, p (t), so that the chaser is commanded to
reach r^ from TQ in a period T with the arrival commanded
target orbit velocity p j U where p j, less than zero, is some pre-
determined safe relative speed of the chaser at the distance
I rj | from the target.
As the distance-to-go, p, diminishes, the speed p must
diminish with it. Here, p is obtained by differentiating p,
to Earth's i0^—- treating the LVLH frame as an inertial non-rotating frame.
center of initial location The following linear relationship between p and p is
mass postulated:
Fig. 3. Three-Axis Multi-Pulse Decelerating p = ap + pT (15)
Inbound Glideslope
the target center-of-mass, be the commanded location of the where a (units: I/second) is the slope of p vs. p (Fig. 4).
chaser on this path at time t, 0 < t < T. Then the boundary The boundary conditions of p and p are
values of rc are
@ t = 0: P = P, P = p<0 (16a)
Ic(0)=I 0 (9a)
p =pT<0 (16b)
rc(T) = rT (9b)
The initial distance-to-go, po, the initial commanded
The vector p(t) emanates from the tip of the vector r/p P
n (distance-to-go) PO
(Fig. 3), and it defines the commanded location of the
chaser on the straight path from TQ to r/p. The boundary t =T
conditions of p(t) are PT

Po t =0
p(T) = 0 (lOb)

and at any time t, Fig. 4. Reduction of the velocity p with time, as a linear
p(t) = r c (t)-r T (11) function of the distances-to-go, p (t)
velocity, po < 0 , and the final commanded arrival velocity,
Since rQ = [XQ y0 ZQ] and r-p = [xp yj zp] , the PT|)> are a
^ know11 °r specified. The
PT <
direction cosines of the vector p are given by slope a is then equal to
n _——P——————————————————
Ci
O~PT ^0
<v \J (17)
cos a =
PO
(12) The commanded path, Eq. (15), corresponds to a varying
commanded acceleration, p = a p , and since | p | is
decreasing with time, the acceleration (actually
deceleration) also decreases with time. These features of the
where PQ = |P0|. The direction of the straight path is then glideslope scheme are desirable. With the boundary
given by the unit vector Up conditions (16), the solution to Eq. (15) is

_ [cosa cosp cosy]iT (13) (18a)

and the scalar distance p, the distance-to-go, along the and the transfer time T is
vector p is
(18b)
p = pu n (14) PO
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

where a < 0 and PQ < p-p < 0. The algorithm to move the
•> x (local horizontal)
chaser from IQ to r/p can be developed now as follows.
Let the number of thruster firings to travel from rQ (p =
PQ) to r/p (p = 0) in time T be N and the uniform interval
between any two successive pulses be At = T/N. The
thrusters are thus fired at time tm = m At (m = 0, 1, ..., N-l), actual path, based
on CW equations
and the mth pulse pushes the chaser from rm (p = pm) to z
(local
Im+1 (P = Pm+l), where vertical) chaser vehicle
(19a) glide-slope path

(e -1) (19b)
Fig. 5. Three-Axis Multipulse Accelerating
Outbound Glideslope
The arrival velocity at mth location is denoted rm, and, in

accordance with Eq. (5), the departure velocity, r^, to

travel from rm to lm+\ is
PT t =T

(20)

where O^ = O^ (At) and (At). The incremental

velocity at rm is then Po t = 0
0 PT
(21)
Fig. 6. Postulated Acceleration of the chaser from the
The chaser will arrive at £ m+ j with velocity r^j equal to initial speed pQ to the final speed pj versus
the distance p
(22)
and ends at r j, and at any time t such that 0 < t < T, the
according to Eq. (7). The actual path of the chaser will not commanded glideslope path p (t) is defined as
be along the vector p , of course, but rather will result from
the Clohessy-Wiltshire equations, illustrated in Figure 3 as = I c (t)-r 0 (24)
the path with humps. Specifically, following Eq. (3a), the
path is given by
where, as before, rc (t) is the vector from the target center
= * n .(t-t m )r m +0 I f(t-t m )r+ (23) of mass to any point on the glideslope path. The boundary
m
conditions of p are now
Because the interval between any two successive pulses is
constant, the spacecraft will move progressively slower as it P(0)=Q (25a)
approaches the target.
Outbound Accelerating Glideslope
The geometry of this glideslope is shown in Fig. 5 and
Fig. 6. Compare this geometry with the geometry of
inbound glideslope shown in Fig. 3 and Fig. 4. The chaser
initial radial vector IQ is now closer to the target than the which yields the direction cosines of p equal to
destination radial vector r-p = r (T). The magnitude of the
velocity TQ at TQ is smaller than the specified speed pj at (26)
PT PT PT
r j. The vector p now emanates from the initial vector TQ
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

