You are on page 1of 7

53rd IEEE Conference on Decision and Control

December 15-17, 2014. Los Angeles, California, USA

Tracking of the UAV trajectory on the basis of bearing-only


observations
Alexander Miller and Boris Miller
Abstract This work considers the tracking of the UAV
(unmanned aviation vehicle) path on the basis of bearingonly observations including azimuth and elevation angles. The
significance of this research becomes clear in the case when
GPS either does not work at all or produce the high level
of the measurement errors. It is assumed that either UAVs
opto-electronic cameras or radar systems are able to capture
the angular position of objects with known coordinates and
to measure the azimuth and elevation angles of the sight
line. Such measurements involve the real position of UAV in
implicit form, and therefore some of nonlinear filters such as
Extended Kalman filter (EKF) or others may be used in order
to implement these measurements for UAV control. However, all
such approximate nonlinear filters produce the estimations with
unknown bias and quadratic errors. This peculiarity prevents
the data fusion in more or less regular way. Meanwhile, there
is well-known method of pseudomeasurements which reduces
the estimation problem to the linear settings. In this article
we develop the modified pseudomeasurement method without
bias and with the possibility to evaluate the second moments
of the UAV position errors which helps to realize the data
fusion. On the basis of this filtering algorithm we develop
the control algorithm for tracking of given reference path
under external perturbation and noised angular measurements.
Modelling examples show the nice performance of the control
algorithm.

I. INTRODUCTION
Modern UAVs navigation systems use the standard elements of INS (inertial navigation systems) along with GPS,
which permit to correct the bias and improve the UAV
localization which is necessary for resolving mapping issues,
targeting and reconnaissance tasks [13]. The performing of
these tasks in autonomous flights needs so-called data fusion
which is a difficult task especially for small UAVs which
are implemented usually at low altitude, and therefore have
relatively high level of the barometric altitude sensor errors.
There are systematic errors which are inherent in GPS usage
at low altitude, and they are also rather important and for
this reason special approaches such as robust nonlinear [11],
[12] or adaptive [5] filtering are necessary. An alternative to
the GPS is the usage of various complementary measurement
systems such as optoelectronic cameras and passive radio- or
This work was supported by Russian Basic Research Foundation Grant
13-01-00406.
Alexander Miller is with A. A. Kharkevich Institute for Information
Transmission Problems, Russian Academy of Sciences, Moscow, Russia.
His work was partially supported by Russian Basic Research Foundation
Grant 13-01-90705. amiller@iitp.ru
Boris Miller is with A. A. Kharkevich Institute for Information Transmission Problems, Russian Academy of Sciences, Moscow, Russia and with
the School of Mathematical Sciences, Monash University, Clayton, 3800,
Victoria, Australia. boris.miller@monash.edu

978-1-4673-6090-6/14/$31.00 2014 IEEE

ultrasonic locators which give direction-finding (or bearingonly) observations. In order to get the metric data from such
observations one needs first to make observations from different positions (i.e. triangulation) and then to use the nonlinear
filtering. However, all nonlinear filters either have unknown
bias [3] or very difficult for on board implementation like
the Bayesian type estimation [4], [6]. Approaches to the
position estimations based on bearing-only observations had
been analysed long ago especially for submarine applications
[7] and nowadays for UAV applications [13], [16].
Comparison of different nonlinear filters for bearingonly observations in the issue of the ground-based object
localization [10] shows that EKF (extended Kalman filter),
unscented Kalman filter, particle filter, and pseudomeasurement filter give almost the same level of accuracy, while
the pseudomeasurement filter is usually more stable and
simple for onboard implementation. This observation is in
accordance with rather old results [7], where all these filters
were compared in the issue of moving objects localization.
It have been mentioned that all these filters have bias which
makes their use in data fusion issues rather problematic. The
principle demand to such filters in data fusion is the nonbiased estimate with known mean square characterization
of the error. Among the variety of possible filters only the
pseudomeasurement filter might be modified to satisfy the
data fusion demands. The idea of such nonlinear filtering
has been developed by V. S. Pugachev and I. Sinitsyn in
the form of so-called conditionally-optimal filtering [14],
which provides the non-biased estimation within the class of
linear filters with the minimum mean squared error. In this
paper we develop such filter for the UAV position estimation
and give the algorithm for the path planning along with the
reference trajectory under external perturbations and noisy
measurements. The structure of the article is as follows: the
next Section 2 is the problem statement, in Section 3 we
develop the conditionally optimal filter for the UAV position
estimation on the basis of the bearing-only observations of
the ground-based objects with known coordinates, in Section
4 we develop the locally optimal control algorithm and in the
Section 5 we provide the results of numerical experiments.
Section 6 is the conclusions and description of directions of
future research.
II. PROBLEM STATEMENT
The problem of the bearing-only filtering is considered to
determine the coordinates of UAV which could observe some
objects with precisely known coordinates.
Assumptions:

