Professional Documents
Culture Documents
9.1 Introduction
The attitude determination and control system (ADCS) is one of key subsystems of space-
craft or satellites, and it consists of an attitude determination system (ADS) and an attitude
control system (ACS). The ADCS provides the stabilization and control of attitude (orien-
tation) of spacecraft using a variety of sensors and actuators in the presence of disturbance
torques. An attitude and orbit control system (AOCS) is also considered as one of space-
craft subsystems. The guidance, navigation, and control (GNC) of spacecraft is an area of
space technology playing a key role to the success of space missions dealing with rendezvous,
docking, and proximity operations.
This chapter introduces the principles of spacecraft ADCS and GNC systems by em-
phasizing practical examples of such systems. The fundamentals of attitude determination
are discussed in Section 9.2 and ADCS hardware descriptions are presented in Section 9.3.
The principles of GNC systems are treated in Section 9.4. A comprehensive treatment of
spacecraft attitude determination theory and applications can be found in [1]. Spacecraft
attitude dynamics and control problems of practical interest can be found in [2, 3].
As an example of the ADS, the Inertial Stellar Compass (ISC), developed by Draper
Laboratory for NASA’s New Millennium Program ST-6 flight validation experiment [4] is
briefly described here. Its performance was successfully flight validated aboard the TacSat-2
satellite launched on December 16, 2006.
The ISC is a miniature, low-power ADS developed for use with low-cost small satellites. It
was designed to be suitable for a wide range of future missions because of its low-mass, low-
power, and low-volume design and its self-initializing, autonomous operational capability.
1
The ISC is composed of a wide field-of-view active-pixel star camera and microgyros, with
associated data processing and power electronics, as illustrated in Fig. 9.1. Periodic updates
from the star camera are used to correct the e↵ect of gyro drift and bias on obtaining the
attitude quaternion information from the rate gyros. The unique feature of the ISC is that
those two miniaturized devices are integrated into a very low-mass and low-power unit with
a microprocessor. It has a total mass of 2.5 kg, a power requirement of 3.5 W, and an
accuracy of 0.1 deg (1 ). A detailed description of the ISC attitude determination system
can be found in [4].
The Guidance, Navigation, and Control (GNC) system controls the motion of the space-
craft centre of mass. The GNC and ADCS sub-systems are analyzed separately as it is
common practice (Deep Space 1 [8], Deep Impact [9], Hayabusa [10]). We will present the
interactions between ADCS and GNC in the application examples. It is worth noting that
in some references the position and attitude determination and control system is referred to
as GNC system. The on-board GNC system is responsible of the following tasks:
• Guidance creates the reference path to achieve the desired goal in nominal conditions.
In order to accomplish these tasks, the GNC system includes the sensors providing mea-
surements, the GNC software implemented in the on-board computer and the actuators.
The GNC software includes the navigation filter, the guidance and control algorithms and
additional functions such as the measurement management and the actuator management.
Figure 9.2 shows the typical GNC architecture. More details will be given in Section 9.4.
2
Figure 9.1: Block diagram illustration of an attitude determination system, called the Inertial
Stellar Compass (ISC), developed by Draper Laboratory under NASA’s New Millennium ST-
6 project [4].
3
Figure 9.2: Block diagram illustration of ADCS/GNC system.
Consider a reference frame A with a right-handed set of three orthogonal unit vectors
{~a1 , ~a2 , ~a3 } and a reference frame B with another right-handed set of three orthogonal unit
vectors {~b1 , ~b2 , ~b3 }. Basis vectors {~b1 , ~b2 , ~b3 } of B are expressed in terms of basis vectors {~a1 ,
~a2 , ~a3 } of A as follows:
2 3 2 32 3 2 3
~b1 C11 C12 C13 ~a1 ~a1
6 7
6 ~b 7 = 6 76
4 C21 C22 C23 5 4 ~a2 7
5=C
B/A 6
4 ~a2 7
5 (1)
4 2 5
~b3 C31 C32 C33 ~a3 ~a3
where CB/A ⌘ [Cij ] is called the direction cosine matrix which describes the orientation
of B relative to A. The direction cosine matrix CB/A is also called the rotation matrix
or coordinate transformation matrix to B from A. Such a coordinate transformation is
symbolically represented as:
CB/A : B A
For brevity, we often use C for CB/A . Since each set of basis vectors of A and B consists
of orthogonal unit vectors, the direction cosine matrix C is an orthonormal matrix; thus, we
have
1
C = CT (2)
which is equivalent to
CCT = I = CT C (3)
T
In general, a square matrix A is called an orthogonal matrix if AA is a diagonal matrix,
and it is called an orthonormal matrix if AAT is an identity matrix. For an orthonormal
1
matrix A, we have A = AT and |A| = ±1.
4
For an arbitrary vector ~r expressed as
y = Cx (5)
where C is the direction cosine matrix of B relative to A, and y and x are the two corre-
sponding component vectors defined as
2 3 2 3
y1 x1
y = 4 y2 7
6
5 ; x = 4 x2 7
6
5
y3 x3
Three elementary rotations about the 1st, 2nd, and 3rd axis, respectively, of the reference
frame A are described by the following rotation matrices:
2 3
1 0 0
6
C1 (✓1 ) = 4 0 cos ✓1 sin ✓1 75 (6a)
0 sin ✓1 cos ✓1
2 3
cos ✓2 0 sin ✓2
6 7
C2 (✓2 ) = 4 0 1 0 5 (6b)
sin ✓2 0 cos ✓2
2 3
cos ✓3 sin ✓3 0
C3 (✓3 ) = 6
4 sin ✓3 cos ✓3 0 7
5 (6c)
0 0 1
where Ci (✓i ) denotes the direction cosine matrix C of an elementary rotation about the ith
axis of A with an angle ✓i .
Euler Angles
One scheme for orienting a rigid body to a desired attitude is called a body-axis rotation; it
involves successively rotating three times about the axes of the rotated, body-fixed reference
frame. The first rotation is about any axis. The second rotation is about either of the
two axes not used for the first rotation. The third rotation is then about either of the two
axes not used for the second rotation. There are 12 sets of Euler angles for such successive
rotations about the axes fixed in the body.
Consider three successive body-axis rotations to describe the orientation of a reference
frame B relative to a reference frame A. A particular sequence chosen here is symbolically
5
represented as:
C1 (✓1 ) C2 (✓2 ) C3 (✓3 )
where Ci (✓i ) indicates a rotation about the ith axis of the body-fixed frame with an angle
✓i .
The rotation matrix to B from A, or the direction cosine matrix of B relative to A, is
then defined as
In general, Euler angles have an advantage over direction cosines in that three Euler
angles determine a unique orientation, although there is no unique set of Euler angles for a
given orientation.
Quaternion
Consider Euler’s eigenaxis rotation about an arbitrary axis fixed both in a body-fixed refer-
ence frame B and in an inertial reference frame A. A unit vector ~e along the Euler axis is
defined as
where ei are the direction cosines of the Euler axis relative to both A and B, and e21 +e22 +e23 =
1.
Then we define the four Euler parameters or the quaternion as follows:
q1 = e1 sin(✓/2) (9a)
6
q2 = e2 sin(✓/2) (9b)
q3 = e3 sin(✓/2) (9c)
q4 = cos(✓/2) (9d)
where ✓ is the rotation angle about the Euler axis. Similar to the eigenaxis vector e =
(e1 , e2 , e3 ), we define a vector q̄ = (q1 , q2 , q3 ) and the quaternion vector q = (q1 , q2 , q3 , q4 )
such that
✓
q̄ = e sin (10)
" 2#
q̄
q= (11)
q4
Note that the four Euler parameters are not independent of each other, but constrained by
the relationship
qqT = q̄T q̄ + q42 = q12 + q22 + q32 + q42 = 1 (12)
The direction cosine matrix can also be parameterized in terms of the quaternion as
follows: 2 3
1 2(q22 + q32 ) 2(q1 q2 + q3 q4 ) 2(q1 q3 q2 q4 )
C B/A
= C(q) = 4 2(q2 q1 q3 q4 ) 1 2(q12 + q32 ) 2(q2 q3 + q1 q4 ) 7
6
5 (13)
2(q3 q1 + q2 q4 ) 2(q3 q2 q1 q4 ) 1 2(q12 + q22 )
where (q1 , q2 , q3 , q4 ) is the quaternion associated with the coordinate transformation: B A.
Consider two successive rotations to A00 from A represented by
C(q0 ) : A0 A (14a)
C(q00 ) : A00 A0 (14b)
where q0 is the quaternion associated with the coordinate transformation: A0 A and q00 is
the quaternion associated with the coordinate transformation: A00 A0 . These successive
rotations are also represented by a single rotation to A00 directly from A, as follows:
where q is the quaternion associated with the coordinate transformation: A00 A, and we
have
C(q) = C(q00 )C(q0 ) (16)
The resulting quaternion transformation relationship can be written as
2 3 2 32 3
q1 q400 q300 q200 q100 q10
6 q2 7 6 q300 q400 q100 q200 76 q20 7
6 7
6 7 =6
6
76
76
7
7 (17)
4 q3 5 4 q200 q100 q400 q300 54 q30 5
q4 q100 q200 q300 q400 q40
7
which is known as the quaternion multiplication rule in matrix form. The 4 ⇥ 4 orthonormal
matrix in Eq. (17) is called the quaternion matrix. Equation (17) can also be written as
2 3 2 32 3
q1 q40 q30 q20 q10 q100
6 q2 7 6 q30 q40 q10 q20 76 q200 7
6 7
6 7 =6
6
76
76
7
7 (18)
4 q3 5 4 q20 q10 q40 q30 54 q300 5
q4 q10 q20 q30 q40 q400
The 4 ⇥ 4 matrix in Eq. (18) is also orthonormal and is called the quaternion transmuted
matrix.
Consider kinematics in which the relative orientation between two reference frames is time
dependent. The time-dependent relationship between two reference frames is described by
the so-called kinematic di↵erential equations.
Consider two reference frames A and B, which are moving relative to each other. The
angular velocity vector of a reference frame B with respect to a reference frame A is denoted
~ B/A , and it is expressed in terms of basis vectors of B as follows:
~ ⌘!
by !
2 3
h i !1
6 7
~ = !1~b1 + !2~b2 + !3~b3 = ~b1 ~b2 ~b3
! 4 !2 5 (19)
!3
Ċ + ⌦C = 0 (20)
where 2 3
0 !3 !2
6
⌦ ⌘ 4 !3 0 !1 7
5 (21)
!2 !1 0
Similar to the kinematic di↵erential equation for the direction cosine matrix C, the
orientation of a reference frame B relative to a reference frame A can also be described by
introducing the time dependence of Euler angles.
Consider the rotational sequence of C1 (✓1 ) C2 (✓2 )
C3 (✓3 ) to B from A. The time
derivatives of Euler angles, called Euler rates, are denoted by ✓˙3 , ✓˙2 , and ✓˙1 . These successive
8
rotations result in
2 3 2 3 2 3 2 3
!1 ✓˙1 0 0
6 7 6 7 6 7 6
4 !2 5 = 4 0 5 + C1 (✓1 ) 4 ✓˙2 5 + C1 (✓1 )C2 (✓2 ) 4 0 75
!3 0 0 ✓˙3
2 32 3
1 0 sin ✓2 ✓˙1
6
= 4 0 cos ✓1 sin ✓1 cos ✓2 7 6
5 4 ✓˙2 5
7
(22)
0 sin ✓1 cos ✓1 cos ✓2 ✓˙3
Note that the 3 ⇥ 3 matrix in Eq. (22) is not an orthogonal matrix because ~b1 , ~a002 , and ~a03
do not constitute a set of orthogonal unit vectors. The inverse relationship can be found by
inverting the 3 ⇥ 3 nonorthogonal matrix in Eq. (22), as follows:
2 3 2 32 3
✓˙1 cos ✓2 sin ✓1 sin ✓2 cos ✓1 sin ✓2 !1
6 ˙ 7 1 6
4 ✓2 5 = 4 0 cos ✓1 cos ✓2 sin ✓1 cos ✓2 7 6 7
5 4 !2 5 (23)
cos ✓
✓˙3 2
0 sin ✓1 cos ✓1 !3
which is the kinematic di↵erential equation for the sequence of C1 (✓1 ) C2 (✓2 ) C3 (✓3 ).
If !1 , !2 , and !3 are known as functions of time, then the orientation of B relative to A as
a function of time can be determined by solving Eq. (23). Numerical integration of Eq. (23),
however, involves the computation of trigonometric functions of the angles. Also note that
Eq. (23) becomes singular when ✓2 = ⇡/2. Such a mathematical singularity problem for a
certain orientation angle can be avoided by selecting a di↵erent set of Euler angles, but it is
an inherent property of all di↵erent sets of Euler angles.
The kinematic di↵erential equation for the quaternion are given by
2 3 2 32 3
q̇1 q4 q3 q2 q1 !1
6 q̇2 7 16 q3 q4 q1 q2 76 !2 7
6 7 6 76 7
6 7 = 6 76 7 (24)
4 q̇3 5 24 q2 q1 q4 q3 54 !3 5
q̇4 q1 q2 q3 q4 0
9
we can rewrite the kinematic di↵erential equation (25) as follows:
1
q̄˙ = (q4 ! ! ⇥ q̄) (26a)
2
1 T
q˙4 = ! q̄ (26b)
2
where 2 32 3
0 !3 !2 q1
! ⇥ q̄ ⌘ 6
4 !3 0 !1 7 6 7
5 4 q2 5
!2 !1 0 q3
In strapdown inertial reference systems of aerospace vehicles, the body rates, !1 , !2 , and
!3 are measured by rate gyros which are “strapped down” to the vehicles. The kinematic
di↵erential equation (25) is then integrated numerically using an on-board flight computer
to determine the orientation of the vehicles in terms of the quaternion. Inertial sensors
such as star trackers or sun sensors are employed to correct state propagation errors caused
by angular-rate measurement uncertainties (e.g., gyro drift and bias), as was illustrated in
Fig. 9.1 for the ISC attitude determination system [4].