In order to accelerate the vehicle while traveling from TQ to z

max - T xmax (33b)
rj, the speed p is postulated to vary linearly with the
which define the size of the flyaround.
scalar distance p, thus,
In-Plane Circular
(27) The geometry and notations for this flyaround are
shown in Fig. 7. The chaser circumnavigates at a radius p
where p (0) = 0, p (T) = pj , and p(0) = pQ . The postulated from the target center of mass, and its instantaneous angular
speed at p(T) is pj , and the transfer time from p = 0 to location in the flyaround circle is denoted 0, where 9 = 0
PT is T. Applying Eq. (27) for time t = T, the positive when the chaser is on the circular orbit in front of the target.
Suppose the flyaround period is T and the circumnavigation
slope a is is effected with N pulses. Then the angle traveled between
the two pulses m and m + 1 (m = 0, 1, ..., N - 1) is
(28) A9 = 2 Ti/N. At t = 0, the chaser is at radius p at an angle OQ.
PT Subsequently, the angle 9 varies linearly as 9 = 9o + 2 n
Solving Eq. (27), one can show now that the distance p t/T, and the commanded location of the chaser in the in-
varies with time, thus, plane flyaround circle is
x = p cos 9 = p cos (9o +2ict/T) (34a)
p(t)=PO ( e *-i) (29)
z = -psin9 = psin(9o+27tt/T) (34b)
The transfer time T is still given by Eq. (18b), though a is
now positive, and py > po > 0.
The algorithm to transfer the vehicle from IQ to r_y is
similar to that for an inbound glideslope. N thruster firings chaser vehicle
effect the transfer, as before, the mth firing taking place at
r(t m ), where
iftm) ~ 10 + P(tm)^ p (30) >x
and p(t m ) is obtained from Eq. (29) by substituting t = tm. (local horiz.)
The calculation of incremental velocities for each firing
proceeds as in Eqs. (20) - (23).
IV. Circumnavigation

In-Plane Natural Elliptic Flyaround z (local

Using the solution, Eq. (1.17), Ref. 19, it can be shown vert.)
that the initial velocity components which a chaser must
have to initiate an elliptic, natural flyaround at any location
(x, z) are Fig. 7. In-Plane Circular Circumnavigation

(3 la) The mth pulse takes place at 9m = 9o + m A9, and the

corresponding chaser location is defined by the vector rm
CO
z = —— x equal to
(31b)
2
Under ideal conditions, the chaser then will traverse the r =p[cos9 m , 0, -sin9 m ] (35)
elliptic path, anticlockwise about the orbit normal y-axis,
= XQ cos cot + 2zo sin cot (32a) The strength of this pulse, causing a discrete increment in
velocity at rm, is calculated by specifying that, in time T/N,
z(t) = — XQ sin cot + ZQ cos cot (32b) the chaser arrives at rm-f i corresponding to the angle 9m+i
where (XQ, ZQ) is the initial location of the chaser at t = 0. = 9o + (m + 1) A9. Having defined rm and r m +l, it is now
The velocity components x,z will vary in accordance with straightforward to calculate the departure velocity r^ at
Eq. (31). The maximum x and z components of the elliptic
path are r m and the arrival velocity 1^+1 at r m + j using Eqs. (20)
2 2\^ 2 - (22), where the transition submatrices are calculated for t
(33a) = T/N.
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