4178

On the ground under the trajectory of UAV there is


a set of objects which can be captured by the UAV
observation system. These object can be either the
set of well recognizable objects or the network of
radio-beacon stations with well specified frequency and
known coordinates. So the UAV control system can
easily recognize which object is captured at the current
time. So for observed objects the UAV can measure the
bearing angles such as azimuth and angle of elevation.
The real task of the UAV may be an inspection of the
territory or a reconnaissance of the particular area. Since
we did not expect the possibility of usage the GPS, like
in military applications, we did not expect the usage
of the active measurement also. So the task is to make
navigation on the basis of bearing-only observations.
Anyway the UAV has the standard set of INS devices,
which enables to perform the flight with some degree of
accuracy which, however, is not enough for the mission
completion:
1) 3-axis gyroscope, measuring the angles with respect to the absolute coordinate systems, such as
yaw, pitch, and roll angles;
2) 3 - axis accelerometer;
3) magnetometer, measuring the azimuth;
4) barometric altimeter;
5) velocity sensor.
The UAV motion is defined by two angles, namely:
(t) (the angle between the projection of velocity to
the plane y0x and the axis OX) and (t) (the angle
between the the vector of velocity and the plane y0x)
(see Fig. 1). We assume the reference motion has the
constant linear velocity and varying altitude.
The UAV control system determines the bearings on the
observed object, that is either the radio-beacon or some
known object in the field of view of the opto-electronic
camera.

B. Measurements
Assume that (Xi , Yi , Zi ) are the coordinates of i - th object
and (i (tk ), i (tk ) are the bearing angles on that object. At
moment tk these angles satisfy the relations:
Yi Y (tk )
I (t ) = tan i (tk ) + 0k
Xi X(tk ) i k
Zi Z(tk )
p
Ii (tk ) = tan i (tk ) + 00k ,
(Xi X(tk ))2 + (Yi Y (tk ))2
(2)
2
0
00
where one can assume that k WN (0, 1 ), k
WN (0, 22 ) are uncorrelated random variables with zero
means and variances 12 , 22 , defined as errors in measurement of tangents of angles i (tk ), i (tk ), and forming the
white noise sequences.
Remark 1: In most of works based on the method of
pseudomeasurements another model is used. It assumes the
measurements of the angles with gaussian errors (see [2], [7]
and most of successful works). However, in the real definition
of the object position in the image or in the matrix of sensors
the system measures the distance between the object image
and the midst of the sensor, that is the tangent of the bearing
angle. This simple observation helps to find the unbiased
estimation of UAV coordinates.
One can rewrite equation (2) for angle i (tk ) as follows
Zi Z(tk )
sin i (tk )
sin i (tk )Ii (tk ) =
+ 00k
Yi Y (tk )
cos i (tk )

(3)

Remark 2: the indicator function Ii (tk ) = 1 if at tk the


bearing of i-th object occurs and Ii (tk ) = 0 otherwise. For
convenience we assume that Ii (tk ) = 1.
So at the moment tk the UAV control system determines
the angles i (tk ) and i (tk ), related with coordinates of
UAV (X(tk ), Y (tk ), Z(tk )) as follows:

A. Model of the UAV motion


We assume 3D UAV motion described by coordinates

(X(tk ), Y (tk ), Z(tk )) at times tk = kt,


1, 2, ..... satisfying the following equations

(Yi Y (tk )) cos i (tk ) (Xi X(tk )) sin i (tk )

k =

(4)
= 0k (Xi X(tk )) cos i (tk ) = 0k

X(tk+1 ) = X(tk ) + V cos (tk ) cos (tk )t + Wkx ,

Y (tk+1 ) = Y (tk ) + V cos (tk ) sin (tk )t + Wky ,

(Zi Z(tk )) sin i (tk ) cos i (tk ) (Yi Y (tk )) sin i (tk )

Z(tk+1 ) = Z(tk ) + V sin (tk )t + Wkz ,

= 00k (Yi Y (tk )) cos i (tk ) = 00k .