The quaternion has no inherent geometrical singularity as do Euler angles. Moreover,
the quaternion is well suited for on-board real-time computation because only products and
no trigonometric relations exist in the quaternion kinematic di↵erential equations. Thus,
spacecraft orientation is now commonly described in terms of the quaternion.
where {r1 , ..., rn } are a set of known reference unit vectors (e.g., the direction of the Earth,
the sun, a star, or the geomagnetic field) in the inertial frame, {b1 , ..., bn } are a set of the
corresponding measured (observed) unit vectors in the spacecraft body-fixed frame, and n
is the total number of measurements. Because these vectors are inaccurate, the weighting
coefficients ai are to be properly chosen to find the least-squares estimate of C. This problem
was first posed by Wahba in 1965 [11] and is often referred to as Wahba’s problem in the
literature.
10
Triad Algorithm
Consider an attitude determination problem using only two vector measurements. The
problem here is to determine the direction cosine matrix C for the two observation unit
vectors, b1 and b2 , corresponding to the two nonparallel reference unit vectors, r1 and r2 ,
as follows:
b1 Cr1 ; b2 Cr2 (27)
Assuming that {r1 , b1 } are more accurate than {r2 , b2 }, one can define two sets of new basis
vectors as
r1 ⇥ r2
x1 = r1 ; x2 = ; x3 = x1 ⇥ x2 (28)
|r1 ⇥ r2 |
b1 ⇥ b2
y1 = b1 ; y2 = ; y3 = y1 ⇥ y2 (29)
|b1 ⇥ b2 |
These are two triads of orthonormal unit vectors. There exists a unique orthogonal matrix
C which satisfies
yi = Cxi ; i = 1, 2, 3 (30)
The solution of Eq. (30) is then obtained as
3
X h ih iT
C= yi xTi = y1 y2 y3 x1 x2 x3 (31)
i=1
which is the so-called Triad algorithm for a simple, but nonoptimal, estimation of C using
two vector observations [1, 12]. Note that Cr1 becomes exactly b1 and that Cr2 doesn’t
become exactly b2 . The necessary and sufficient condition for b2 ⌘ Cr2 is
A major drawback of the Triad algorithm is its ad hoc nature. The two measurements are
heuristically combined to obtain an attitude estimate but the combination is not optimal
in any statistical sense. The error covariance matrix associated with an estimated C is
often computed in terms of Euler angles. An efficient method for computing the covariance
matrix of the Triad algorithm is presented in [12]. Although the Triad algorithm has been
implemented in numerous space missions [12], it cannot be easily applied to star trackers
that provide many simultaneous vector measurements.
11
The q-Method and QUEST Algorithm
where
n
X
B= ai bi rTi (36)
i=1
we transform the problem of minimizing L to the problem of maximizing G.
Because the nine elements of C are subject to six constraints, it is better to parameterize
C in terms of the quaternion. The direction cosine matrix given by Eq. (13) can be written
as
C = (q42 q̄T q̄)I + 2q̄q̄T 2q4 Q (37)
where 2 3 2 3
q1 0 q3 q2
q̄ = 6 7 6
4 q2 5 ; Q = 4 q3 0 q1 7
5
q3 q2 q1 0
where " #
S I z
K=
zT
12
and
n
X
= tr[B] = ai bTi ri (39)
i=1
n
X
S = B + BT = ai (bi rTi + ri bTi ) (40)
i=1
n
X h iT
z= ai (bi ⇥ ri ) = B23 B32 B31 B13 B12 B21 (41)
i=1
The problem of determining C is transformed to finding the quaternion vector q that max-
imizes the gain function G expressed by Eq. (38) subject to
Adding this constraint to the gain function with a Lagrange multiplier gives
G = qT Kq (qT q 1) (43)
By di↵erentiating this new gain function with respect to q, we obtain the necessary condition
as
Kq = q (44)
Thus, the eigenvector of K becomes the optimal estimation of q. The eigenvector associated
with the largest eigenvalue maximizes the gain function G because
G = qT Kq = qT q = qT q = (45)
we have
max =1 L (47)
Thus, a good approximation of the optimal eigenvalue is
max ⇡1 (48)
13
Once the optimal eigenvalue max is found using a Newton-Raphson iteration starting from
1 as the initial estimate, the next step of the QUEST algorithm is to solve the following
eigenvector problem
Kq = max q (49)
where q is the optimal quaternion. For the QUEST algorithm, this eigenvector equation is
rewritten as
p = [( max + )I S] 1 z (50)
where p is the Gibbs vector or Rodriguez parameters defined as
q̄
p= = e tan(✓/2) (51)
q4
Instead of inverting the matrix in Eq. (50), one may utilize Gaussian elimination to solve
the following equation
[( max + )I S]p = z (52)
After finding the optimal p, the optimal quaternion is then simply obtained as
" #
1 p
q= q (53)
1 + pT p 1
A simple solution to the time-varying attitude estimation problem was proposed in [14]. It
is based on propagating and updating the matrix B, as follows:
nj
X
B(tj ) = µ 3⇥3 (tj , tj 1 )B(tj 1 ) + ai bi rTi (54)
i=1
14
where 3⇥3 (tj , tj 1 ) is the state transition matrix of the attitude rotation matrix C, µ < 1 is a
fading memory factor to be properly chosen, and nj is the total number of vector observations
at time tj . The optimal attitude estimate at tj is then computed by the QUEST algorithm
for B(tj ).
The REQUEST algorithm proposed in [15] propagates and updates Davenport’s matrix
K as nj
X
T
K(tj ) = µ 4⇥4 (tj , tj 1 )K(tj 1 ) (tj , tj 1 ) + ai Ki (55)
i=1
where 4⇥4 (tj , tj 1 ) is the quaternion state transition matrix and Ki is Davenport’s matrix
K for a single observation.
A major disadvantage of such QUEST-based recursive methods is the use of a simple
fading memory scalar approximation for the sensor and process noises. The performance
of the QUEST-based methods have been significantly improved by an extended QUEST
algorithm [16]. The Extended QUEST algorithm solves the attitude estimation problem by
finding the attitude quaternion qj and the auxiliary state vector xj , which minimize the loss
function
j n
1X 2 1 1
J(qj , xj ) = i |bi C(qj )ri |2 + |Rww(j 1) wj 1 |
2
+ |Rqq(j 1) (qj 1 q̂j 1 )|2
2 i=1 2 2
1
+ |Rxq(j 1) (qj 1 q̂j 1 ) + Rxx(j 1) (xj 1 x̂j 1 )|2 (56)
2
subject to the quaternion propagation equation
xj = fx (tj , tj 1 ; qj 1 , xj 1 , wj 1 ) (58)
and the quaternion normalization constraint: |qj |2 = qTj qj = 1. The a posteriori estimates
of q and x at time tj 1 are denoted as q̂j 1 and x̂j 1 , respectively, the process noise vector is
denoted by wj 1 , and the standard deviations associated with bi measurements are denoted
by i.
The Extended QUEST algorithm employs two separate computational phases, as follows
[16].
(i) Dynamic Propagation Phase:
15
x̃j = fx (tj , tj 1 ; q̂j 1 , x̂j 1 , 0) (60)
The loss function after the propagation phase becomes
nj
1X 2 1
J(qj , xj ) = i |bi C(qj )ri |2 + |R̃qq,j (qj q̃j 1 )|2
2 i=1 2
1
+ |R̃xq,j (qj q̃j ) + R̃xx,j (xj x̃j )|2 (61)
2
where R̃ matrices are computed via a QR factorization.
(ii) Measurement Update Phase: The optimal xj is given by
1
xj = x̃j R̃xx,j R̃xq,j (qj q̃j ) (62)
The best estimate q̂j is then obtained by minimizing this modified loss function, Eq. (63).
The best estimate x̂j is then obtained as
1
x̂j = x̃j R̃xx,j R̃xq,j (q̂j q̃j ) (64)
where x is the state vector and w is the process noise vector. It is assumed that the process
noise is a Gaussian white noise whose mean and covariance are characterized as
E[w(t)] = 0 (66)
16
The initial mean of the state vector and the initial covariance of the state estimation error
vector are given by
E[x(t0 )] = x̂(t0 ) = x̂0 (68)
T
E{[x(t0 ) x̂0 ][x(t0 ) x̂0 ] } = P(t0 ) = P0 (69)
The estimated state vector satisfies the following equation
where (t, t0 ) is the state transition matrix with the following properties
@
(t, t0 ) = F(t) (t, t0 ) (77)
@t
(t0 , t0 ) = I (78)
The covariance matrix P(t) satisfies the Riccati equation
The estimated state vector and the state estimation error covariance matrix are then
propagated as
x̂j = (tj , x̂+
j 1 , tj 1 ) (81)
17
Pj = (tj , tj 1 )P+
j 1 (tj , tj 1 )T + Nj 1 (82)
where Z tj
Nj 1 = (tj , ⌧ )G(⌧ )Q(⌧ )GT (⌧ ) T
(tj , ⌧ )d⌧ (83)
tj 1
yj = h(xj ) + vj (84)
with
E[vj ] = 0 (85)
E[vj vjT ] = Rj (86)
and its measurement sensitivity matrix is obtained as
@h(x)
Hj = (87)
@x x̂j
x̂+
j = x̂j + Kj [yj h(x̂j )] (88)
P+
j = [I Kj Hj ]Pj (90)
The detailed applications of the EKF to the spacecraft attitude estimation problem with
the state vector consisting of the attitude quaternion, the gyro bias vector, and other uncer-
tain parameters can be found in the literature [17-20 ].
The EKF is widely employed for the state estimation of nonlinear dynamical systems. How-
ever, the unscented Kalman Filtering (UKF) is known to be performing better than the EKF
because the UKF reduces the linearization errors of the EKF. The UKF algorithm [18-21] is
briefly discussed as follows.
18
Consider a discrete-time nonlinear system described by
where xj is the state vector, yj is the measurement vector, wj is the process noise vector, and
vj is the measurement noise vector. It is assumed that wj and vj are zero mean uncorrelated
Gaussian noise processes with covariance matrices Qj and Rj , respectively.
The UKF is initialized as
x̂+
0 = E[x0 ] (92a)
P+
0 = E[(x0 x̂+
0 )(x0 x̂+ T
0) ] (92b)
The next step is to obtain a set of sigma points using the current best estimate of the mean
and covariance as
x̂ij 1 = x̂+ i
j 1 + x̃j 1 (93a)
q T
x̃ij 1 = nP+
j 1 ; i = 1, ..., n (93b)
i
q T
x̃jn+i1 = nP+
j 1 ; i = 1, ..., n (93c)
i
Using the propagated sigma point vectors x̂ij , we obtain a priori state estimate x̂j and error
covariance Pj as
where ai are weighting coefficients. Sigma points are re-computed using the current best
estimate of the mean and covariance, as follows:
19
The predicted observation vector ŷj and the covariance matrices are computed as
ment vector yj as
x̂+
j = x̂j + Kj (yj ŷj ) (97a)
1
Kj = Pxy(j) Py(j) (97b)
P+
j = Pj Kj Py(j) KTj (97c)
References
[1] Wertz, J. R. (ed.), Spacecraft Attitude Determination and Control, Kluwer Academic
Publishers, The Netherlands, 1978.
[2] Sidi, M. J., Spacecraft Dynamics and Control: A Practical Engineering Approach,
Cambridge University Press, Cambridge, 1997.
[3] Wie, B., Space Vehicle Dynamics and Control, American Institute of Aeronautics and
Astronautics, Second Edition, 2008.
[4] Brady, T., Tillier, C., Brown, R., Jimenez, A., and Kourepenis, A., “The Inertial
Stellar Compass: A New Direction in Spacecraft Attitude Determination,” SSC02-II-1, 16th
Annual USU Conference on Small Satellites, 2002.
[5] ADCS/GNC reference to be added
[6] ADCS/GNC reference to be added
[7] ADCS/GNC reference to be added
[8] Riedel, J. E., Bhaskaran, S., Desai, S., Han, D., Kennedy, B., Null, G. W., “Au-
tonomous Optical Navigation (AutoNav) DS1 Technology Validation Report,” Jet Propul-
sion Laboratory, California Institute of Technology, 26 April, 2004.
20
[9] Bank, T., Frazier, W., Blume, W., Kubitschek, D., “Deep Impact: 19 Gigajoules
Can Make Quite an Impression,” 24th AAS Guidance and Control Conference, AAS 01-022,
January 2001
[10] Uo, M., Shirikawa, K., Hasimoto, T., “Hayabusas Touching-Down to Itokawa- Au-
tonomous Guidance & Navigation,” 16th AAS/AIAA Space Flight Mechanics Conference,
AAS 06-214, January 2006
[11] Wahba, G., “A Least-Squares Estimate of Satellite Attitude,” SIAM Review, Vol. 7,
No. 3, July 1965, p. 409.
[12] Shuster, M. D. and Oh, S. D., “Three-Axis Attitude Determination from Vector
Observations,” Journal of Guidance, Control, and Dynamics, Vol. 4, No. 1, 1981, pp. 70-77.
[13] Markley, F. L., “Attitude Determination Using Vector Observations and the Singular
Value Decomposition,” Journal of the Astronautical Sciences, Vol. 36. No. 3, 1988, pp. 245-
258.
[14] Shuster, M. D. , “A Simple Kalman Filter and Smoother for Spacecraft Attitude,”
Journal of the Astronautical Sciences, Vol. 37. No. 1, 1989, pp. 89-106.
[15] Bar-Itzhack, I. Y., “REQUEST: A Recursive QUEST Algorithm for Sequential At-
titude Determination,” Journal of Guidance, Control, and Dynamics, Vol. 19, No. 5, 1996,
pp. 1034-1038.
[16] Psiaki, M. L., “Attitude-Determination Filtering via Extended Quaternion Estima-
tion,” Journal of Guidance, Control, and Dynamics, Vol. 23, No. 2, 2000, pp. 206-214.
[17] Le↵erts, E. G., Markley, F. L., and Shuster, M. D., “Kalman Filtering for Spacecraft
Attitude Estimation,” Journal of Guidance, Control, and Dynamics, Vol. 5, No. 5, 1982,
pp. 417-429.