Circular Flyaround in Any Plane Plane of fly-around

Since in-plane circumnavigation appears to be more
common, and since it does not excite the out-of-plane
motion, the y-axis appears to be a natural reference for
orienting the axis of an arbitrary flyaround plane. Two chaser
angles, 0za about the z-axis and 0Xa about the once- c \ /initial
displaced x-axis (x'-axis, Fig. 8), define the orientation of locatic
of chaser
the flyaround axis y" normal to the arbitrary
circumnavigation plane x" z" (Fig. 8). When both angles are
zero, the y"-axis aligns with the y-axis, and the plane x" x"
coalesces with the orbit plane xz, as desired. Table 1 records
three pairs of special values of the angles 0xa and 9za which
bring y"-axis in alignment with the LVLH x, y, or z-axis.
The components of the unit vector a along the general y"-
axis expressed in the LVLH frame are
u xa -cos 0xa sin 0za Fig. 9. Location of the chaser vehicle in a general
VI
= cos 0xa cos 0za (36) flyaround plane x "z "
u
za sin 0xa
t = 0, the chaser is located at XQ, yo, ZQ which corresponds
target > x
to the angles 0za, 0xa and 0y = 0yQ in the circumnavigation
orbit (local horiz.) plane x" z", in this order, in Fig. 9. To determine 0yQ, the
initial vector TQ =[XQ yo ZQ] T in the LVLH frame is
expressed in the circumnavigation frame x" y" z".
One can then show that, in this frame,
to the plane
x
of fly-around o X
0 +YO S0
za
H
(local vert.) yo 0 (38)
»
Z
0
Fig. 8. Orientation of the axis of flyaround in a
general plane
where c_ = cos_, s_ = sin_. In Eq. (38), because x" T!' is the
circumnavigation plane of the chaser, the y"-component of
Table 1. Three special flyaround axes and the the chaser location is zero, by definition. The initial angular
corresponding orientation angles location of the chaser, the angle 0yQ, is calculated from
fly-around
axis plane (rad) (rad) (39)
0 -71/2 X
0
xz 0 0
xy 71/2 0 At any later time t, the angle 0y in the x" z" plane equals
0 y =0yO+27it/T (40)
Sometimes it is easier to specify the unit vector a rather than
specify the angles. In that circumstance, the orientation where T is the period specified for the 2n radian
angles are determined from circumnavigation. The commanded, instantaneous location
of the chaser is now completely specified by the rotation
0xa = sin" A
za (37a) angle 0y, the orientation angles 0za and 0xa, and the
circumnavigation radius rc. In the LVLH xyz frame, this
instantaneous location r is given by
0za =tan" (37b)
l
ya c0za c0 y -s0 za s0xa s0y
where tan~l denotes the 4-quadrant arctangent function, ); 0xa, 0 za j= rc s0za C0 y +c0 z a s0xa s0y (41)
often denoted atan2. The commanded location of the chaser
CV/YJI S Wv
in the flyaround plane is defined by the angle 0y (Fig. 9). At
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
The relative location, r, Eq. (41), of the chaser is used for
the purpose of effecting the circumnavigation with N pulses.
As in the case of in-plane circumnavigation, the mth pulse
takes place at
t = 0,1,..., N-l) (42)
Substituting B v m in Eq. (41), we calculate the vector rm.
From then on, this general circumnavigation is commanded
similar to the commanding of the previously described in-
plane circumnavigation or glideslopes.
V. Relative Navigation
This section is concerned with the estimation of the
position and velocity of the chaser relative to the target. The
estimation is accomplished by measuring the range of the
target, the location of its image on the focal plane of the
chaser, and the attitude of the chaser focal plane in an
inertial frame or in the LVLH frame of the chaser. While the
focal plane measurements are obtained from cameras,
visible or infrared sensors, relative range measurements are
provided by lidar, and chaser attitude is furnished by IMU.
Blending all these measurements suitably, we determine the
location of the target relative to the chaser. A simple model
of these blended measurements is developed below.
Measurements Model
Fig. 10 depicts the chaser and the target in two
neighboring orbits, target located at rcT from the chaser.
Using measurement of the target image centroid on the focal
plane and focal plane attitude measurement from IMU
relative to the chaser LVLH x j y\ z\ frame, we can express
the vector ICT in the triad F\: x\ y\ z\ as
-I r
CTy r
CTz J (43)
chaser
Fig. 10. Target line-of-sight measurement in the chaser
LVLH frame
Because the lidar, visible and infrared sensors measure the
target location in terms of range and line-of-sight (LOS) angles,
Fig. 10 shows the two LOS angles 9yi (azimuth) and 0zi
(elevation), measured from the LVLH xi yi z\ frame, which
bring the local horizontal axis xi in alignment with the LOS
vector r^j. The angle 9yi about the axis yi is in the chaser
orbit plane and is thus called azimuth. The angle Qz\ about the
once-displaced zi-axis (that is, z\ -axis) accounts for any out-
of-orbit-plane component of the chaser-target vector r^ j and
is thus called elevation. In terms of these angles and the range r
|, the vector IT, Eq. (43), can be written also as
r£T =r [c0 y l c9 z l , s0 z l , -S0 y l c9 z l J T (44)
For the purposes of guidance using Clohessy- Wiltshire
equations, the angles 6yi and Qz\ and the range r are the
aforementioned blended measurements related to the
components ? x, ? y, z as follows:
\
~ rCTz 1
6yl tan ' V'
.= »(0,GV)
r Y
CTx J
l
6Z, sin +V
z' (45)
where Vy and vz are angle measurement noises comprised
of noise from the visible and infrared sensors and IMU, and
vr is the range measurement noise of lidar. In the absence of
any other more suitable model of noise, these noises are
assumed to be uncorrelated, white, with zero mean and
standard deviations equal to ay, az and ar, respectively.
The Clohessy-Wiltshire equations govern the position
(x, y, z) of the chaser satellite measured from the target
center of mass in the LVLH xyz frame located at the target
center of mass. Consequently, the LOS vector r^y and the
measurements, Eq. (45), must be all expressed in the target
LVLH xyz frame. The chaser LVLH xi yi z\ frame and the
target LVLH xyz frame, both shown in Fig. 10, are
generally not the same. However, at an altitude of 400 km,
when the target and chaser are 1 km or so apart, their orbital
angular separation is ~150 jirad, of the same order of
magnitude as the visible sensor noise variance (la =
120 jirad) or infrared sensor noise variance (la = 850 jirad
to 2 mrad). Hence, for such a relatively small separation
between a chaser and a target, the two LVLH frames are
parallel, and the LOS vector components in the target
LVLH frame will be x = -r^^ x> Y = ~rCT Y? z = ~rCT z?
changing the measurement equations (45) to
+v = hi(x,y,z)+v y
= h 2 (x,y,z) + vz (46)
r= = h 3 (x,y,z) + vr
Because these range and angle measurements are related
nonlinearly to the position coordinates (x, y, z), the
measurement functions hi, h2, h3 in Eq. (46) are nonlinear.
The estimation of position and velocity of the chaser using
these measurements and an extended Kalman filter requires
partial derivatives of the functions hi, h2, h3 with respect to
the state vector x:
x = [x y z x y z] (47)
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