(1)
where (Wkx , Wky , Wkz ) are uncorrelated random perturbations acting along axis (OX, OY, OZ), having zero
means and variances (x2 , y2 , z2 ), correspondingly;
the controls (tk ) and (tk ) are the angles between
projection of vector-velocity on the the plane y0x and
the axis OX, and between vector of velocity and
the axis OX, correspondingly, these angles define the
nominal UAV motion.

(5)

III. M ODIFIED METHOD OF PSEUDOMEASUREMENTS


A. Linear measurements model
The idea of the pseudomeasurement method is to separate
in (4) and (5) the observable and non observable values,
which gives the following observation equations:

4179

1) Prediction: The prediction is obtained by assuming


that at the moment tk+1 the values of i (tk+1 ) and i (tk+1 )
will be known

m0k = Yi cos i (tk ) Xi sin i (tk ) =

k+1 ) = X(t
k ) + V cos (tk ) cos (tk )t,
X(t

Y (tk+1 ) = Y (tk ) + V cos (tk ) sin (tk )t,


k+1 ) = Z(t

Z(t
 0 k ) +V sin (tk )t,
m
k+1
m
k+1 =
m
00k+1

Y (tk ) cos i (tk ) X(tk ) sin i (tk )


+0k (Xi X(tk )) cos i (tk )
m00k = Zi sin i (tk ) cos i (tk ) Yi sin i (tk )
= Z(tk ) sin i (tk ) cos i (tk ) Y (tk ) sin i (tk )

Z(t
k+1 ) sin i (tk+1 ) cos i (tk+1 )
Y (tk+1 ) sin i (tk+1 )

+00k (Yi Y (tk )) cos i (tk )


So the measurement vector has the following form:

mk =

m0k
m00k

Y (tk ) cos i (tk ) X(tk ) sin i (tk )


+0 (Xi X(tk )) cos i (tk )
k

Z(tk ) sin i (tk ) cos i (tk ) Y (tk ) sin i (tk )


+00k (Yi Y (tk )) cos i (tk )

(6)
Thereby we obtain the system (6) of linear measurement
equations, though the noise variance depends on unobservable coordinates. By using V. S. Pugachev method [14]
one can obtain the unbiased estimation and the variance
evaluation with the aid of prediction-correction filter [9].

Assume that at the moment tk we have unbiased estima k ), Y (tk ), Z(t


k ) such that
tion X(t
(7)

Assuming that the motion perturbations and the UAV position are uncorrelated we obtain
i
h
k+1 ))2 = P xx (tk ) + x2 ,
P xx (tk+1 ) = E (X(tk+1 ) X(t
P xy (tk+1 ) = P xy
h (tk ),
i
yy

P (tk+1 ) = E (Y (tk+1 ) Y (tk+1 ))2 = P yy (tk ) + y2 ,


h
i
k+1 ))2 = P zz (tk ) + 2 ,
P zz (tk+1 ) = E (Z(tk+1 ) Z(t
z
P xz (tk+1 ) = P xz (tk ),
P yz (tk+1 ) = P yz (tk ).
To calculate values
k+1 ))(mk+1 m
P xm (tk+1 ) = E(X(tk+1 ) X(t
k+1 )T ,
ym

P (tk+1 ) = E(Y (tk+1 ) Y (tk+1 ))(mk+1 m


k+1 )T ,
zm
k+1 ))(mk+1 m
P (tk+1 ) = E(Z(tk+1 ) Z(t
k+1 )T ,
P mm (tk+1 ) = E(mk+1 m
k+1 )(mk+1 m
k+1 )T

B. Prediction-correction estimation

k ), Y (tk ), Z(t
k )) = (X(tk ), Y (tk ), Z(tk ))
E(X(t

Y (tk+1 ) cos i (tk+1 )


k+1 ) sin i (tk+1 )
X(t

we use the following relations


 0

0k+1
mk+1 m
00k+1
m00k+1 m

with the following matrix of the mean-square errors

(Y (tk+1 ) Y (tk+1 )) cos i (tk+1 )


k+1 )) sin i (tk+1 )
(X(tk+1 ) X(t
0
+k+1 (Xi X(tk+1 )) cos i (tk+1 )