[18] Crassidis, J. L. and Markley, F. L., “Unscented Filtering for Spacecraft Attitude
Estimation,” Journal of Guidance, Control, and Dynamics, Vol. 26, No. 4, 2003, pp. 536-
542.
[19] Crassidis, J. L. and Junkins, J. L. Optimal Estimation of Dynamic Systems, Chapman
& Hall/CRC, Boca Raton, FL, 2004.
[20] Crassidis, J. L., Markley, F. L., and Cheng, Y., “Survey of Nonlinear Attitude
Estimation Methods,” Journal of Guidance, Control, and Dynamics, Vol. 30, No. 1, 2007,
pp. 12-28.
[21] Simon, D., Optimal State Estimation, John Wiley & Sons, 2006.
21
9.3 Attitude Determination and Control Systems (ADCS)
9.3.1 Introduction
The Attitude Determination and Control System (ADCS) stabilizes, controls and positions a
satellite in a desired orientation despite any external or internal disturbances acting on it. The
satellite’s payload requires a specific pointing direction whether the payload is a camera,
science instrument or an antenna. Satellites also require orientation for thermal control or to
acquire the sun for their solar panels. The ADCS uses sensors in order to determine a
satellite’s attitude and actuators to control the vehicle to a required direction. The ADCS
needs to achieve the various mission and payload objectives such as pointing accuracy,
stability, rotational rate (slew) and sensing with many physical constraints such as mass,
power, volume, computer power/storage, space environment, robustness/lifetime and cost.
The ADCS is a synthesis of two subsystems: the Attitude Determination System (ADS) and
the Attitude Control System (ACS) of a satellite as depicted in Figure 1.
1
9.3.2 ADCS Requirements and Stabilization Methods
Satellites come with very diverse attitude determination and control requirements depending
on their mission, orbit and payload. For example, earth observation satellites require very
high levels of pointing and stability in order to ensure that images are not blurry or in order to
transmit onboard data to specific ground stations. Actuators such as reaction wheels, thrusters
or electro-magnets (magnetorquers) which can be used to manoeuvre a spacecraft (change its
attitude) can cause disturbance torques themselves which requires careful selection and
design of both actuators and sensors. Table 1 lists the typical ADCS requirements for a
satellite mission.
2
Gravity Gradient Torques: It exploits Newton’s law of general gravitation and can always
keep a spacecraft nadir pointing. This is achieved by using a boom extending a small distinct
mass (usually a magnetometer in order to minimize magnetic interference) from the
spacecraft (which becomes the second distinct mass) by a distance of 3-6m. These two
masses which are connected by a thin and light boom can then be used to exploit the
difference in gravitational pull on the main satellite platform and the additional mass
(magnetometer) due to the difference in their distance from Earth. This small difference can
be sufficient to enable the satellite/additional mass system to be aligned with the radius vector
at all times as an orbiting pendulum. The gravity gradient stabilization scheme can be
beneficial for coarse pointing (~5 deg) around the nadir axis while still the other two axes
will need to be stabilized. Early UoSAT satellites used this passive scheme for low cost space
missions in the 1980’s for store and forward communications as nadir pointing was only
needed for these missions [6].
3
(a) (b) (c)
Figure 3: Stabilization Schemes (a) METEOSAT Spinner (b) Boeing-Hughes Dual Spinner (c) SSTL
Nigeriasat-2 3-axis
Spinners: Spinning a satellite generates an angular momentum vector which remains nearly
fixed in inertial space. The angular momentum generated provides gyroscopic stiffness to a
spinning satellite, making it less prone to external disturbances and more stable for
thruster/apogee motor firings. This passive stabilisation technique was popular in the 1970’s
with GEO communication satellites and is still used for interplanetary missions due to its
simplicity and systems benefits for thermal and communication purposes. Detailed
dynamical formulations for spinners can be found in References [1-5].
Dual Spin: As a variation of the spin stabilisation scheme a dual spin satellite has two parts
of its structure spinning at different angular rates about the same axis. In this case, one
section of the satellite spins to provide angular momentum, while the other part (platform) is
despun and stays pointing in one fixed direction for example towards the Earth. Such a
scheme can be beneficial for a spacecraft in which the structure (diameter) of the platform is
required to be ‘thin’ to fit in a launch vehicle fairing. The disadvantage of this scheme is the
added complexity for carrying bearings and slip rings between the rotating parts of the
satellite. Dual spin satellites where also popular in the 1970’s and used in GEO satellites
where the satellite antennae could stay fixed towards Earth.
Bias Momentum: As a 3-axis system, momentum bias systems use one momentum wheel
aligned about the pitch axis normal to the orbit plane. Gyroscopic stiffness is used in order to
control the vehicle by keeping the momentum wheel spinning at a constant rate. Small
4
variations in wheel speed allow the control of the pitch axis. Roll coupling for nadir pointing
bias momentum systems can be used to control the other two axes.
Zero momentum: Most spacecraft today use the 3-axis zero momentum stabilization scheme
as they can provide higher accuracy. In these systems, reaction wheels are used for each axis
in order to compensate for external disturbances and in order to complete various
commanded manoeuvres. A pointing error is used to make the reaction wheels accelerate
from an initial zero value and then the wheels move to a small spin rate which keeps
increasing due to the manoeuvres required and due to secular disturbances, making them
reach their saturation limits. The increase in angular momentum to saturation levels requires a
desaturation strategy which is called ‘momentum dumping’ or unloading. This is achieved by
using magnetorquers or thrusters thus enabling the wheels to go to zero values.
5
used to estimate the atmospheric disturbance as the atmosphere’s density increases with
lower altitudes.
A suite of sensors is required to determine the attitude of a spacecraft including rates and
angular position despite constraints such as eclipses. The attitude information needs to be
provided continuously with sufficient accuracy. There are two categories of sensors: (i)
reference sensors which provide a reference or a ‘datum’ of a direction of an object such as
the sun, a planet or a star, though this could be interrupted by an eclipse. (ii) inertial sensors,
which can provide continuous attitude readings. Due to the various sensor concepts and
constraints, a suite of sensors which combines reference and inertial sensors is used for a
spacecraft using a balance of performance, mass/power consumption and cost.
6
/absence of Sun to sophisticated technologies which pin point the Sun’s direction to the
accuracy of few hundredth of a degree. These sensors can be classified into three basic
categories: the Sun presence detector, analogue sensor and the digital sensor. The following
sections briefly explain the operation of the sensors including the hardware involved for
better understanding.
As the name indicates the Sun presence detector outputs a signal when the Sun vector is in its
field of observation. The configuration of the sensors slits and the field of view varies for the
particular application for which the sensor is being used. Normally these sensors are used in
cases when a particular component onboard is sensitive to Sun and has to be switched ON /
OFF relative to the Sun’s presence / absence. For example equipment such as the space
telescopes and star trackers, needs to be protected from direct sunlight. Various
configurations of Sun presence sensors have been developed are given in Reference [1].
Here the photocell lies beneath the slits and generate an output signal when the Sun vector
lies on the plane of the slits. Two of such slit detectors are normally used in spin stabilized
platforms to detect the spin axis of the spacecrafts.
A particular configuration of the Sun sensor called the V-slit sensor has been widely used to
find the spin axis orientation of spin stabilized spacecraft. A V-slit Sun sensor has two slits,
the meridian slit parallel to the spin axis and the skew slit inclined at an angle i to the spin
axis. During each rotation of the spacecraft, the Sun vector crosses the plane of the meridian
slit once and the skew slit once.
The basic principle of operation for an analogue sun sensor is that the total energy flux on the
surface of the photo cell will be proportional to the cosine of the incidence angle of the Sun
vector. Since the current generated from the photo cell follows the cosine law, it is also
known as the cosine law detector. Using one photo cell a two dimensional representation of
the Sun vector can be computed, thereby making it a one axis Sun sensor. A two axis Sun
sensor is a combination of two single axis Sun sensors giving a complete 3 dimensional
7
representation for the Sun vector. Figure 4 shows a Sun sensor developed by SSTL with a
field of view of ±50° with an accuracy of 1° weighing 0.3 kg. Figure 10b shows a Sun sensor
developed by TNO and Bradford Engineering with a field of view of 128° x 128° with an
accuracy of <1° weighing 365g.
The digital Sun sensors provide precise measurements, therefore providing higher accuracy in
comparison to analogue sun sensors. In order to obtain a complete description of Sun vector
orientation two single axis digital sun sensors with their optical plane at 90° apart is required.
The digital Sun sensor comprises of an optical head and a signal processing unit. The optical
head has a narrow slit for the Sun rays to pass through and fall on the reticle slits organized to
represent any of the codes such as gray code or binary code. A more detailed description of
the choice and usage of gray binary code is given by References [1, 3]. The encoded output
from the reticle slits is decoded in order to obtain the Sun vector orientation using the signal
processing unit.
With the evolution of MEMS technology (Micro Electro Mechanical System) miniature
sensors using Active Pixel Sensors (APS) have been developed recently. Photo cell detectors
used in these devices can be based on CMOS and CCD technologies. A CMOS sensor
module is comparatively smaller with less power consumption than a CCD image sensor,
although this benefit is at the expense of the image quality. For microsatellites and
nanosatellites where the size and mass plays a vital role, the CMOS sensors are useful and
can also be cheaper than other solutions.
8
Digital Sun Sensor Examples
Figure 5a shows the sensor from Adcole co-operation can be used in spin stabilized
platforms. The field of view is ±87.5° with an accuracy of ±0.25°and weighs 0.25 kg. Figure
5b shows a sensor from Adcole co-operation can be used in for three axis stabilization mode.
The field of view is 128° x 128° with an accuracy of ±0.25° [9]. Figure 5c shows a sensor
from Officine Galileo can be used in for three axis stabilization mode. The field of view is
128° x 128° with an accuracy less than 0.02° weighing less than 400grams.
a) Miniature Spinning Sun Sensor b) Two Axis Digital Sun Sensor c) Two Axis Digital Sun Sensor
The Earth provides a reference direction for determining the relative attitude of a spacecraft.
Unlike the Sun, Earth cannot be approximated as a point source target, as for a spacecraft in
low Earth orbit, around 40% of the satellites’ vision is filled up by Earth. It is sufficiently
easy to ‘see’ the Earth from the spacecraft through a wide range of spectral bands. But in
order for the sensor to differentiate Earth from cold space, all along the orbit, the radiance
emitted should have a uniform energy distribution over a range of the spectrum. Earth’s
albedo lies mostly in the visual region and it varies widely depending on the reflecting
surface (ice, snow, forest, water, soil, etc.) and also on the time of the day, therefore causing
ambiguity in the measurements while using visible spectral range. A better spectral region
that is useful and that which matches the requirements is the infra red region. The spectral
range of 14 to 16μm (CO2 band) is used by most horizon sensor as in this region the energy
emitted has a uniform energy distribution irrespective the day/night terminator and the
Earth’s reflecting surface (as most of the radiation are from atmosphere above the Earth’s
surface) [10-12].
9
Principle of Operation of Earth Sensor
The Earth sensor normally operates by scanning the sky to detect the Earth’s horizon. The
Earth sensor has an optical system, detector along with signal processing unit. If the Earth
sensor is on a spin stabilized platform, as the spacecraft spins the sensor can scan the sky to
detect for infrared radiation emitted by the Earth. Otherwise the sensor is provided with the
steerable scanning mechanism to scan the sky. Figure 6 shows the basic components.
Figure 8a shows the STD 15 EADS Sodern sensor, used for GEO orbits to measure pitch and
roll angles. It is a dual conical scanning sensor, working at an altitude of 15000 to 140000
km, with an operating nominal depointing pitch range of ±12 deg (roll = 0) and roll range of
±2.9 deg (pitch = 0). With accuracy budget 3σ, bias amounts to 0.035 deg and noise typical
noise to 0.015 deg [11]. Figure 8b shows the STD 16 EADS Sodern sensor, used for LEO
orbits to measure pitch and roll angles. It is a dual conical scanning sensor, working at an
10
altitude of 300 to 6000 km, with an operating nominal depointing pitch range of ±17 deg (roll
= 0) and roll range of ±33 deg (pitch = 0). With accuracy budget 3σ, bias amounts to 0.06
deg and noise typical noise to 0.042 deg [11].
a) STD15 b) STD 16
Star Sensors
Star cameras can provide accurate absolute attitude information by imaging stars and
matching them to catalogue positions. They provide the most accurate attitude information of
all satellite sensors – an estimation accuracy of 20 arc seconds and less is typical. Star camera
performance depends on the ability to detect dim light sources, and the attitude accuracy
improves with the number of stars that can be detected. CCD devices are usually used for
imaging (as opposed to CMOS imaging devices) because they are more sensitive to the
incoming photons. Star trackers have large apertures to allow as much light as possible to
enter the lens, and also make use of baffles to suppress stray light from the sun or light
reflected from the Earth and moon. A star camera can be used in eclipse and daylight portions
of the orbit as long as the boresight points away from these bright objects. The imaging
devices used in star camera are susceptible to radiation effects. Radiation particles will
typically damage pixels on the sensor, reducing its star detection capabilities over time.
11
Figure 8: Star Camera Schematic
A star tracker will match observed stars to catalogue stars. Each match yields a vector pair:
an observed vector in the camera coordinate frame (which can be converted to the spacecraft
body coordinate frame) and a vector in the inertial coordinate frame (obtained from the
catalogue itself). As long as two or more of these vector pairs are available, a full attitude
estimate can be obtained. The TRIAD algorithm uses only 2 vector pairs but QUEST
(quaternion estimation) is an optimal algorithm which determines the attitude that achieves
the best weighted overlap of the reference and observation vectors [16].
Reference star positions are obtained from a catalogue, such as the Hipparchos and Bright
Star Catalogues (BSC). Detected stars are matched to catalogue stars by finding pairs of stars
with the same angular separation and the same apparent magnitude. A matching tolerance is
used to allow for detection variations. Multiple pairs of catalogue stars may be matched with
a detected pair. These matches are then further pruned using a constellation matching
algorithm [17].