In particular, rewriting the measurement Eq. (46) in requires a detailed investigation, possibly along the lines of
standard Kalman filter vector notations z = h(x) + v, we Refs. 5, 21, 22, this bias error in the knowledge of CO will be
require ignored here. Let wx, Wy, wz be the white, random
acceleration acting on the chaser, in addition to the
deterministic acceleration ocx, ay, az. The intensities of the
z 0 x random accelerations are denoted GWX* CTwy, Gwz,
4 "4 respectively, where
xy TIP yz
H=—= = o (48)
r 2 r, P r2 r2 rr
r
IP
-3x3 = x,y,z) (49)

_x y. _Z
r r r with 8 being the Dirac delta operator. The deterministic
Jr(-) equations (1) then become

tangential, forward (x-axis):

where rjp equals in-plane range, r^ = x 2+ z2 , H is a 3x6
matrix, and Q3x3 is a 3x3 matrix of zeros. The noise vector x-2coz =oc x + w x wx
v equals y = [Vy vz vr]T? and the associated measurement
noise matrix R for Kalman gain computations is R = diag cross-track (y-axis):
ar2] . y+G)2y =ocy + wy wy

radial, downward (z-axis):

Process Noise Model
The Clohessy-Wiltshire equations (1) assume that the
accelerations ax, av, ocz are deterministic and known
exactly. When thrusters are fired, however, there is likely to
be a minute randomness in the force produced. Also, the
estimate of the target orbit rate, CO, is likely to contain a
slight random bias error. Since target orbit estimation
z + 2 c o x - 3 c o 2 z =az + wz wz ~ ^ (0, awz)
Eqs. (50) are rewritten in standard first-order state-space
form, with state x defined by Eq. (47). The corresponding
6x6 discrete process noise matrix Q k can be then shown to
be23, 24

TT3

wy

(51)

Qk =
0

where T = sample period. The coupling between the in- 0) ~ 0.001 rad/s, and for sample period T = 1 s, ooT = 0.001
plane coordinates x and z is retained in Eq. (51). This rad « 1 rad. For this reason, the coupling terms in Q k,
coupling, though, is weak because for an altitude of 400 km,
Eq. (51), can be ignored. If we further assume that the
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

random acceleration has the same intensity in all three axes, z vs. x (pulse behavior)
-150
the matrix Q k then simplifies to
-100
: T^
- 0 0 0 0
j 2
li3 o ! o zi2 0

m.
T3
——
3
! o 0 zi2 (52)

! T 0 0

sym. T 0
-100 -200 -500
x(m)
: sym. T Figure 11. Inbound, outbound V glideslopes an
four-pulse in-plane circumnavigation