(Z(tk+1 ) Z(t
k+1 )) sin i (tk+1 ) cos i (tk+1 )

(Y (t

k+1 ) Y (tk+1 )) sin i (tk+1 )


00
+k+1 (Yi Y (tk+1 )) cos i (tk+1 )

k ) T
k ) X(tk ) X(t
X(tk ) X(t
P (tk ) = E Y (tk ) Y (tk ) Y (tk ) Y (tk )
k)
k)
Z(tk ) Z(t
Z(tk ) Z(t

xx
P (tk ) P xy (tk ) P xz (tk )
= P xy (tk ) P yy (tk ) P yz (tk )
P xz (tk ) P yz (tk ) P zz (tk )

the identities

(8)
Problem 1: The next problem is to get the estimation
of the UAV position at time tk+1 on the basis of pre k ), Y (tk ), Z(t
k ), P (tk ), observations
vious estimations X(t
i (tk+1 ), i (tk+1 ), known position of i-th observable object
(Xi , Yi , Zi ), and known parameters of the motion equations
(1) in the interval [tk , tk+1 ]. In other words one needs to
k+1 ), Y (tk+1 ), Z(t
k+1 )
find the unbiased estimations of (X(t
and the matrix P (tk+1 )) on the basis mk and the motion
parameters. These estimates must satisfy (7),(8).

k+1 ) (X(tk+1 ) X(t


k+1 )),
Xi X(tk+1 ) = Xi X(t
Yi Y (tk+1 ) = Yi Y (tk+1 ) (Y (tk+1 ) Y (tk+1 )),
k+1 ) (Z(tk+1 ) Z(t
k+1 )),
Zi Z(tk+1 ) = Zi Z(t
where we consider the position of i-th object is to be known,
and use the noncorrelatedness of 0k+1 , 00k+1 and differences
k+1 )), (Y (tk+1 ) Y (tk+1 )) (Z(tk+1 )
(X(tk+1 ) X(t
k+1 )).
Z(t

4180

Finally we get
h
iT
P xm (tk+1 )

Remark 3: The estimates obtained by (12), (11) are unbiased [9],[14] and give the best linear estimates, they are
kept constant until measurement at the time tk+1 > tk and
they must be updated by formulas (12), (11) at that moment.
Of course, these estimates are not equal to the conditional
expectations, but they are projections on the set of preceding
measurements m1 , ..., mk , therefore, they are orthogonal to
the linear space L{m1 , ..., mk }.


 0

0k+1
k+1 )) m00k+1 m
= E (X(tk+1 ) X(t
=
00k+1
mk+1 m

P xy (tk+1 ) cos i (tk+1 )


P xx (tk+1 ) sin i (tk+1 )

P xz (tk+1 ) sin i (tk+1 ) cos i (tk+1 )


P xy (tk+1 ) sin i (tk+1 )

IV. C ONTROL OF UAV

(9)
and similar to (9) we get the analogous equations for
P xm (tk+1 ) and P zm (tk+1 ). In a similar way we calculate
(P mm (tk+1 ))1 (see equation on the top of thext next page),
where


0k+1 )2
E (m0k+1 m
= P yy (tk+1 ) cos2 i (tk+1 ) P xy (tk+1 ) sin 2i (tk+1 )
k ))2
+P xx (tk+1 ) sin2 i (tk+1 ) + 12 ((Xi X(t
xx
2

+P (tk+1 )) cos i (tk+1 )




00k+1 )
0k+1 )(m00k+1 m
E (m0k+1 m
= P yz (tk+1 ) sin i (tk+1 ) cos i (tk+1 ) cos i (tk+1 )
P yy (tk+1 ) cos i (tk+1 ) sin i (tk+1 )
P xz (tk+1 ) sin2 i (tk+1 ) cos i (tk+1 )
+P xy (tk+1 ) sin i (tk+1 ) sin i (tk+1 )


00k+1 )2
E (m00k+1 m
= P zz (tk+1 ) sin2 i (tk+1 ) cos2 i (tk+1 )
P yz (tk+1 ) sin i (tk+1 ) sin 2i (tk+1 )
+P yy (tk+1 ) sin2 i (tk+1 ) + 22 ((Yi Y (tk ))2
+P yy (tk+1 )) cos2 i (tk+1 )