SSTL Rigel Star Tracker and Draper Lab Inertial Stellar Compass (ISC)
In this section two Star trackers are presented, the SSTL Rigel Star tracker and the Draper
ISC. The Draper ISC is not a pure star tracker – it has built in gyroscopes. It integrates
12
gyroscope readings to find the attitude and only uses star vector measurements periodically to
remove the gyroscope errors and to calibrate the gyroscopes. The SSTL Rigel star camera is a
new development that combines the heritage of the optics of SODERN star cameras for the
Camera Head Unit (CHU) with the low cost electronics (DPU-Data Processing Unit)
developed by SSTL. The Rigel Star camera can be used in a configuration with multiple
CHU for increased robustness and coverage. The two systems have been built as low
cost/mass sensors for multiple space missions [18-20].
Figure 9: Draper ISC, SSTL Rigel Star Camera with single and multiple CHU
Magnetometers
Magnetometers have become one of the most commonly used attitude determination sensors
for satellites in LEO. This is primarily due to their simplicity, robustness, low cost and small
mass. They are used to measure the strength and direction of the local magnetic field. When
this information is combined with a model of the earth’s magnetic field such as the IGRF
model, the attitude of the satellite can be determined. Due to the fact that the field is not well
mapped and has many anomalies, magnetometers can only provide coarse measurements and
are usually combined with other sensors such as star cameras and sun sensors.
Magnetometers are extensively used in the detumbling phase of the satellite when
magnetorquers are used for that purpose. The firing of the magnetorquers needs to be timed
so as to allow the magnetic field to breakdown before taking any readings. Special
consideration needs to be taken as to the placement of the magnetometer. It must be placed
away from any sources which might cause noise. For this reason they are often placed at the
ends of extensible booms. The most common type of magnetometers for attitude
determination purposes is the fluxgate magnetometer. These magnetometers usually have a
sensitivity of ±10nT with a range of ±60μT. They tend to have an accuracy of between 0.5-5
deg and are only usable for altitudes of below 6000km [13].
13
Rate Gyros
Spinning gyros are one of the oldest and popular sensors used on-board satellites and
aerospace vehicles. Gyros measure angular rates of a vehicle without the need of any
knowledge of an external or absolute reference. That means that if the attitude of a spacecraft
is determined with an earth or sun sensor, then the angular rates of the satellites principal
axes can be obtained by differentiating the angular position outputs of the sensors. However,
this will be problematic if the spacecraft enters in an eclipse where it will no longer be
possible without the use of rate gyro sensors to measure the satellite’s attitude and therefore
be able to control it. Another reason for using rate gyro sensors is the need to control the
angular rate of a spacecraft in addition to the angular position. Differentiating angular
position outputs of a satellite from other sensors in order to get the angular rates, can lead to
noisy results which will affect the stability and pointing of the ADCS system.
The gyro consists of a spinning wheel which reacts and measures imposed attitude rotations
of a vehicle. The most common type of a gyro is the rate-gyro (RG) and the rate integrating
gyro (RIG) which are based on the gyroscopic stiffness of revolving moments of inertia. The
biggest disadvantage of a gyro is its reliance on moving parts-gyro-which has a limited
lifetime. However, advances in mechanics, microelectronics and space components have
contributed to the development of new rate sensors-gyros, based on new physical concepts
with no moving parts such as the laser gyros, quartz rate sensors, MEMS sensors, Fiber Optic
Gyros (FOG) and Hemispherical Resonator Gyros (HRG).
In order to be able to understand and compare rate gyros which come with inherit noise
problems, we need to define various design and performance parameters:
• Range: The larger the range of measurement, the larger the noise level of the sensor.
The smaller the range, the better the accuracy we can get, ranges of 1 deg/s-100 deg/s
are feasible
• Bias (drift): The most important characteristic of a gyro inherit to the technology of a
gyro, ranging from 0.01 deg/hr to 1 deg/hr
• Output noise: specified per frequency band
• Scale factor: Important for rate integration and has a strong influence on the
achievable attitude accuracy
Gyros are commonly used in clusters, one per axis plus a fourth unit in a skewed
configuration for redundancy. This particular configuration is also called an Inertial
14
Reference Unit (IRU) and the combination of gyros and accelerometers can give us in
addition position/velocity measurements and are called Inertial Measurement Units (IMU).
GPS
GPS or GNSS signals can be used for orbit determination (position) but also for attitude
determination using multiple antennae layouts. A set of antennae is used connected to a GPS
receiver on the top panel of a spacecraft in LEO facing the GEO/Galileo navigation
constellations placed in MEO orbit. Using the phase difference between the antennae for each
of them allows the reconstruction of the attitude of a spacecraft. Despite various technical
issues such as multi-path and noise, accuracies of 0.1-1 deg have been achieved [14, 15].
1. Momentum Wheels (MW): They provide constant angular momentum for gyroscopic
stabilisation. Orientation of the spin axis is fixed with respect to inertial space.
Attitude Control is achieved by varying the spin speed of the wheel about some
nominal value.
2. Reaction Wheels (RW): They provide torque to a vehicle by increasing or decreasing
the speed of the wheel, with the wheel nominally at rest.
3. Control Moment Gyroscopes (CMG): A momentum wheel gimballed in one or two
axes. Control torques are generated by changing the direction of the momentum
vector, by changing the direction of the spinning wheel’s axis.
15
4. Magnetic torquers (MT): Magnetic coils or electromagnets that generate magnetic
dipole moments, M. A magnetic torquer produces torque proportional (and
perpendicular) to Earth’s magnetic field, B. It is often used as a second actuator on
spacecraft to desaturate momentum exchange systems.
5. Thrusters: Produce a thrust (force) or torque around the centre of mass by expelling
mass.
Momentum/Reaction Wheels
h = hB + hw = 0 [1]
Therefore, hB = - hW [2]
where h is the total momentum of the satellite, hB = IBωB is the angular momentum of the
rigid body (satellite) with an inertia IB and angular rate of ωB and hw = IWωW is the angular
momentum of the MW/RW with a flywheel inertia IW rotating at a speed ωW. The ratio
between the satellite and flywheel inertia is selected such that it fulfils the attitude control
(agility) requirements of a specific mission, keeping into consideration mass, volume and
power constraints for actuators to be used on a satellite. Using the attitude dynamics
equations derived from Euler’s momentum equation [1-3]:
[3]
where N is the torque. From Eq. 3 it is possible to derive the various mathematical
descriptions of attitude control schemes employing various actuators such as MW/RW and
CMGs.
16
A cut-out view of a typical wheel is shown in Figure 11. It consists of a precision engineered
fly wheel with most of the mass concentrated at the tip/rim of the disc to achieve maximum
wheel inertia for aluminium mass. A brushless DC motor is usually used to rotate the wheel.
The complete wheel assembly with integrated electronics is housed in a cage. This cage has
a twofold task; it helps protect the spacecraft in case something goes wrong with the spinning
wheel but also sometimes acts as a pressure vessel to keep the lubricant from outgassing.
The bearing assembly which is required for mechanical support and operations is what limits
the lifetime of reaction wheels to about 5-15 years, depending on the duty cycle of the wheel,
type of lubrication used and motor technology. In the last decade, new mechanical and
mechatronic developments have enabled the design of miniature, low cost and low-jitter
MW/RW with longer lifetimes. Reaction wheels are used when accurate time optimal rapid
manoeuvres are required. They allow continuous and smooth control of torque. They can
accelerate in both directions and normally have a zero speed. Due to sticking friction though
they display a non linear response at very low spin rates (rpm) which might cause an irregular
motion of the space craft. This is usually solved by running the wheel with a small bias.
Momentum wheels are essentially the same as reaction wheels but have a large nominal spin
rate. This provides a constant angular momentum which causes gyroscopic stiffens around
two axes which help in maintaining the attitude of the spacecraft.
Figure 11: SSTL Microwheel 100SP-M with CFRP casing, internal electronics and specification [22]
A minimum of three orthogonal wheels are required for full three axis control. In order to
avoid single point failure, usually a fourth wheel is added in a skewed configuration. Due to
17
the addition of a fourth wheel which is usually added at an equal angle with the other three
wheels, additional torque and momentum authority may be required. The wheels need to be
properly sized so as to not saturate by the expected worst case disturbance torques. When
momentum builds up, external torque actuators such as magnetorquers/thrusters need to be
used to dump some of the momentum. The torque capability of the reaction wheels is
determined by the desired slew rates.
CMGs are gimballed wheels which can generate large amounts of torque/angular momentum.
They are considered to be ‘torque amplifiers’ as they can use the stored angular momentum
in a flywheel and ‘convert’ it to large torques by gimballing the flywheel appropriately.
A CMG consists of two parts:
• The momentum wheel which produces a large and constant angular momentum
(magnitude)
• The gimbal motor (or set of gimbal motors), on which the momentum wheel is
mounted, so that the angular momentum vector of the wheel can be changed to the
desired direction.
Torquing the gimbal results in a precessional torque that is normal to the gimbal axis and
momentum wheel spin axis.
NCMG
18
.
where, h is the angular momentum vector and δ is the gimbal rate
Most CMG’s are used on large spacecraft and space stations, mainly due to their high angular
momentum storage capability, which is used to provide increased stabilisation under large
external disturbance torques. They can also produce substantial torques and are very heavy
(typically 55-150 kg for 300-1000 Nms momenta and 100-1000 Nm output torques [2]).
Recently, Astrium and Honeywell have begun to work on smaller CMG’s (Table 4), for new
families of spacecraft in the 500-2000 kg range that require high precision pointing and fast
slew capabilities [2, 24-26].
Figure 13: Ball Aerospace WorldView 1 and 2 Spacecraft with Honeywell M50 CMG
Astrium is also building a CMG based on a Teldix RW. The compact CMG has been
designed for the French Pleiades spacecraft, a new platform in the 1 ton class designed for
agile, high resolution imaging in a constellation of imaging (provided by France) and radar
(provided by Italy) spacecraft. Pleiades 1 and 2 are currently being prepared for launch in
2011 and mid-2012 [26].
19
Figure 14: Astrium Pleiades-1 Satellite and Astrium’s 15-45S CMG
Magnetorquers/Magnetic Control
Magnetic control has been used in many spacecraft missions. The simplicity, inexpensive
hardware and reasonably good attitude control (0.5o to 5o in all axes) makes magnetic control
very attractive to use, especially for small satellites primarily for attitude control and
momentum dumping of reaction/momentum wheels.
NM = M × B [5]
The direction of M can be controlled on average by a proper sequence of magnetorquers
firings, but the B field vector is dependent on the orbital location. As a result, the torque NM
which always is orthogonal to B and M is not necessarily favourable for control of the
attitude of a specific spacecraft axis, in certain regions of the orbit. Another drawback of
magnetorquers is that it is possible that a desirable control torque for a certain attitude axis
(pitch, roll, yaw), might generate undesirable disturbance torques for the other axes. The
Earth’s magnetic is predominately a magnetic dipole. The magnetic field can be expressed
mathematically by a spherical harmonic model, the so-called IGRF (International
Geomagnetic Reference Field) model. For purposes of simulation a first order dipole model is
utilised in order to represent the geomagnetic field vector. This dipole vector is expressed as:
⎡ R TMe ⎤ Me
B = ∇⎢ 3
[
⎥ = 3 1 − 3RR
T
] [6]
⎢⎣ Rs ⎥⎦ Rs
where,
∇ is the vector gradient operator
20
Rs is the length of the geocentric position vector
R is the unit geocentric position vector
Me is the geomagnetic strength of the dipole vector
1 is the identity matrix
Magnetorquers can produce a torque based on the magnetic moment it can produce which
depends on the number of coil windings n, the cross-sectional area A of the coil, and on the
amount of current I that passes through the coil in the unit vector along the coil’s axis u:
M = nIAu [7]
NM = nIA(u x B) [8]
Usually, 3 magnetorquers are used on spacecraft, one per each axis for coarse attitude control
and mainly for angular momentum unloading. Their utility decreases the higher the altitude a
spacecraft is placed on due to the decreasing strength of the magnetic field. The field’s
strength and direction also varies. A specific feature of magnetorquers is that they cannot
produce a torque component about the local field direction. For example in a polar orbit any
required direction can always be achieved at some point in the orbit since the field direction
changes round the orbit. In the equatorial plane this would be problematic due to the field
lines always being in a north-south direction. Magnetorquers do not require any propellant,
require very limited power levels and have an unlimited lifetime, as well as no moving parts.
Therefore they are very popular, simple to manufacture and low cost actuators. Figure 15
shows SSTL’s MTR-5 magnetorquers with over 50 units used for small satellite spacecraft,
with a 6.2 Am2 moment, mass of 500g and dimensions of 251 x 30 66 mm.
21
Thrusters
External disturbances acting on satellite can be countered by using small thrusters, thus
controlling the total momentum of the spacecraft. They are mounted in clusters on the
surfaces of a satellite in various configurations in order to provide the required direction of
thrust about each axis. The disadvantages of using a thruster, especially when compared to
magnetorquers is the use of consumables (propellant), increased mass, complexity and cost.
However, using thrusters is independent of altitude and of the Earth’s magnetic field and the
thrusters can be used for fine/precise attitude control, station-keeping of GEO satellites and as
an orbit control system in many cases. Europe’s ATV (Automated Transfer Vehicle) built by
EADS Astrium for ESA is an automated robotic tug which is used as a ferry to transport
cargo to the ISS using an autonomous ranging and rendezvous system which requires very
fine and precise attitude control. The ATV which has a mass of 20,750 kg at launch and uses
28 220N bipropellant thrusters for attitude control operations with two missions (Verne and
Kepler) achieved to date. Figure 16 shows the thruster layout for the ATV which uses 4
clusters of 2 thruster and 4 clusters of 5 thrusters [34].
Figure 16: Astrium/ESA ATV with 5-200N Cluster and 200N thruster Firing Test [34]
22
9.3.5 ADCS Design Case Studies
UoSAT-12 was launched in April 1999 in a 650 km circular, 64.5° inclination orbit. UoSAT-
12 supports a wide range of sensors for attitude determination and a multi-channel GPS
receiver for onboard orbit determination and accurate time synchronisation. The GPS receiver
also has an experimental attitude determination capability, through baseline sensing of an
array of 5 patch antennas. A redundant set of three 3-axis flux-gate magnetometers are used
to measure the geomagnetic field vector in the satellite’s body co-ordinates. These
measurements are used to determine the magnetic coil torque vector and in combination with
a magnetic field model to estimated the full attitude and angular rates of the satellite. Four 2-
axis (azimuth and elevation) sun sensors measure the sun vector angle to high accuracy.