where a'w 2 2a^ = ua2

w =~ ^wx ~ wyY —uwz The three axes in this and (0, ±100 m) in the xz plane. The larger the number of
simplified Q k are now uncoupled. The process noise matrix pulses, the better the approximation to a circular flyaround,
as illustrated later. The phase plane x vs. x corresponding
associated with each axis agrees with the 2x2 single-axis to the glideslopes and flyaround in Fig. 11 is shown in
process noise matrix in Ref. 23. Fig. 12. The dashed x vs. x straight glideslope paths
VI. Illustrations illustrate the p vs. p linear variations in Fig. 4 and Fig. 6.
This continuous increase/decrease in the velocity x is
The preceding guidance algorithms are illustrated realized, on the average, by discrete increments/decrements
below for different scenarios, first with neither noise nor a caused by thruster firings. The circumnavigation over one
Kalman filter, and then with noise and a Kalman filter. The orbit period is effected with four pulses, taking place at
target is assumed to be flying at a 400 km altitude (oo ~ (±100 m, 0) and (0, ±100 m) in the orbit plane. At z =
0.001 rad/s, orbit period ~ 92.56 minutes). ±100 m, AV is along z , as shown in the phase plane z vs.
z, in Fig. 13. The exponential decrement of x, and its
Scenario 1. Inbound/Outbound V Glideslope and In- average, discrete counterpart, vs. time during inbound
plane Circumnavigation glideslope over 9 minutes (t = 2000 to t = 2540 s), are
Fig. 11 illustrates this scenario. Inbound 1 glideslope shown in Fig. 14.
starts at x = -500 m behind the target and ends at x = -100
m, effected by ten pulses in 9 minutes decelerating the xdot vs. x (pulse behavior)
chaser from several m/s to 0.05 m/s (shown in detail later).
Then, after perfect stationkeeping at x = -100 m, the chaser
performs a 100 m four-pulse circumnavigation in about one
orbit period (~92.53 minutes). At the end thereof, the
inbound 2 glideslope is performed, taking the chaser from x
= -100 m to -25 m in the next ten pulses in 9 minutes. The
chaser then perfectly stationkeeps behind the target for 3
minutes, followed by an outbound accelerating glideslope
from -25 m to -500 m in ten pulses over 9 minutes. One
observes in Fig. 11 that the theoretically straight paths,
corresponding to the vectors p in Fig 3 and Fig. 5, are, in
reality, accompanied with humps, diminishing in the case of commanded
£ glideslope
the decelerating glideslope and growing in the case of the
accelerating glideslope. While the theoretical straight path
assumes continuous thrusting, the actual glideslope is
performed with a finite number of pulses, with coasting 100 -100 -200 -300 -500
x(m)
between them, and the humps arise during coasting. We also
observe that four pulses do not effect a flyaround with a Figure 12. Phase Plane x vs. x during three V
constant radius, the radius being 100 m only at (±100 m, 0) glideslopes and a four-pulse 100m flyaround

10
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

-0.2 -400

-0.15
^glidesbpes -200

-0.1

-0.05
200
-z-pulse-.:.... z-pulse
I 0.05

0.1 600
flyaround

0.15

0.2
t=0
100 50 0 -50 -100 -150
z(m)
1000 800 600 400 200 0 - 2 0 0 - 4 0 0
x(m)
Figure 13. Phase plane Z vs. z during three V
glideslopes and four pulse 100 m flyaround Figure 15. 3-axis inbound glideslope and 300 m 24-pulse
flyaround in (local horizontal) xy plane

z vs. y (circumnavigation behavior)

2600

300 -200 -300

Figure 14. Exponential discrete deceleration in
inbound 1 glideslope
Scenario 2. 3-Axis Inbound Glideslope and Flyaround
in Local Horizontal Plane
Figure 15 illustrates a three-axis glideslope and a 300 m
flyaround in 30 minutes using equi-spaced pulses. At t=0,
the chaser is located at [1, 1, 1]* km station, from where it
is commanded to arrive at [300, 0, 0] * m on the x-axis using
four pulses in 5 minutes, decelerating the vehicle from 11.56
m/s to 0.084 m/s. The projection of this motion in the local
horizontal xy plane is shown in Fig. 15. Unlike the four-
pulse in-plane flyaround in Fig. 11, the 24-pulse 30-minute
flyaround in Fig. 15 following the glideslope is essentially a
circle, commanded here in the xy plane (which is turning
about y-axis in an inertial frame at the orbit rate). Because
of coupling between x and z, the chaser experiences local
vertical motion along z while flying around in the xy plane.
This motion, under control by z-component of the pulses, is
shown in Fig. 16.
Figure 16. Z-motion induced in a 300 m 24-pulse
flyaround in xy plane
Scenario 3. Four-pulse 100 m Flyaround in 30 minutes
Using Sensors and Kalman Filter
Figs. 17-23 illustrate this scenario. A total of four
sensors, comprising two visible sensors, one infrared sensor,
and one lidar sensor, measure the relative location of the
target. The visible and infrared sensors provide angle and
passive range measurements sequentially at approximately
4 Hz, whereas the lidar sensor provides range measurements
at 1 Hz. The measurement noise matrix, R, for each sensor is:
= dia
^visible S [(0.001)2 rad2? (0.001)2 rad2, (0.08r)2 m2]
Binfrared = Bvisible (53)
Blidar = (°-5 meter)2
where r = range. The initial estimation error in the state
vector x (position: m, velocity: m/s) are