Control of UAV which ensures its motion along the reference trajectory may be determine on the basis of standard
deterministic linear-quadratic approach [15], however, the
problem of control on the basis of bearing-only observation
is defferent from standard one. It should be stressed that this
problem is non-linear and cannot be solved in the standard
way. The problem of the optimal control for system (1) is
stochastic one with incomplete information and do not have
the explicit solution. However, for practical reasons one can
simplify it if to consider the locally optimal control. Here
we discuss two problems:
Problem 2: Find the locally optimal controls (tk ) and
(tk ) at constant velocity V aimed to keep the motion of
UAV along the reference trajectory.
Problem 3: Find
the
locally
optimal
controls
(tk ), (tk ), V (tk ) aimed to keep the motion of UAV
along the reference trajectory.
A. Solution of Problem 2
Assume that the reference trajectory is defined by equations
Xnom (tk+1 ) = Xnom (tk ) + V cos (tk ) cos (tk )t

2) Correction: After getting mk+1 (more precisely the


tangents of angles i (tk+1 ) and i (tk+1 )) one can obtain
the estimate of the UAV position at the time tk+1 . Therefore,
the solution of Problems 1 has a form:

k+1 )
X(tk+1 )
X(t
Y (tk+1 ) = Y (tk+1 )
k+1 )
k+1 )
Z(t
Z(t

P xm (tk+1 )
+ P ym (tk+1 ) (P mm (tk+1 ))1 (mk+1 m
k+1 )
P zm (tk+1 )
(10)
and the matrix of the mean square errors is equal:

Ynom (tk+1 ) = Ynom (tk ) + V cos (tk ) sin (tk )t


Znom (tk+1 ) = Znom (tk ) + V sin (tk )t,
(12)
where functions (tk ) and (tk ) are given in advance and
velocity V is constant.
The real trajectory differs from the nominal one due to
the presence of perturbations and the controls to be applied
c (tk ) and c (tk ):
X(tk+1 ) = X(tk ) + V cos c (tk ) cos c (tk )t + Wkx
Y (tk+1 ) = Y (tk ) + V cos c (tk ) sin c (tk )t + Wky

P (tk+1 ) = P (tk+1 )

P xm (tk+1 )
P ym (tk+1 ) (P mm (tk+1 ))1
P zm (tk+1 )

Z(tk+1 ) = Z(tk ) + V sin c (tk )t + Wkz

T
P xm (tk+1 )
P ym (tk+1 ) .
P zm (tk+1 )
(11)

(13)
The locally optimal controls c (tk ) and c (tk ) are obtained
Y , Z)
in order to
on the basis of current estimates (X,
minimize the deviation from the reference path at the next
time tk+1

4181

 
(P mm (tk+1 ))1 = E

00k+1 )
0k+1 )(m00k+1 m
(m0k+1 m
0k+1 )2
(m0k+1 m
2
00
00
00
00
0
0
k+1 )
(mk+1 m
k+1 )
k+1 )(mk+1 m
(mk+1 m


= cos c (tk ) cos c (tk )

min

c (tk ),c (tk )

+ cos c (tk ) sin c (tk )

(14)
Minimization of (14) gives the solution of Problem 2 (see
(15), (16) below):
where


+ sin c (tk )


X(tk )
+ V cos (tk ) cos (tk )
t

Y (tk )
+ V cos (tk ) sin (tk )
t


Z(tk )
+ V sin (tk ) .
t
(19)

V. E STIMATION OF THE UAV POSITION AND CONTROL .


M ODELLING RESULTS .

k ),
X(tk ) = Xnom (tk ) X(t
Y (tk ) = Ynom (tk ) Y (tk ),
k ).
Z(tk ) = Znom (tk ) Z(t
Remark 4: We omit the derivations since they are presented in the article submitted to Automation and Remote
Control [8].
B. Solution of Problem 3
Here we consider the Problem 3 where the reference
trajectory is defined by the same equation (12), while the
controlled trajectory is defined by equations:
X(tk+1 ) = X(tk ) + V (tk ) cos c (tk ) cos c (tk )t + Wkx
Y (tk+1 ) = Y (tk ) + V (tk ) cos c (tk ) sin c (tk )t + Wky
Z(tk+1 ) = Z(tk ) + V (tk ) sin c (tk )t + Wkz