During the nominal nadir pointing attitude mode, small pitch and roll angles can be measured
with a 2-axis infrared horizon sensor. The highest attitude measurement accuracy is obtained
from a dual set of opposite looking star sensors. The sensors supply star measurement vectors
and matched star catalogue vectors at a rate of once per second to an attitude and rate
estimation filter. A solid state angular rate sensor is mounted in one axis of the satellite to
flight qualify the sensor for future mission. Table 1 lists all the sensors used on UoSAT-12
for attitude determination and their most important characteristics.
Twelve magnetorquer coils are positioned within the satellite to give some level of
redundancy and to deliver full 3-axis magnetic dipole moment control. These coils are
controlled by using dual polarity current pulse width control to deliver the required averaged
level of magnetic moment per sample period. The magnetorquers are used for:
23
• Detumbling of body angular rates after launch
• Momentum dumping of the reaction wheels
• Momentum maintenance on the momentum wheel during Y-spin stabilisation
• Nutation damping during spin stabilisation
• Libration damping and yaw spin/phase control after deployment of a backup gravity
gradient boom
Table 2 lists the various actuators used on UoSAT-12 for attitude and orbit control and their
respective characteristics.
24
Table 2: Attitude and orbit determination sensors on UoSAT-12
UoSAT-12 was launched on the 21st of April 1999 from the Baikonur Cosmodrome in
Khazakstan. The initial telemetry of the magnetometer indicated a tumbling rate of about 2
deg/s. The next day the ADCS software was loaded on the 186-OBC and the Rate Kalman
filter confirmed the initial tumbling rate to be mainly around the Y-axis. This result
confirmed the slightly higher Y-axis MOI and the cross-products of inertia.
25
Figure 19: Roll and pitch pointing during zero-bias 3-wheel control
After 2 orbits the body angular momentum of UoSAT-12 was almost completely dumped.
The magnetorquer controller was left running for 2 more orbits until the satellite was in the
required Y-Thomson attitude. The cross-products of inertia prevented the satellite to reach
the target rate exactly and the true Y-rate estimated was approximately -0.8 deg/s with small
residual X and Z-rates of less than 0.2 deg/s.Following transition to nadir pointing the ADCS
system was put in zero momentum mode. The pitch and roll pointing errors experienced are
very small, the 1-s deviation is 0.13deg and the maximum peaks during this period less than
0.5 deg. The yaw error is worse (1-s deviation is 0.62 deg) due to a lack of accurate yaw
information close to the polar region. The reason being the use of the magnetometer as the
only source of yaw information to the EKF estimator.
The EUROSTAR family of GEO platforms has been a very successful platform for GEO
platforms built by EADS Astrium. This section presents the architecture of the EUROSTAR
3000 AOCS system flown on the CNES STENTOR, Amamzonas-2, Arabsat 5A/C, Interlsat-
10, Hotbird 10, YahSat 1A/1B missions and many others.
26
Figure 20: Astrium EUROSTAR 3000 Intelsat-10
The EUROSTAR 300O AOCS was designed by the then MATRA MARCONI Space (now
Astrium) company based on the successful EUROSTAR 2000+ AOCS concept and hardware
with a focus on performance enhancement and reducing operational workload. The AOCS
system was designed to also be scalable for medium to very large GEO platforms. The AOCS
system was validated in the CNES STENTOR mission flown in 2002. Heritage for the
EUROSTAR 3000 AOCS system was also drawn from the EUROSTAR 1000/2000 AOCS
including Solar Sailing which was used for smooth roll/yaw attitude control through the
modulation of Solar Radiation Pressure (SRP) on the solar arrays equipped with fixed
patented flaps. With the use of electric propulsion (EP) the AOCS system achieved improved
pointing stability. With the use of electric propulsion (EP) the AOCS system achieved
improved pointing stability. Yaw sensing becomes important in EP station-keeping
manoeuvres, and this is achieved using solar array sun sensors in a gyroless control mode.
THE EUROSTAR 3000 AOCS subsystem uses a centralised computer concept. The AOCS
system uses a 4 wheel skewed configuration composed of 2 MW and 2 RW. The Chemical
27
Propulsion Subsystem (CPS) with a strong heritage uses 14 10N thrusters in two branches
and a Liquid Apogee Engine (LAE). The Plasma Propulsion Subsystem (PPS) is composed of
two small platforms with plasma thrusters used for North/South station keeping manoeuvres
and orbit eccentricity correction. The thruster direction is controlled by commanding stepper
motors of the 2-axis Thruster Orientation Mechanisms (TOM). The solar arrays equipped
with flaps produce long term inertial torques through the offset of each wing with respect to
the Sun direction.
The Transfer and Acquisition phase which is inherited from the EUROSTAR 2000+
sequence is based on three axis control. For the EUROSTAR 3000 an improved design is
28
utilised allowing for simpler operations without any loss on safety. In the transfer phase, all
operations from launcher separation to the end of the last LAE firing sequence are included in
which solar arrays are partially deployed. In this phase only sun and earth acquisition is used
to reach three axis pointing. The pointing information is then used for gyro stabilisation in
order to perform the LAE firing. When on-station the attitude is then based on a new 2-DOF
wheel system which deviates from the momentum bias system used on EURPSTAR 2000+.
Here the solar sail mode is used exclusively for long term wheel off-loading. For large GEO
satellites ii is not easy to separate the short term/long term movements due to nutation
frequency. Thus a new robust momentum/control design is implemented for global large
band control extending the bandwidth of the control. Another innovation of the EUROSTAR
3000 AOCS system is the ability to provide gyroless yaw estimation during the EP station-
keeping phase based on Earth and sun sensor measurements.
References
[1] Wertz, J. R. (ed.), Spacecraft Attitude Determination and Control, Kluwer Academic
Publishers, The Netherlands, 1978.
[2] Wie, B., Space Vehicle Dynamics and Control, American Institute of Aeronautics and
Astronautics, Second Edition, 2008.
[3] Sidi, M. J., Spacecraft Dynamics and Control: A Practical Engineering Approach,
[4] Wertz, J. R., Larson, Space Mission Analysis & Design , Microcosm, Torrance,
California, 1999
[5] Berlin, P., Satellite Platform Design, Department of Space Science, University of Lulea,
5th Edition, 2007
[6] M.S. Hodgart "Gravity Gradient and Magnetorquing Attitude Control for Low Cost Low
Earth Orbit Satellites- the UoSAT Experience", Ph.D. Submission at University of Surrey,
June 1989
29
[10] EADS Sodern Horizon Sensor,
http://www.sodern.com/site/FO/scripts/siteFO_contenu.php?mode=&noeu_id=61&lang=EN,
[Accessed 29 March 2011].
[13] Steyn, WH, A Multi-mode Attitude Determination and Control System for Small
Satellites, PhD Thesis, Stellenbosch, 1995
[14] Lightsey, G.E., “Spacecraft Attitude Control Using GPS Carrier Phase”, Global
Positioning System: Theory and Application, Parkinson, B.W. and Spilker, J.J. (ed), Vol. II,
1996.
[15] Lightsey, G.E., Cohen, C.E., Feess, W.A. and Parkinson, B.W., “Analysis of Spacecraft
Attitude Measurement Using Onboard GPS”, Advances in the Astronautical Sciences, Vol.
86, 1994.
[16] Shuster M.D., Oh S.D., "Three-Axis Attitude Determination from Vector Observations",
Journal of Guidance and Control, Vol.4, No.1, pp.70-77, Jan.1981.
[17] Van Bezooijen, R.W.H, A Star Pattern Recognition Algorithm for Autonomous Attitude
Determination, IFAC Automatic Control in Aerospace, Japan, 1989
[20] T. Brady, "Technology Validation of the Inertial Stellar Compass (ISC)," Proceedings of
the Sixth Annual NASA Earth Science Technology Conference (ESTC 2006), College Park,
MD, USA, June 27-29, 2006, URL:
http://www.esto.nasa.gov/conferences/ESTC2006/presentations/C6p3.pdf
30
[23] SST: Smallwheel 200SP Data sheet,
www.sstl.co.uk/Downloads/Datasheets/SmallWheel, Accessed 23 March 2011
[24] Lappas, Vaios J (2002) A Control Moment Gyro (CMG) Based Attitude Control System
(ACS) For Agile Small Satellites Doctoral thesis, University of Surrey.
[26] Defendini, A., Lagadec, K., Guay, P., Blais, T., Griseri, G., “Low cost CMG-based
AOCS designs”, Proc. 4th International Conf. on Spacecraft Guidance, Navigation and
Control Systems, pages 393-398, 2000
[28] Burt, Richard “AAS 03-072 “Failure analysis of International Space Station Control
Moment Gyro” 26th Annual AAS Guidance and Control Conference, Breckenridge,
Colorado.
[30] http://www51.honeywell.com/aero/common/documents/myaerospacecatalog-
documents/M50_Control_Moment_Gyroscope.pdf, Accessed 23 March 2011
[33] A. Bradford, “BILSAT-1: A Low Cost, Agile, Earth Observation Microsatellite for
Turkey” International Astronautical Federation, October, Houston, USA, 2002
[35] Steyn, W.H., Hashida Y., “An Attitude Control System for a Low-Cost Earth
Observation Satellite with Orbit Maintenance Capability”, Proceedings of the 13th Annual
AIAA/USU Conference on Small Satellites, Utah State University, Logan, Utah, August
1999.
31
[36] Martin, J.; Benoit, A.; D'Allest, C.; Chaudron, F., Eurostar 3000 AOCS Design,
Spacecraft Guidance, Navigation and Control Systems, Proceedings of the 4th ESA
International Conference, held 18-21 October, 1999 in ESTEC, Noordwijk, the Netherlands.
Edited by B. Schürmann. ESA SP-425. Paris: European Space Agency, 2000, p.53
32
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
operations and to survive critical situations without relying on ground segment intervention ([6]).
The on-board autonomy depends on the specific mission requirements and constraints. The
autonomy level can vary between a very low level of autonomy, involving a high level of control
from ground, to a high level of autonomy, whereby most of the functions are performed on-board.
The autonomy of the space segment has an impact on total life cycle cost in as much as increased
autonomy can increase development costs, but decreases operations and maintenance costs.
Therefore, the adoption of specific autonomy goals for a given mission is decided by careful
balancing of costs, risks, and schedules for both the development and the operation & maintenance
phases. The needs for autonomy are very different from one mission to another and there are three
different factors that have a strong influence in the degree of autonomy required for a mission:
- Communication delays, when the mission characteristic time is much shorter than ground
control response time (including communication signal round-trip time).
- Environment uncertainty, related again to the safety of the mission that demands high
reactivity to unknown disturbances, for instance in missions to NEOs.
- Costs and operation teams downsizing, very important in long routine phases.
In summary, autonomous GNC systems are complex and critical for space programs (in terms of cost,
risk, and schedule). Therefore, nowadays autonomous GNC systems are implemented when are the
only feasible option or in technology demonstration missions. Some examples where mission
feasibility depends on autonomous GNC are robotic rendezvous in Mars (e.g. Mars Sample Return),
pinpoint entry, descent and landing (EDL) in a solar system body or hypervelocity impact or fly-by of
small bodies (e.g. Deep Impact). There have been advancements to reduce the operational costs of
aerobraking in Mars by means of increased level of GNC autonomy (e.g. Mars Reconnaissance
Orbiter [7]) which takes several months and have many uncertainties and risks.
Some examples of technology demonstration missions implementing autonomous GNC are Deep
Space 1, in which the autonomous flight of long interplanetary low thrust arcs reduces operational
costs, or PRISMA [17] for rendezvous and formation flying.
9.4.2.1 Filter
The navigation function is typically implemented as a digital filter in the on-board computer. The
navigation filter shall provide the necessary parameters to the guidance and control (e.g. current
vehicle state). If just one sensor can provide all the required information, then the navigation
algorithm can be just a low-pass filter. In general, the filter shall
- Provide states not directly observed by the sensors and additional dynamical parameters
(e.g. current thrust level delivered by the propulsion system), and/or
- Process measurements from different sensors, and/or
- Provide the vehicle state during intervals without measurements.
To achieve such objectives there are different types of filters, usually classified as,
- Dynamic or kinematic filters regarding the integration of dynamical equations or only the
Kinematics. In general, kinematic filters require lower number of operations but are far less
flexible and robust than dynamic filters.
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
where yk is the filter output at time tk, xk is the input, and ρk is the a-priori residual (difference
between the input and the predicted output). This filter requires only one configuration parameter,
the LOS-rate gain G2.
Examples of batch filters are the weighted least squares filter and the Square-Root Information Filter
(SRIF) ([9]). The formulation of SRIF can be summarized as
S k− Z k− S k+ Z k+
T =
h y k 0 ρk
k
(S k− ) T S −k = Pk− (2)
Z k− = S k− X −k
where Xk is the parameter matrix at epoch tk, Pk the covariance matrix, yk the observation vector, T()
represents the Householder reflection, hk is the observation matrix, and ρk the vector of
measurements residuals. This filter can be used in sequential mode if the batch only includes the
newest measurement. A priori information can be included as an additional measurement.
The most popular filter for on-board navigation is the Kalman filter ([10]). It is a dynamic filter that is
usually used in sequential estimation, although batch filtering is possible. There are several
formulations of the Kalman filter (some of them described in section 9.2),
- Linear Kalman filter or just Kalman filter that consider linear time invariant dynamical model
(analytical transition matrix)
- Extended Kalman filter that considers non-linear propagation of average state
- Kalman UD as the extended Kalman but with UD decomposition of the covariance matrix to
assure the conservation of the positive definiteness in the time and measurement updates
- Unscented Kalman filter in which the statistics are propagated and constructed from a set of
wisely selected points (sigma points).