11
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

zest/z true Vs. Time

3 so

I -50

-100

200 400 600 800 1000 1200 1400 1600 1800

Time (Seconds)
vzest/Vz true Vs. Time

50 0 -50 0 200 400

x-Vbar (Meters)

Figure 17. Four-pulse 30-minute in-planeflyaround Figure 19. z and z and their estimates z and z
using range, azimuth, and elevation angle vs. time for flyaround
measurements and Kalman filter
x est/x true Vs. Time
y est/y true Vs. Time

I -50

-100

200 400 600 800 tOOO 1200 1400 1600 1800

Time (Seconds) 800 1000 1200 1400 1600 1800
Time (Seconds)
vx est/vx true Vs. Time
3 vyest/vy true Vs. Time
10-
5

-5 ......;.......;.. --•••-------:•••-•--------••••-•--••-•-;---•--

^^-i-^-^J

0 200 400 800 1000 1200 1400 1600 1800

Time (Seconds) -20o 200 400 600 800 1000 1200 1400 1600 181
Time (Seconds)

Figure 18. x and x and their estimates x and x vs. time Figure 20. y and y and their estimates y and y vs. time
for flyaround
for flyaround
The results in Figs. 17-23 pertain to the just-mentioned
x = [ 1 0 m , 0.5m, 7m, -0.001 m/s,
specifications. The impulses for circumnavigation are
-0.0001 m/s, -0.003 m/s] T (54)calculated using the estimated position, either from the
and the covariance matrix P is initialized as Kalman filter or from knowledge of the initial conditions,
not the true position of the chaser. As such, the chaser is not
P = diag [33.3 m2, 33.3 m2, 33.3 m2, likely to reach its intended place exactly. In Fig. 17, AV
0.0001 (m/s) 2, 0.0001 (m/s)2, 0.0001 (m/s)2] (55)calculated at t = 0 to move the chaser from (100, 0) m to (0,
0, -100) m in t = 450 s using the initial position estimate
The incremental velocity vector, AV, demanded by the according to the estimation error (54) takes the vehicle, not
guidance algorithm, is produced by thrusters with 1% error to its intended place exactly, but in the neighborhood
and measured by the accelerometer with 2% error. These thereof. Aided by the sensors, however, particularly the
errors are compensated for by specifying the process noise lidar, the initial estimation errors subside to an optimum
acceleration variance a^xT^CJ^z^ == (0.004 m/s)2 and level determined by the ratio of the process noise matrix Q

a =10" 3 wx in Eqs. (51-52). The sample period T and the measurement noise matrix R specified above. See
X(

in Fig. 17 that at t = 900, 1350, and 1800 s, the chaser

varies but is nominally equal to 0.25 s. essentially reaches its intended location. The variation of the

12
 c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
Estimated Error Function x est-x true Vs. Time
600 800 1000 1200 1400 1600 1800
Time (Seconds)
Estimated Error Function z est-z true Vs. Time
Time (Seconds)
Figure 21. Position estimation errors x and z and the
square root of their variances vs. time for flyaround
Estimated Error Function vx est-vx true Vs. Time
600 800 1000 1200
Time (Seconds)
Estimated Error Function vz est-vz true Vs. Time
0 200 400 600 800 1000 1200 1400 1600 1800 Eqs. (54)-(55) are possibly unrealistic. As an experiment, all
Time (Seconds)
Figure 22. Velocity estimation errors x and y and the
square root of their variances vs. time for flyaround
true and the estimated position coordinates x, y, z and
x, y, z, respectively, and the velocity components x, y, z ,
and x, y, z , respectively, versus time is shown in Figs. 18-
20. Because of active range measurements with lidar, the
initial position estimation errors, Eq. (54), are removed
quickly, for the true and the estimated position coordinates
in Figs. 18-20 do not appear distinguishable. Some
estimation errors in velocity estimates persist, however,
particularly due to the 2% bias error in accelerometer
measurements of AV, as seen in Figs. 18-20. The two
significant discrete steps in x and its estimate x at t = 0
and 900 s in Fig. 18 change the direction (slope) of
x-motion near x = ±100 m, seen in Fig. 17. Likewise, the
discrete steps in z and z in Fig. 19 at t = 450 s and 1350 s
change the direction of z-motion near z = ±100 m in Fig. 17.
The in-plane (x, z)-motion is not coupled with the y-motion,
but because the chaser is initially 5 m out-of-plane, it is
brought in-plane, and y is decreased to nearly zero in just
the first three pulses, at t = 0, 450, and 900 s, as depicted in
Fig. 20. Finally, the position and velocity estimation errors,
x, y, z , and x, y, z , respectively, and the positive and
2 2
negative square roots of their variances, P(x ), P(y ),
P ( z 2 ) , and P(£ 2 ), P(^ 2 ), P(? 2 ), respectively, are
shown in Figs. 21-23. Interestingly, all variables concerning
x and z exhibit a cyclic behavior, reflecting the cyclic
variation of x and z in the flyaround, and the estimation
errors remain within the bounds of their respective standard
deviations.
Estimated Error Function y est-y true Vs. Time
1400 1600 1800
Figure 23. Estimation errors y and y and their
variances vs. time for flyaround
The velocity estimation errors and their variances in
three components of the velocity errors were increased to -
0.02 m/s and the corresponding variances to 0.0025 (m/s)^,
but the results shown in Figs. 17-23 did not change
noticeably.
Scenario 4. Inbound/Outbound Glideslopes,
Circumnavigation, and Stationkeeping Using Sensors
and Kalman Filter
Figs. 24-26 illustrate a scenario similar to Scenario 1,
but now using sensors and Kalman filter, as described in
Scenario 3, and a circumnavigation using 24 pulses. Fig. 24
depicts all phases of motion: inbound and outbound
glideslopes, flyaround, and Stationkeeping. The duration of
each phase appears in Fig. 24. Because of active range
measurement and relatively small measurement errors, the
true and the estimated x and z position components in Fig.
24 are almost indistinguishable. The three glideslopes are
shown at a magnified scale in Fig. 25, where the estimates
x and z are seen to track the true x and z well. The random
drift during Stationkeeping in the interval t = 5 to t =
6 minutes at 100 m caused by thruster/accelerometer
13
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.