(17)
As it was done above we determine velocity V (tk )
and angles c (tk ),c (tk ) on the basis of the current es Y , Z)
in order to minimize the deviation
timates (X,
from the reference path on the next step, so the controls
(c (tk ), c (tk ), V (tk )) give the minimum to the expression
E{(X(tk+1 ) Xnom (tk+1 ))2 + (Y (tk+1 ) Ynom (tk+1 ))2
+(Z(tk+1 ) Znom (tk+1 ))2 }

V (tk )

E{(X(tk+1 ) Xnom (tk+1 ))2


+(Y (tk+1 ) Ynom (tk+1 ))2 + (Z(tk+1 ) Znom (tk+1 ))2 }

1

min
c (tk ),c (tk ),V (tk )

(18)
It gives the solution of the Problem 3. Angular controls
remain the same as above (15), (16) while velocity has a
form:

It is very difficult to evaluate the quality of the estimation


and control since the system 1 is nonlinear. So we use the
modelling under the following conditions:
The initial point of motions is (X0 , Y0 , Z0 ) = (0, 0, 0).
Nominal velocity of UAV is constant V = 50.
Number of intervals of the length t equals n = 200.
t = 10.
The standard deviations of noises in dynamic equations
equal to x = y = z = 100.
Standard deviations of the angles measurement errors
are equal to = 0.003, that is approximately 10 that
is achievable for radio-measurement and much higher
than that can be achieved with optical measurements.
For referent trajectory the laws of the angle change are
as follows
k
k
cos( )
cos( )
20 , (t ) =
20 .
(tk ) =
k
3
4
We assume the presence of three observable objects
installed at different heights with the following coordinates:
1) (X1 , Y1 , Z1 ) = (10000, 7000, 6000) visible from
distance R1 = 4000,
2) (X2 , Y2 , Z2 ) = (25000, 5000, 0), R2 = 7000,
3) (X3 , Y3 , Z3 ) = (50000, 6000, 0), R3 = 7000.
Initial conditions for estimates are as follows
1 ), Y (t1 ), Z(t
1 ))
(X(t
=
(X0 , Y0 , Z0 ) so their
variances are equal P xx (t1 ) = P xy (t1 ) = P xz (t1 ) =
P yz (t1 ) = P yy (t1 ) = P zz (t1 ) = 0.
Next Fig. 2 shows the dependence of X(t), Y (t), Z(t) for
nominal control (without feedback) (left) and for angular
control only, as given by solution of Problem 2 (right). First
level of the picture gives the value of the indicator function
which is equal 1 if the UAV can observe the object and
0 otherwise. Next three levels show the dependence of X,
Y , Z and their estimates for the times tk . The last three
show the dependence of the mean square errors P xx , P yy ,

4182

tan c (tk ) =

Y (tk ) + V t cos (tk ) sin (tk )


X(tk ) + V t cos (tk ) cos (tk )

(15)

and
tan c (tk ) =
(16)
Z(tk ) + V t sin (tk )
.
cos c (tk )(X(tk ) + V t cos (tk ) cos (tk )) + sin c (tk )(Y (tk ) + V t cos (tk ) sin (tk ))

P zz upon the time. We should stress the abrupt reduction of


mean square errors on the intervals of the angular-position
measurements.
Even if the estimation is good enough there is substantial
error in the path control. In order to remove it we use the
additional velocity control given by solution of Problem 3.
The corresponding results are shown on Fig. 3. The positions
of pictures are the same as in above example. One can
observe the substantial reduction of the tracking error with
respect to above case.
VI. C ONCLUSIONS
So the use of this sub-optimal filter provides the good
quality of the UAV trajectory estimation. The angular and
velocity controls permit to realize the good tracking in the
area of bearing-only measurement. However, we do not aim
to demonstrate the possibility of good control on the basis of
bearing-only observation. Of course, this type of observations
must be used with other measurement systems and our results
really open the way to the data fusion. The principal reason
for this is that we obtain the unbiased UAV position estimate
with known mean square error. Only both of these properties
permit to make the data fusion in an optimal way.