The steps of the Kalman filter are schematically shown in Fig. 1. The time update propagates the
average state and its covariance matrix from the last epoch (can be the initialization epoch and
state) to the current measurement time (a priori state and covariance). Then, a test of hypotheses is
done on the input measurements. If the measurements are accepted, the measurement update
provides the a posteriori state and covariance. The measurement update might be iterated to
smooth the non-linearities. The a posteriori residuals and covariance are checked to check the
hypotheses again and the convergence. The output is prepared in the proper format and some
information for aiding the measurement management (e.g. for image processing).
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
INITIALIZATION MEASUREMENT
NAV. FDIR
MONITOR
[
K j = P j− H Tj R j + H j P j− H Tj ]
−1
where xj is the augmented state vector at time tj, Pj is the state covariance matrix, δyj is the
difference between real and estimated measurement, Hj is the observation matrix (partial
derivatives of measurements wrt state), and Rj is the covariance matrix of the zero-mean, time-
uncorrelated, additive measurement noise. The configuration parameters are the initial covariance,
the model (or process) noise (for the time update of the covariance matrix), and the measurement
uncertainty (usually a dynamic input).
Different filters might be used in different phases of a mission. More usually, different configurations
of a filter are required in different phases. The filter design must take into account
- The unmodeled physical effects and their impact in the propagation and observations,
- The uncertainty in the parameters of the considered dynamics and measurement models,
- The inclusion of multiple sensors in the navigation chain, and
- The allocated computer resources.
The augmented state vector refers to the uncertain parameters considered in the filter (can be
solved-for or consider parameters). The selection of the augmented state vector components shall
consider the above mentioned issues. It typically includes the SC state (in proper coordinates), and
uncertain parameters from the dynamics and measurements models.
The dynamics equation for each uncertain parameter depends on the parameter itself and the
application. The most usual models for uncertain parameters from dynamics and measurements are
biases (constant average value), drift (linear time dependency) or colored noises (e.g. exponentially
correlated random variables [9]). Additive white Gaussian noise is usually included as process and
measurement noise. These noises come from unknown or unmodeled sources and contribute to
increase the a priori covariance and to decrease the innovation vector (damp the effect the
measurements update). A high value of such noises to avoid divergence of the filter is sometimes
called artificial noise.
frequency, and their quality. Other system performances (e.g. pointing error and stability) affect the
performances of the on-board sensors.
Often a single sensor cannot provide all the required observables during all the operational ranges of
a mission. Thus, a sensor suite is mounted. In each phase, different sensors are used, or the same
sensor with different processing algorithms (e.g. camera providing LOS at far distances, or providing
LOS and range at close distances). The filter processes different measurement types to estimate the
uncertain parameters (data or sensor fusion). An example is the hybridation of inertial navigation
(IMU) with other sensor such as GPS or camera or altimeter. For the selection of the sensor suite,
different issues shall be taken in consideration. The following considerations shall be traded in order
to optimize the sensor set. For instance, for far-range operations the nominal sensor might have a
large FOV with coarse accuracy. However for close operations, the nominal sensor is switch to one
with narrower FOV and higher accuracy.
- What observables are needed for navigation.
- Which sensors provide the observables (note that some sensors can provide several
observables simultaneously).
- What accuracy is required (one order of magnitude more accurate than overall GNC
requirement, but if not feasible, at least two or three times).
- What is the operational range (distance, FOV, velocity, angular rate …)
- What are the system-level implications.
o System constraints as power, size, mass, cost.
o Operational constraints as illumination, on-board vs. ground-based processing.
There are several sensors that might provide the same observable in a given mission phase. The
combination of the different observables shall provide the information required by the navigation
filter (the system shall be observable). Note that not all the estimated parameters need to be
directly observed. For instance, the velocity can be estimated by a dynamic filter after a certain time
using only position-related observables (LOS, range). A list of different sensors that can provide the
most usual observables is given below. As technology evolves, new sensors appear providing more
observables and more accurate (for instance flash LIDAR or 3D TOF camera). Thus, the list below is
not exhaustive but includes the most used sensors for autonomous GNC applications.
- Range measurements are directly provided by radar altimeters (as in ESA’s Huygens), laser
range finders (as in NASA’s NEAR-Shoemaker), or RF sensors (as in CNES’s Formation Flying
Radio Frequency system on PRISMA mission). Range can be derived from camera image in
certain applications by means of proper image processing (as in ESA’s ATV).
- Range-rate can be provided by RF sensors measuring the Doppler shift (as the Russian Kurs
system on Soyuz and Progress spacecraft).
- Line-of-sight can be provided by LIDAR sensors (like the scanning LIDAR in ATV that provide
simultaneously range and LOS), RF sensors (the Kurs system) or by optical cameras (as in the
approach phase of JAXA’s Hayabusa mission).
- Horizontal velocity (normal to the LOS) can be measured by a Doppler radar (as in NASA’s
Mars Science Laboratory) or processing series of camera images (like DIMES in NASA’s MER).
- Complete position is provided by some sensors like space-qualified GPS (or GNSS) receivers
that provide PVT solution of the spacecraft in the vicinity of the Earth (as in JAXA’s HTV), or
scanning LIDAR that can provides complete relative position.
Vision-based navigation is often used in many autonomous GNC systems. The main reasons are the
low cost and system requirements (mass, size, power) of the sensor. Cameras can provide accurate
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
LOS and range measurements (thus full state), and even relative attitude. As an example, Fig. 2
shows different observables that can be obtained from images of the Moon. These observation
types were traded for the design of the backup optical navigation system for manned missions [13].
The main drawback is the image processing algorithms required to derive the observables. Often,
some aids are used in order to make simpler image processing algorithms (more details on section
9.4.5).
Fig. 2. Illustration of star and limb related observables and image matching (positioning) from Moon images (© GMV).
modulated thrusters) and/or the thrust duration is required to have full controllability. In addition,
during long thrusting arcs it might be necessary to allocate short ballistic arcs for navigation tasks.
If the maneuvers are applied in open-loop (no control function), the guidance dynamical model shall
be accurate enough to achieve the required goal. If the maneuvers are executed in closed-loop, the
control function can compensate small unmodeled effects in the guidance function.
Analytical algorithms providing the solution in closed-form are preferred for on-board
implementation. These solutions are not always available to achieve the guidance objective within
the allocated error budget. It is important to note that the reference trajectory generated by the
guidance algorithm shall fulfill the operational constraints. Thus, in many cases the guidance
problem is formulated as a constrained optimization problem.
One of the most useful guidance methods for impulsive maneuvers is based on the differential
guidance. It was originally introduced for interplanetary navigation ([12]) but can be applied to
rendezvous, formation flying, orbit maintenance or D&L on small bodies. The basic formulation is
δrN 0 δr0 0
≡ = Φ N,0 +
δv N 0 δv 0 + ∆V1 ∆V 2 (4)
∂x
Φ N,0 ≡ N
∂x 0
that considers an initial deviation from the reference trajectory (δr0 , δv 0 ) and two impulses,
- The initial delta-V (∆V1) cancels the final position deviation at a fixed final time δrN .
- The final delta-V (∆V2) cancels the final velocity deviation δv N .
The linear system of equations defined by Eq. (4) can be solved explicitly. The key issue is the
computation of the transition matrix Φ N,0 . For linear time-invariant systems the transition matrix
can be obtained analytically. In more complex dynamics, the transition matrix can be computed by
numerical differences or by integration the variational equations of motion. Note that the second
delta-V might never be applied since at the time of arrival to the final point a new maneuver can be
calculated to achieve the next guidance objective (similar to a receding horizon control). The
formulation can be extended to consider undefined final time. In this case, the final time is solved
minimizing the total delta-V.
The differential guidance can be applied at intermediate points of the trajectory to cancel
perturbations that might grow to unacceptable level at the final time (Fig. 3). These trajectory
corrective maneuvers (TCM) can be applied at any instant if the navigation filter has converged after
the previous maneuver.
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
When the effect of finite thrust on the trajectory is not negligible, impulsive guidance cannot be
applied. If the thrusting duration is not too long compared with the guidance horizon, then a simple
parameterization of the maneuver can be used. The small number of parameters permits the
optimization of the maneuvers with low computational cost. An extreme case was implemented in
Deep Space 1 ([1]). This mission used electric propulsion to fly by small bodies. Long low thrust arcs
were executed. The thrust profile in spherical coordinates was discretized as piece-wise linear
expansion. A parameter optimization problem was solved analytically. The gain matrix depends only
on the sensitivity matrix of the final state deviation to the maneuver parameters.
When the thrust arcs are comparable to the guidance horizon (e.g. electric propulsion) more refined
parameterizations are needed (for instance the guidance method in [15] was also applied to the D&L
on small bodies). In some problems the optimal control theory ([11]) can be applied. The parameters
of the optimal control (initial adjoints value) are solved to get the thrust profile that fulfills the
guidance objective. This option can increase significantly the guidance computation time because
the co-states (aka adjoints) are propagated simultaneously. There are exceptions when the adjoint
dynamics can be solved analytically (e.g. optimal lunar descent and landing).
In some landing missions, hazard avoidance is required. The hazard avoidance system is usually not
considered part of the GNC system. The measurements from the same sensors can be used by both
but the guidance will take as objective the site selected by the hazard avoidance system. In turn, the
hazard avoidance must consider the current navigation estimated state and the guidance capabilities
to compute the reachable site locus.
For typical thruster management algorithms, the translational control shall provide forces in
spacecraft body axes. The desired force can be achieved controlling the thrust level during a given
duty cycle (throttling), or the thrusting time with constant thrust magnitude (pulse-width in pulsed
mode or total thrusting duration in steady mode). The thruster management will select the optimal
thrusters to provide the force, and also the torque requested by the attitude control. The thruster
management will also define the thrusting times or thrust level of each thruster. Attitude constraints
are considered in the guidance and controller algorithms. It shall be noted that different thrusters
might be used in different mission phases, for instance thrusters providing hundreds of Newton for
large transfer maneuvers and Newton-level thrusters for proximity operations.
The control function monitors the imparted acceleration in order to fulfill the guidance and the
controller commands. The control of the maneuver execution can be done in closed-loop or in open-
loop. In open-loop the firing duration is computed from the expected acceleration. In closed-loop,
state update at high frequency is usually needed. The applied delta-V during one thruster control
cycle can be measured with accelerometers if they are sensitive enough. If the acceleration provided
by the thrusters is smaller than the resolution of the accelerometers, then other indirect
measurements might be used (e.g. pressure, voltage).
In case of simple thruster architecture, the maneuver execution control can consider thruster
saturation (maximum thrust) and minimum impulse bit (MIB), i.e. minimum thrusting time, in the
computation of the commanded delta-V. For finite thrust (usually low-thrust) a controller is usually
implemented. The maneuver execution control is less demanded since the closed-loop controller
updates frequently the thrust profile and can compensate execution errors.
In some missions, the control function includes a closed-loop controller proper. The controller shall
cancel deviations produced by maneuver execution errors, dynamical perturbations and state
uncertainty. Thus, a controller increases the accuracy and robustness of the GNC at the expense of
increased complexity and usually higher propellant expenditure. The presence of a translation
controller is required when there are tight final delivery requirements (e.g. terminal RDV or precise
landing).
The basic objective of the closed-loop controller for translational motion of a space vehicle is to
achieve the required performance and stability with low sampling rates. The performances of the
controller are expressed in terms of different metrics often competing with each other, such as low
steady-state error and transient response error, sufficiently fast response time, and low propellant
consumption. The controller is implemented in software on the on-board computer. Hence, a
discrete control is required. Discretization of continuous-time systems (e.g. bilinear or Tustin
transformation) is often valid but better performances are achieved designing the GNC system
directly in the discrete domain.
Different trajectory control algorithms can be applied to different mission phases. A common
approach in the controller design is the linearization around the reference trajectory provided by the
guidance. Some controllers often used for translation motion control are regulators and terminal
controllers ([11]).
A terminal controller aims to reach the desired conditions at a terminal time. An example is finite-
horizon optimal control (problem of model predictive control) that has been proposed for different
applications (pin-point landing on Mars ([14]), low-thrust interplanetary trajectory control ([15]), or
precise landing on asteroids ([16]). The formulation of the discrete finite-horizon optimal control
problem is
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
1 N −1
min J = ∑ δu K
2
, subject to
δ {u K } 2 K =0
x& K = A K x K + B( u K + δu K ) + Bg~K , ∀k = 0 ,..., N − 1
x = x (5)
N f
u + δu ≤ a
K K K max , ∀k = 0,..., N − 1 ,
−u +δu ( t −t ) / I g
where a K max = Tmax / m K , mK = m K −1e K −1 K −1 K K −1 SP 0
The dynamics has been linearized at reference nodes xk of the reference trajectory. The thrust
acceleration uk is discretized using the zero-order hold approach. The resulting dynamics is a
piecewise linear time-invariant (LTI) system. The other constraints are to achieve the desired final
state xf and the maximum thrust provided by the thruster (formulated as limited control
acceleration aKmax). The cost function J is the sum of squares of the control corrections δuk, in order
to minimize the deviations from the optimal thrust profile (quadratic objective function is
convenient to solve guidance and control algorithms).
These problems admit analytical transition matrix and closed-form solution for the linear
constraints. However, they need to be iterated to fulfill the non-linear constraints. Apart from the
maximum thrust, attitude constraints are also non-linear and needs to be considered ([1], [15]).
Nevertheless, the short computation time permits on-board implementation at a frequency high
enough to track the reference trajectory. A receding horizon implementation is convenient to avoid
singularities close to the terminal time.
A regulator aims to maintain a reference condition (e.g. a fixed position or an orbit). A state
feedback regulator compares the reference state (guidance) with the estimated state (navigation)
and generates an acceleration to cancel the error signal. A schematic diagram of a closed-loop
control system is depicted in Fig. 4 (feedforward and direct link are not depicted). The plant includes
the actuators, the real world dynamics, and the sensors. Every component of the system introduces
disturbances not always additive (e.g. scale factors, cross-couplings). The closed-loop system must
be stable in the presence of such disturbances. The disturbances from the sensors and actuators
might be specified in the GNC design. In some cases these components are fixed and the GNC
system must cope with their performances.