imperfections and estimation errors in x, y, z, is shown

24-pulse
30 minute more clearly in Fig. 26. For tighter control of the drift of the
fly a round
chaser while stationkeeping, a closed-loop translation
: / inbound 2 : controller is required but is not included here.
glideslope •
inbound 1 38 < t < 43 mjn ;
» glideslope Conclusions
stationkeep .
stationkeep : *——— :;43 < t < 4 4 min The autonomous guidance algorithms developed and
i:<t<6rtiin \ outbound "
0 <t< 1 min
/ \ 44<t<46mjn illustrated in the preceding for glideslope and
stationkeep
;.6<t <.7 rain ; circumnavigation are simple, for they are based on classical
:37<t<38min Clohessy-Wiltshire equations, their solutions, and elementary
: 46 < t < 47 mirt
vector and matrix analysis of kinematics. Different NASA
Centers and industries seem to be in possession of similar
algorithms but, apparently, the algorithms are not available in
the literature. Notwithstanding the simplicity of the
algorithms, however, their implementation entails complex
350 300 250 200 150
x(m)
100 50 0 -50 -100
sensor technology for inertial and relative navigation. The
Figure 24. In-plane inbound/outbound glideslopes, relative navigation is briefly treated in the paper, utilizing
24-pulse circumnavigation, and stationkeeping using range, azimuth and elevation measurements and extended
sensors and Kalman filter Kalman filter. But the initial relative position and velocity
estimates, to be used in relative navigation, are extracted from
inertial navigation of the chaser and the target, and this
100m
complex subject is not considered in the paper. Also, in
fjyaround addition to sensing the target by visible or infrared sensors,
the designer/analyst must consider IMU measurements,
attitude dynamics of both chaser and target, their attitude
determination, and attitude control of chaser to fire the
thrusters in the right direction. Some of these topics will be
addressed in the future.
inbound 1
glideslope Acknowledgement
The authors are grateful to Mr. Thomas A. Mulder and
his associates, Rendezvous and Proximity Operations
Group, The Boeing Company, Houston, for sharing their
rendezvous techniques used on the Space Shuttle over the
outbound
stationkeep
glideslope
last nearly two decades. Also, the authors are indebted to
2U——————————I——————————I——————————I————-.-.

300 250 200 150

- ...1-————————I——————————L_.__._.
100 50 0
.....I. _
-50
J. Brazzel, Boeing, F. Clark and P. Spehar, Lockheed
x(m) Martin, Houston, Texas, for their briefing material entitled
Figure 25. Glideslopes magnified RPOP Guidance Algorithms, February 24, 2000. Their work
is the cornerstone of the study reported here.
References
1. Pearson, D.J., "The Glideslope Approach," Advances in
the Astronautical Sciences, American Astronautical
Society Paper No. AAS 89-162, pp. 109-123.
2. Zimpfer, D., and Spehar, P., "STS-71 Shuttle/Mir GNC
Mission Overview," Advances in the Astronautical
Sciences, Space Flight Mechanics 1996, Paper No. AAS
96-129, Vol. 93, Part I, pp. 441-460.
3. Young, K.A., "Apollo Lunar Rendezvous," AlAA J.
Spacecraft and Rockets, Vol. 7, No. 9, September 1970,
pp. 1083-1086.
4. Cruzen, C.A., Lomas, J.J., and Dabney, R.W., "Test
Results for the Automated Rendezvous and Capture
System," Advances in the Astronautical Sciences,
110 108 106 104 102 100 98 96 Guidance and Control 2000, Paper No. AAS 00-003, pp.
35-56, Vol. 104, 2000.
5. Upadhyay, T., Cotterill, S., and Deaton, A.W.,
Figure 26. Glideslopes and stationkeeping in front of the "Autonomous Reconfigurable GPS/INS Navigation and
target at x = 100 m Pointing System for Rendezvous and Docking," AIAA