[9] B. M. Miller and A. R. Pankov. Theory of random processes [in


russian]. Moscow, Phizmatlit, 2007.
[10] B. M. Miller, K. V. Stepanyan, A. B. Miller, K. V. Andreev. and
S. N. Khoroshenkikh. Optimal filter selection for UAV trajectory
control problems. Proceeedings of the 37-th Conference on Information Technology and Systems - 2013. Conference for Young Scientists
and Ebngineers. IITP RAS, 1-6 September 2013, Kaliningrad, Russia.
2013, pp. 327333.
[11] A. Nemra, A. Nabil, A. Tsourdo, and B. White. Robust nonlinear
filtering for INS/GPS UAV localization. Proceedings of the 16th Mediterranean Conference on Control and Automation Congress
Centre, Ajaccio, France June 25-27, 2008, pp. 695702.
[12] A. Nemra and N. Aouf. Robust INS/GPS Sensor Fusion for UAV
Localization Using SDRE Nonlinear Filtering. IEEE Sensors Journal,
vol. 10, no. 4, 2010, pp. 789798.
[13] R. W. Osborn, III and Y. Bar-Shalom. Statistical Efficiency of Composite Position Measurements from Passive Sensors. IEEE Transactions
on Aerospace and Electronic Systems, vol. 49, no. 4, 2013, pp. 2799
2806.
[14] V. S. Pugachev and I. N, Sinitsyn. Stochastic Differential Systems.
Analysis and Filtering. Wiley, 1987.
[15] P. B. Sujit, S. Saripalli, and J. Borges Sousa. Unmanned aerial vehicle
path following: A survey and analysis of algorithms for Fixed-Wing
unmanned aerial vehicless. IEEE Contol Systems Magazine, vol. 34,
no. 1, 2014, pp. 42-59.
[16] S. Zhang and Y. Bar-Shalom. Efficient Data Association for 3D Passive
Sensors: If I Have Hundreds of Targets and Ten Sensors (or More).
Proceedings of the 14th International Conference on Information
Fusion, Chicago, IL, 2011.

R EFERENCES
[1] K. S. Amelin and A. B. Miller. An Algorithm for Refinement of
the Position of a Light UAV on the Basis of Kalman Filtering of
Bearing Measurements. Journal of Communications Technology and
Electronics, vol. 59, No. 6, 2014, pp. 622-631.
[2] V. J. Aidala and S.C. Nardone. Biased Estimation Properties of
the Pseudolinear Tracking Filter. IEEE Transactions on Aerospace
Electronic Systems, vol. 18, no. 4, 1982, pp. 432441.
[3] D. Belfadel, R. W. Osborne, III, and Y. Bar-Shalom. Bias Estimation
for Optical Sensor Measurements with Targets of Opportunity. Proceedings of the 16-th International Conference on Information Fusion
Istanbul, Turkey, July 9-12, 2013, pp. 18051812.
[4] A. N. Bishop, B. Fidan, B. D. O. Anderson, K. Dogancay, and P. N.
Pathirana. Optimality analysis of sensor-target localization geometries.
Automatica, vol. 46, 2010, pp. 479492.
[5] W. Ding, J. Wang, and A. Almagbile. Adaptive Filter Design for UAV
Navigation with GPS/INS/Optic Flow Integration. Proceedings of the
2010 International Conference on Electrical and Control Engineering,
2010, pp. 26232626.
[6] C. Jauffet, D. Pillon, and A. C. Pignoll. Leg-by-leg Bearings-Only
Target Motion Analysis Without Observer Maneuver. J. of Advances
in Infromation Fusion, vol. 6, no. 1, 2011. pp. 2438.
[7] X. Lin, T. Kirubarajan, Y. Bar-Shalom, and S. Maskell. Comparison
of EKF, Pseudomeasurement and Particle Filters for a Bearing-only
Target Tracking Problem. Signal and Data Processing of Small Targets
2002, Eds. Oliver E. Drummond, Proceedings of SPIE, vol. 4728,
2002, pp. 240250.
[8] A. B. Miller. Development of the motion control on the basis of
Kalman filtering of bearing-only measurements. Submitted to Automation and Remote Control. 2014.

4183

Fig. 1.

Position od UAV and the observed object

Fig. 2. Angular control. Dependence of the indicator function, coordinates


X, Y, Z, and the mean square errors P xx , P yy , P zz upon tk .

Fig. 3. Angular plus velocity control. Dependence of the indicator function,


coordinates X, Y, Z, and the mean square errors P xx , P yy , P zz upon tk .

4184

You might also like