+ x u
GUIDANCE CONTROL PLANT
−
NAVIGATION
FILTER
where the state x is the error signal defined in Fig. 4, u is the control acceleration, and K is the
control gain matrix. Several methods exist to compute the control gain and fulfill the performance
and stability requirements. The on-board implementation only needs to select the gain
corresponding to the current mission phase and GNC mode.
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
A classical Proportional Integral Derivative (PID) controller is one of the most popular control
techniques for SISO systems. In the PID controller, the control acceleration is defined in Eq. (7). The
proportional term reacts to a current error and is related to the response time (how fast the
controller compensates a certain error). The integral term cancels the steady-state error of a pure
proportional control. Integral term wind-up often appears in GNC systems (thruster saturation). The
derivative term is related to the damp of future oscillations or overshoots, at the expense of slowing
the transient response. If the derivative line is present, it is convenient that navigation filter
estimates the derivative of the state minimizing the high frequency noise of derivation. The tuning of
the controller consists of setting the different gains to achieve the requirements. There are several
methods to tune the PID controller, for instance the Ziegler-Nichols method, but usually require
manual trial and error.
dx
u = KP x + KD + K I ∫ x(t )dt (7)
dt
The controller often includes notch filters to avoid the excitation of lightly damped flexible modes. In
addition, lag-compensation techniques are included when the delays introduced in the system
reduce too much the stability margins.
The main advantage of PID control is that it does not require knowledge of the plant model.
However, PID control is only applicable when the channels can be decoupled. When the cross
couplings prevent the use of SISO control, modern MIMO methods are used. The optimal space-
state methods require knowledge of the plant model in order to minimize a cost function. One
simple optimal control technique is the Linear-Quadratic Regulator (LQR). This regulator assumes a
linear (time-invariant) dynamics and a quadratic cost function J to compute the feedback control
gain K that minimizes J. The formulation for continuous control is
∞
J = ∫ (x T Qx + u T Ru)dt
0
(8)
dx
= Ax + Bu
dt
where the control u and state x are vectors, and the weight matrices Q and R are the controller
tuning parameters (the cross-coupling of control and state in the cost is usually not included). There
are many software packages to obtain the optimal control gain K. This gain is associated to a given
linearized dynamics (i.e. A, B). If there are different reference states, then a set of gains are pre-
computed and stored on-board.
When there are uncertainties on the plant parameters, robust control methods are more
convenient. These methods are mathematically cumbersome and require certain knowledge of the
plant. On the other hand, they can provide graceful degradation of performances in the presence of
bounded uncertainties (if the deviation of the uncertain parameter from its nominal value is too
large the system becomes unstable).
The non-linearities such as thruster saturation can introduce problems for stability and require
proper treatment. In addition, dead-bands might be introduced in order to reduce the propellant
consumption. Finally, it is important to highlight the coupling between the navigation and guidance
algorithms, and of course the sensor and actuator equipment. For instance, the navigation filter
estimates the parameters requested by the guidance and control, removing the high-frequency
noise from the sensors. Then, the guidance or control includes the estimated biases (zero-frequency
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
assures the achievement of the required final conditions for mating. The terminal RDV
(sometimes known as final approach) might end with a free drift to cover the last meters
(due to contamination issues or simply sensors availability).
During the design of the RDV trajectory, some ‘time-flexible elements’ [5] needs to be included in
order to synchronize the RDV timeline with external events (visibility or communication windows) or
schedules (crew or ground operations). The RDV timeline needs to be modified in-flight because of
the uncertainties and errors/deviations that appear in real-world operations. Typical time-flexible
elements are hold points (constant relative position) and free drift orbits.
The RDV problem has been traditionally analyzed in circular orbit. However, elliptic rendezvous is
nowadays often considered, if not for nominal operations (elliptic orbits might yield better mission
performances) at least for contingency scenarios (orbit injection error). For the most useful
maneuvers in circular RDV ([5]), equivalent maneuvers in elliptic RDV have been defined ([18]),
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
including time-flexible elements (hold points are now periodic orbits around the hold point). In
addition, the analytical transition matrix of the Hill-Clohessy-Wiltshire equations for circular orbits
has an analogous close-form solution for the elliptic RDV ([19]).
In order to estimate the relative position of the chaser wrt the target, a sensor suite needs to be
selected. Different sensors are available for relative navigation in different scenarios. A trade-off for
each particular mission is needed. Table 9-1 summarizes the current sensors available for RDV. Note
that the orbit of the target or the chaser needs to be known in order to formulate the equations of
the relative motion. The reference orbit can be known from the ground-based orbit determination
system or using satellite navigation systems like GPS (only in Earth orbits). The relative state can be
obtained from differences of absolute measurements (if available from both vehicles and known by
the chaser GNC) but lead to larger errors than direct observations of relative state.
Providing specific values for sensor selection is difficult because there are many parameters that are
mission dependent and shall be assessed specifically. For instance, the size of the target has a
significant impact on the operational range of some sensors (optical camera, RF and LIDAR on target
without reflectors). However, some guidelines for sensor assessment can be provided for a
rendezvous GNC system.
- RF-sensors are good for medium to long distances (mainly homing and closing) including
acquisition and contingency (omni-directional or scanning antennas).
- GPS is a reliable well known technology but requires cooperative target and is only available
operationally in Earth orbits. Relative Kinematic GPS (using phase) provides good
performances (several cm with multi-path) that makes it suitable for closing and terminal
RDV.
- Cameras with different FOV can cover the entire RDV phase. The main problem for target
acquisition and orbit synchronization is the visibility constraints (illumination conditions,
faint target). The visibility windows for a NAC and for RF-sensor (omni-directional antennas
on target) are shown for a Mars RDV scenario in Fig. 7.
- Scanning LIDAR is a mature, robust, precise sensor that can provide relative position and
attitude measurements at relatively high frequency (~10 Hz). Its main drawbacks are the
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
limited operational range (only for close RDV) and the high cost, mass, and power demand
compared to the rest of relative sensors.
Vision-based navigation is one of the most promising technologies for autonomous RVD, in particular
when the target is non-cooperative. In this context, a non-cooperative target is a spacecraft that
does not incorporate aids for relative navigation. The cameras are used often in conjunction with
other sensors to increase the robustness (for instance omni-directional RF-sensors for higher
localization probability in case of large uncertainty or LIDAR for terminal phase). For redundancy,
different optical heads and processing units are mounted. It is convenient that the MVM can
configure any combination of optical head and processing unit.
Different cameras might be necessary to cover the entire RDV phases. For instance, for optical far-
range a camera with narrow or moderate FOV (higher resolution) and slightly defocused
(straightforward obtaining of sub-pixel accuracy) is best fitted. During the far imaging, the object
appears as a point source, and only the line of sight (LOS) can be computed. The maximum range
depends on the camera sensitivity, usually defined by the camera limit magnitude. The image
integration time is a critical parameter for far-range target detection. Long exposure time permits
detection of fainter object. However, a short exposure time is preferable to relax attitude stability
requirements. Some techniques permit to trade these competing requirements.
In optical close-range operations, the object is extended and a well focused camera with larger FOV
is preferred. In this phase, the shape of the object is distinguishable with sufficient resolution for the
image processing to provide distance to the target and even relative attitude. Different image
processing techniques are available to provide the required observables.
Fig. 7. Visibility arcs with nominal NAC constraints (16º exclusion angle, 11 mag limit, eclipses) (©GMV)
Hill’s equations are a system of linear equations that approximate the relative motion between two
bodies in orbit. These equations have been largely used since the early space missions to compute
the dynamic of vehicles in a rendezvous scenario. Two hypotheses are assumed to derive the general
equations of motion, close distance between both spacecraft, and circular orbits. Hill’s equations are
expressed as,
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
&x& − 2ωz& = FX m
2
&y& + ω y = FY m (9)
2
&z& + 2ωx& − 3ω z = FZ m
where x, y, z are the chaser position with respect to the target in the target local orbital frame (Fig.
8), ω is the angular orbital rate of the target satellite, m is the mass of the chaser vehicle, and FX, FY,
FZ are the differential forces acting on the chaser.
This system can be represented as a linear time invariant system in state space given below. Note
that the out-of-plane motion (y) is decoupled from the in-plane motion (x,z). The range of validity of
the Hill’s equation is increased if the reference frame is formulated in curvilinear coordinates instead
of Cartesian coordinates.
0 0 0 1 0 0 x 0 0 0
0 0 0 0 1 0 y 0 0 0
FX
dx 0 0 0 0 0 1 z 0 0 0
= Ax + Bu = + FY (10)
dt 0 0 0 0 0 2ω x& 1 m 0 0
FZ
0 ω 2 0 0 0 0 y& 0 1 m 0
2
0 0 3ω − 2ω 0 0 z& 0 0 1 m
The homogeneous solution (zero-input) of the Hill’s equations (Eq. (9) or Eq. (10)) is known as the
Clohessy-Wiltshire (CW) equations (sometimes Eq. 10 are known as CW equations in the literature).
The CW equations in the state space result
r ( t F ) Φrr (t F , t0 ) Φ rv ( t F , t0 ) r (t0 )
= (11)
v ( t F ) Φ vr (t F , t0 ) Φ vv (t F , t0 ) v (t0 )
where,
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
These equations are very convenient for use in a linear Kalman filter since the transition matrix is
analytical. Hence, the time update (or propagation of state and covariance matrix) is very fast. In
addition, the analytical transition matrices are well suited for the differential guidance presented in
Eq. 4. This discrete control can be applied in case of impulsive transfers. Multiple rendezvous
strategies for the circular case are given in chapter 3 of [5]. Some of them are extended for the
elliptic rendezvous in [18]. For instance, in the case of V-bar hopping (Fig. 9), the nominal maneuvers
for each hop are given by,
ω
∆VZ 1 = ∆VZ 2 = ∆x (13)
4
where ∆x is the actual distance to be traversed. The duration of the hop maneuver is half an orbital
period T. At an intermediate point a correction maneuver can be applied using the transition matrix
presented in Eq. 12.
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
Fig. 9. Guidance and control of V-bar hopping (or radial impulse transfer along V-bar).
In the case of forced motion along V-bar (Fig. 10), the task of the guidance is to compute the
reference R-bar acceleration that maintains a constant approach velocity VX considering the actual
V-bar position from the navigation. The approach velocity VX is usually defined from safety
considerations. The continuous R-bar acceleration γZ is given below. The duration ∆t depends on the
approach velocity and the traversed distance ∆x. Continuous control algorithms could be applicable
in this continuous thrust maneuvers with the considerations for discrete implementation mentioned
in paragraph 9.4.4.
γ Z = 2ωV X
∆x (14)
∆t =
VX
The terminal phase poses different requirements and constraints for high-speed impact and
rendezvous missions. A fly-by of a small body presents many similarities to impact missions (for
instance NASA-JPL AutoNav system [23] was used for Deep Space 1 fly by comet Borrelly, Stardust fly
by asteroid Annefrank, and Deep Impact ([20]) impactor and flyby spacecraft). The usually
demanding delivery requirements, uncertain environment, and tight operational constraints make
the GNC system for the terminal phase a critical enabling technology of missions to small bodies. An
example of the tight requirements is an impact with a small body of few 100 m, the delivery
accuracy for high success probability shall be few or several 10 m.
Any mission to a small, irregular body, such as Itokawa the target NEO of Japanese mission Hayabusa
([21], [22]), will have to cope with uncertain environmental conditions (SC dynamical environment,
asteroid shape, rotational state and surface characteristics). In addition, the forces acting on the SC
are all small. Thus, the uncertainties in the dynamical model have a significant impact in trajectory
prediction and in accuracy-critical phases (e.g. descent and landing). Another important factor is the
communication delay, typically of tens of minutes, which is critical in hypervelocity impact or in
descent and landing. Thus, autonomous operations in certain critical phases are mandatory.
The most generic methods to obtain relative observations are, images with a camera (in the visual or
near-infrared spectral range), and/or range measurements with an altimeter (radar or laser). The
range output can be a Digital Elevation Map using a scanning or flash LIDAR. Typically, the camera
measurements can be taken at a further distance than the range measurements.
At the beginning of the terminal phase, the first objective of the GNC system is to detect and identify
the small body against the starry background. Small, irregular objects observed at long distances
(∼106 km) are point sources with low visibility (faint and highly variable as can be seen in Fig. 11).
The observability depends on the asteroid rotational state and the Sun-asteroid-SC relative
geometry. Long image integration times increases the signal-to-noise ratio. For center of brightness
(CoB) calculation, tight relative pointing error (RPE), aka pointing stability, is required to concentrate
the photons around a pixel. During long integration times with no tight RPE, point sources will
produce a characteristic pattern in the image. For this imaging strategy, Deep Space 1 implemented
an image processing technique called ‘multiple cross-correlation’ ([23]). Such IP algorithm correlates
each object in an image with a mask template extracted from the same image for center-finding.
For target detection and point-source tracking, a Narrow-Angle Camera (FOV ∼< 5°) is the best
option because provides higher sensitivity and LOS accuracy. A star-tracker (FOV ∼10°) can be a good
alternative if reduces the equipment and chain and the slightly defocused optics provides sub-pixel
accuracy (it is also possible with a NAC and over-sampling of image stacks).
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
Fig. 11. Integrated magnitude and offset between the center-of-brightness (CoB) and the center-of-mass (CoM) for a
simulated 240-m (433) Eros-like asteroid (©GMV).
Early maneuvers are more effective for correcting deviations, but navigation accuracy is usually
worse at larger distances. Hence, it is necessary to design in conjunction the navigation chain, the
guidance strategy and the control system to achieve the delivery requirements with optimal use of
the on-board resources.
In the case of an impact or fly-by mission the time-to-go (time to impact or minimum distance, or
equivalently relative distance and speed) is not controlled. Therefore, only divert maneuvers are
required to control the impact point. Several guidance and control strategies can be applied [24],
involving different propulsion systems.
- Predictive-impulsive guidance, where impulsive maneuvers are executed at predefined
times. This strategy is fitted for relatively high thrust (e.g. tens of N for vehicles of several
hundred kg).
- Proportional navigation, where continuous thrust is proportional to the LOS rate and
approach speed. This is a well-known missile guidance method. It is particularly fitted for low
thrust level (e.g. less than 1 N for vehicles of several 100 kg). Special care must be taken
about the minimum thrust level required to compensate the initial deviation of the impact
point.