14
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.
92-1390, Space Programs and Technologies Conference,
March 1992, Huntsville, AL.
6. Howard, R.T., Bryan, T.C., Brook, M.L., and Dabney,
R.W., "The Video Guidance Sensor: A Flight Proven
Technology," AAS 99-025, pp. 281-298.
7. Calhoun, P., and Dabney, R., "A Solution to the Problem
of Determining the Relative 6 DOF State for Spacecraft
Automated Rendezvous and Docking," SPIE Vol. 2466,
1995, pp. 175-185.
8. Kawano, L, Mokuno, M., Kasai, T., and Suzuki, T.,
"Result and Evaluation of Autonomous Rendezvous
Docking Experiments of ETS-VII," Proceedings, AIAA
Guidance, Navigation, and Control Conference, Paper
No. 99-4073, August 1999, Portland, OR.
9. Philip, N.K., Ananthasayanam, M.R., and Dasgupta, S.,
"Study of Relative Position and Attitude Estimation and
Control Scheme for the Final Phase of an Autonomous
Docking Mission," IF AC Automatic Control in
Aerospace, Seoul, Korea, 1998, pp. 185-193.
lO.Serrano-Martinez, J.B., "Use of Simulation Tools and
Facilities for Rendezvous and Docking Missions,"
AGARO Flight Vehicle Integration Panel Symposium on
Space Systems Design and Development Testing, Cannes,
France, Vol. CP-561, October 3-6, 1994, pp. 17.1-17.12.
11.Prussing, J.E., and Chiu, J.H., "Optimal Multi-Impulse
Time-Fixed Rendezvous Between Circular Orbits,"
AIAA J. Guidance, Control, and Dynamics, Vol. , Jan.-
Feb. 1986, pp. 17-22.
12.Kaplan, M.H., Modern Spacecraft Dynamics and
Control, John Wiley & Sons, Inc., 1976, Sec. 3.6,
pp. 108-115.
13.Prussing, I.E., and Conway, B.A., Orbital Mechanics,
Oxford University Press, New York, 1993, Chapter 8.
15
14.Chobotov, V.A., (Ed.), Orbital Mechanics, AIAA
Education Series, 1991, Chapter 7.
15.Wiesel, W.E., Space/light Dynamics, The McGraw-Hill
Companies, 1997, Section 3.5.
16.Wie, B., Space Vehicle Dynamics and Control, AIAA
Education Series, 1998, Section 4.6.
17.Vallado, D.A., Fundamentals of Astrodynamics and
Applications, The McGraw-Hill Companies, 1997,
Section 5.8.
18.Mullins, L.D., "Initial Value and Two-Point Boundary
Value Solutions to the Clohessy-Wiltshire Equations,"
The Journal of the Astronautical Sciences, Vol. 40, No.
4, October-December 1992, pp. 487-501.
19.Bryson, A.E., Jr., Control of Spacecraft and Aircraft,
Princeton University Press, Princeton, New Jersey, 1994,
Chapter 1.
20.Klumpp, A.R., "Trajectory Shaping Rendezvous
Guidance," IEEE Aerospace and Electronics Systems
Magazine, 2 (2): 17-22, 1987.
21.Park, Y.W., Brazzell, J.P., Jr., Carpenter, J.R., Hinkel,
H.D., and Newman, J.H., Flight Test Results from Real-
Time Relative Global Positioning System Flight
Experiment on STS-69, NASA TM-104824, Nov. 1996.
22. Garrison, J.L., and Axelrod, P., "Application of the
Extended Kalman Filter for Relative Navigation in an
Elliptic Orbit," Advances in Astronautical Sciences,
Paper No. AAS 96-142, pp. 693-711.
23.Bar-Shalom, Y., and Li, X-R, Estimation and Tracking:
Principles, Techniques, and Software, YBS 1998, pp.
262-263
24.Gelb. A., (Ed.), Applied Optimal Estimation, The M.I.T.
Press, 1974, Section 3.6.