- Hybrid scheme, which implements mid-course predictive–impulsive guidance and terminal
proportional-navigation. It is designed for missions with intermediate thrust level.
The navigation filter is closely related to the impact guidance and control strategy. For the
abovementioned strategies, two different types can be implemented:
- Estimation of LOS and its time derivatives. Some applicable filters are digital fading memory
or batch-sequential least squares. These filters are more sensitive to the IP performances.
- Estimate the complete relative state vector with a dynamic filter (e.g. a Kalman-Schmidt
filter). These filters are prone to over-estimate the innovation vector and to numerical
problems due to the non-observability of some components of the state vector.
The rendezvous missions present notable differences. During the approach phase most of the
relative velocity is cancelled by means of braking maneuvers. The duration of this phase is long
enough to involve the ground control in the navigation chain. This relaxes the requirements in the IP
although due to the communication limitations still some critical pre-processing operations might be
done on-board. Thus, full autonomous GNC is not mandatory (different levels of autonomy might be
used depending on the particular mission).
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
This approach can be split in two phases. During the far approach, the celestial object is a point
source in the image. It starts after successful detection and identification of the small body.
Asteroids of ∼1 km are typically resolved (illuminates several pixels) at a range of ∼1000 km. Laser
range finders used in past missions to small bodies achieved acquisition at 250 km (NEAR-
Shoemaker) and 50 km (Hayabusa). Photometry can be used to infer range from the brightness of
the object. However, the measurements are subject to very large error (up to 100% depending on
the knowledge of the asteroid lightcurve and properties).
Thus, the navigation function shall estimate the relative distance and speed without direct range
measurements. The more accurate options are,
- Long ballistic flight allowing long observation arcs (spanning a significant portion of the
object orbit). It takes long time and the relative trajectories must assure good observability
conditions.
- Execute ‘dog-leg’ maneuvers that rapidly change the observation geometry ([25]). Proper
design of the approach trajectory (and thus maneuvers) is mandatory.
It is important to note that due to the irregular shape of the asteroid there might be a risk of losing
the asteroid tracking early in the far approach (Fig. 12). The image processing and navigation filter
shall be robust against such event ([25]). In all the terminal phases of missions to small bodies, the
accuracy of the navigation filter depends on the knowledge of certain asteroid characteristics. The
sequence of operations must assure that there is enough knowledge of the body characteristics to
accomplish the next phase objectives.
Fig. 12. Simulated images of an Eros-shaped asteroid at the point of peak brightness (minimum integrated magnitude),
NEA axis of rotation perpendicular to the approach plane (©GMV).
In the close approach the asteroid appears as an extended object in the camera frame. In addition,
at a certain point range measurements might begin to be available from a laser altimeter (if
available). The IP algorithm is based on CoB computation (other more accurate and complex
techniques requires size and shape knowledge). Therefore, the CoB-CoM offset must be accounted
for in the navigation filter. The GNC algorithms can be applied in certain modes of the proximity
operations.
In the proximity operations the gravity of the asteroid becomes non negligible and at short distances
the most significant force. Still, the dynamics is slow and a fully autonomous GNC is not mandatory,
except in the descent and landing. The weak, irregular gravity field opens new possibilities for
‘orbiting’ the asteroid, for instance hovering (station keeping) or self-stabilized orbits (aka photo-
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
gravitational orbits). The selection of the sequence of orbits more appropriate for a certain mission
depends mainly on the uncertainties of the gravity field (safety issues), surface observability (for
science and navigation), and propellant expenditure.
The asteroid has a large angular size and a Wide Angle Camera (FOV ∼> 20°) is the best option for
vision-based navigation. The image processing can use different techniques to obtain relative
measurements. The image processing might be done on ground (as was demonstrated by Hayabusa
mission). It if is done autonomously, a separate processing unit is usually used not to load the on-
board computer. If good knowledge of the size and shape of the asteroid is available, limb-related
measurements, such as limb-star angular distance or star occultation time, or for image matching
can be applied.
If the images have enough resolution of the surface, then known landmark identification or
unknown feature tracking can be used (Fig. 13). Known landmark mapping permits direct positioning
of the SC state in asteroid body-fixed frame. Unknown landmark tracking provides measurements of
the velocity relative to the surface surface-relative frame. Thus, these observations must be used in
combination with other measurements for complete state estimation. The unknown landmark
measurements are used as complementary measurements, in particular when there are not enough
identified known landmarks. An interesting option to simplify the image processing is to deliver
markers on the surface (Hayabusa strategy). The markers can include LEDs or can be illuminated
with a flash. The objective is the markers to be the brightest objects in the image.The markers serve
as beacons (artificial landmarks) to land in a given position close to them.
Fig. 13. Feature extraction and tracking algorithm for known and unknown landmark observations (©University of
Dundee)
The use of an altimeter for navigation is mandatory if information on the shape and size is not
available from science observations or previous phases. In any case, altimetry increases significantly
the robustness and accuracy of the navigation system. If a scanning LIDAR can be used, then the
DEM (2D images of the distances) can be used for terrain relative navigation. LIDAR images provides
more information than combined optical images and altimetry but need to be traded with the total
cost, mass, power and volume (note that redundancy in the navigation chain is needed for a one-
failure tolerant system).
The objective of the descent and landing (D&L) can be to land softly or to hover at very close
distance of a selected site. The GNC must be fully autonomous below the so-called low gate (analogy
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
with lunar missions).Some navigation sensors (e.g. optical cameras and LIDAR) needs to avoid large
angular rates and to maintain the landing site continuously in view. Thus, for trajectory control the
propulsion system should be able to provide pure force (no torque in nominal conditions).
The orientation of the SC wrt the surface must assure that there is no risk of collision of solar arrays
or tumbling after touch-down. Thus, LIDAR or multiple tilted altimeters provides information of the
SC orientation wrt the surface. The accommodation of the sensors needed for the approach,
proximity and landing operations must assure good visibility during all phases. For instance star
tracker must not be blinded or dazzled during the proximity operations, or legs cannot appear in the
WAC images. A possible sensor configuration is shown in Fig. 14 for an asteroid sample return
mission (communication antennas and solar arrays not shown for clarity).
Fig. 14. Possible sensor configuration for proximity and landing operations around small asteroid (©OHB-System AG)
Further reading
1. Battin, R. H., An Introduction to the Mathematics and Methods of Astrodynamics, AIAA
education series, American Institute of Aeronautics and Astronautics, Inc, Reston, VA, USA,
1999
2. Wie, B., Space Vehicle Dynamics and Control, AIAA education series, American Institute of
Aeronautics and Astronautics, Inc, Reston, VA, USA, 1998
3. Fehse, W., Automated Rendezvous and Docking of Spacecraft, Cambridge University Press,
Cambridge, UK, 2003.
4. Simon, D., Optimal State Estimation, John-Wiley & Sons, Inc., Hoboken, NJ, USA, 2006
5. Bryson, A. E., and Ho, Y. C., Applied Optimal Control – Optimization, estimation, and control,
2nd ed., Hemisphere Publishing Corporation, Washington D.C., 1975, Chap. 5.
List of figures
Fig. 1. Typical GNC architecture. .............................................................. Error! Bookmark not defined.
Fig. 2. Functional architecture of a Kalman filter for autonomous navigation....................................... 5
Fig. 3. Illustration of star and limb related observables and image matching (positioning) from Moon
images (© GMV). .................................................................................................................................... 7
Fig. 4. Illustration of differential guidance including intermediate corrective maneuvers (©GMV) ..... 9
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
Glossary
AIV Assembly, Integration and Validation
APE Absolute Pointing Error
CoB Center of Brightness
CoM Center of Mass
DEM Digital Elevation Map
D&L Descent and Landing
DIMES Descent Image Motion Estimation System
ECSS European Cooperation for Space Standardization
EDL Entry, Descent and Landing
FDIR Failure Detection, Identification and Recovery
FOV Field Of View
GPS Global Positioning System
GNC Guidance, Navigation and Control
GNSS Global Navigation Satellite Systems
HIL Hardware-in-the-loop
HW Hardware
IP Image Processing
LED Light Emission Diode
LEO Low Earth Orbit
LIDAR Light Detection and Ranging
LOS Line-of-Sight
LQR Linear-Quadratic Regulator
MER Mars Exploration Rovers
MIB Minimum Impulse Bit
MIL Model-in-the-loop
MIMO Multiple Input Multiple Output
MVM Mission and Vehicle Management
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
References
[1] Riedel, J. E., Bhaskaran, S., Desai, S., Han, D., Kennedy, B., Null, G. W. et al. “Autonomous Optical Navigation
(AutoNav) DS1 Technology Validation Report,” Jet Propulsion Laboratory, California Institute of Technology, 26 April,
2004.
[2] Bank, T., Frazier, W., Blume, W., Kubitschek, D.et al. “Deep Impact: 19 Gigajoules Can Make Quite an Impression”,
24th AAS Guidance and Control Conference, AAS 01-022, January 2001
[3] Uo, M., Shirikawa, K., Hasimoto, T. et al. “Hayabusa’s Touching-Down to Itokawa- Autonomous Guidance &
Navigation,” 16th AAS/AIAA Space Flight Mechanics Conference, AAS 06-214, January 2006
[4] Brooks, C. G., Grimwood, J. M., and Swenson Jr., L. S., “Astronavigation - The First Apollo Contract,“ Chariots for
Apollo: A History of Manned Lunar Spacecraft, The NASA History Series, NASA SP-4205, Washington DC, 1979
[5] Fehse, W., “The Onboard Rendezvous Control System,” Automated Rendezvous and Docking of Spacecraft,
Cambridge University Press, Cambridge, UK, 2003, pp. 171–217.
[6] ECSS Ground systems and operations — Telemetry and telecommand packet utilization, ECSS-E-70-41A, 30/01/2003
[7] Roy. E. Gladden. “Mars Reconnaissance Orbiter: Aerobraking Sequencing Operations and lessons Learned”.
Proceedings of ESA/EUMETSAT/AIAA’s SpaceObs 2008 Conference
[8] P. Zarchan. “Tactical and Strategic Missile Guidance, 2nd Edition”, Progress in Aeronautics and Astronautics, Vol. 157,
1994
[9] Bierman, G. J., Factorization Methods for Discrete Sequential Estimation, Academic Press, New York, 1977, pp. 150-
152
[10] Simon, D., Optimal State Estimation, John-Wiley & Sons, Inc., Hoboken, NJ, USA, 2006
[11] Bryson, A. E., and Ho, Y. C., Applied Optimal Control – Optimization, estimation, and control, 2nd ed., Hemisphere
Publishing Corporation, Washington D.C., 1975, Chap. 5.
[13] Prieto-Llanos, T., Gil-Fernández, J., Corral, C., “Autonomous Optical Navigation for Manned Lunar Missions”, 7th
International ESA Conference on Guidance, Navigation & Control Systems, 2-5 June 2008, Tralee, County Kerry, Ireland.
International Handbook of Space Technology Attitude Determination, Guidance, Navigation, and Control
[14] Acıkmese, A. B., “Enhancements on the Convex Programming Based Powered Descent Guidance Algorithm for Mars
Landing”, AIAA-2008-6426, , AIAA/AAS Astrodynamics Specialist Conference and Exhibit, 18 - 21 August 2008, Honolulu,
Hawaii.
[15] Gil-Fernandez, J., and Gomez-Tierno, M. A. “Optimal Guidance of Low-Thrust Trajectories”, Journal of Guidance,
Control, and Dynamics, Vol. 33, No. 6, November-December 2010, pp. 1913-1917.
[16] Acıkmese, A. B., and Carson III, J. M., “Small Body GN&C Research Report: A Robust Model Predictive Control
Algorithm with Guaranteed Resolvability”, Jet Propulsion Laboratory Internal Document, JPL D-32947, September 2,
2005
[17] Chasset, C., Larsson, R., Nilsson, F. et al., “Guidance, Navigation, and Control Experiments on the PRISMA In-Orbit
th
Test Bed”, IAC-07-C1.6.01, 58 International Astronautical Congress, Hyderabad, India, 24-28 September 2007
[18] Peters, T., Strippolli, L., “Guidance For Elliptic Orbit Rendezvous”, AAS 10-175, 20th AAS/AIAA Space Flight
Mechanics Meeting, San Diego, USA, 14-17 February 2010
[19] Yamanaka, K., Ankersen, F., “New State Transition Matrix for Relative Motion on an Arbitrary Elliptical Orbit”,
Journal of Guidance, Control and Dynamics, Vol. 25, No. 1, 2002, pp. 60-66
[20] W. M. Owen et al. “Optical Navigation for Deep Impact”, AAS 06-176, Advances in the Astronautical Sciences,
Volume 124 Part II, 2006.
[21] T. Kominato, M. Matsuoka, M. Uo et al. “Optical Hybrid Navigation in Hayabusa – Approach, station Keeping &
Hovering”, AAS 06-209, AAS/AIAA Space Flight Mechanics Conference, Tampa, Jan. 2006
[22] M. Uo, K. Shirikawa, T. Hasimoto et al. “Hayabusa’s Touching-Down to Itokawa – Autonomous Guidance and
Navigation”, AAS 06-214, AAS/AIAA Space Flight Mechanics Conference, Tampa, Jan. 2006
[23] Bhaskaran, S., “Autonomous Navigation for Deep Space Missions”, ACGSC Meeting, March 1-3, 2006
[24] J. Gil-Fernandez, R. Panzeca, C. Corral, “Impacting small Near Earth Objects,” Journal of Advances in Space Research.
COSPAR, Elsevier Ltd., vol. 42, No. 8, pp. 1352-1363, 15 October 2008 (doi:10.1016/j.asr.2008.02.023).
[25] J. Gil-Fernández, R. Cadenas-Gorgojo, T. Prieto-Llanos, M. Graziano, R. Drai, “Autonomous GNC Algorithms for
Rendezvous Missions to Near-Earth-Objects,” AIAA 2008-7087, AIAA/AAS Astrodynamics Specialist Conference and
Exhibit, 18 - 21 August 2008, Honolulu, Hawaii.