Content uploaded by Jörg Friedrich Wagner
Author content
All content in this area was uploaded by Jörg Friedrich Wagner on Aug 30, 2016
Content may be subject to copyright.
Wagner 123
From Bohnenberger's Machine to Integrated Navigation Systems,
200 Years of Inertial Navigation
JÖRG F. WAGNER, Stuttgart
ABSTRACT
In 1817, F. Bohnenberger presented an apparatus consisting of a fast spinning rotor with cardanic suspension. His
invention formed the basis for L. Foucault’s epochal studies on gyroscopic sensors in the middle of the 19th century.
The development of instruments like the artificial horizon and the directional gyro is a direct result of this work. Later
on, gyro technology led also to stabilized platforms forming the early variants of inertial navigation systems.
Foucault’s initial intention of using Bohnenberger’s invention as a sensor was the provision of a device offering an
inertial direction reference for surveying and navigation. Due to technical limitations, a gyro is, however, suitable for
such a task during only a limited period of time. To enable a long-term usage, additional, complementary measurements
are required. An artificial horizon therefore includes levelling sensors, and a directional gyro is typically coupled with a
magnetic compass. On the other hand, the main feature of integrated navigation systems is the fusion of data from dis-
similar signal sources. Thus, artificial horizons and directional gyros represent elementary forms of such systems.
Modern integrated navigation systems of high performance consist of the combination of inertial strapdown systems
(being the successor of stabilized platforms) and GPS receivers. In photogrammetry, they have become an essential tool
for direct georeferencing. Nevertheless, their system structure is still traceable to the early gyro instruments mentioned.
This classical system design being outlined in the following opens perspectives for new developments.
1. HISTORICAL DEVELOPMENT OF GYROSCOPIC INSTRUMENTS
Working on a simple experimental proof for the rotational motion of the earth, the French physicist
Léon Foucault introduced in 1852 the term Gyroscope for an instrument being able to observe such
movements (Foucault, 1852). Besides his well-known pendulum, these studies concentrated on gy-
ros with cardanic suspension (Figure 1). Foucault recognized especially that a well-directed re-
straint of the motion of the gimbals (like blocking one degree of freedom of the suspension) leads to
specific indicators or sensors detecting different rotation components (Broelmann, 2002). With that,
he paved the way for such important gyro instruments like the artificial horizon, the directional
gyro, and the gyrocompass (Sorg, 1976). Moreover, this was also the basis for the development of
stabilized platforms (Wrigley, 1977) leading into modern inertial navigation systems.
L. Foucault is, however, not the originator of gyros with cardanic suspension. He was familiar with
this mechanical principle because such instruments were already employed in many French schools
to explain the precession of the earth rotation axis. This matter was born of the initiative of the
French mathematician Pierre-Simon Laplace, who referred likely to an initial specimen of the École
Polytechnique in Paris (Poisson, 1813). The inventor of this device was J.G. Friedrich Bohnenber-
ger (1765 – 1831), a former Professor for mathematics, astronomy, and physics at the University of
Tübingen, Germany. Using the drawing shown in Figure 1, he explained retroactively but for the
first time the design and the use of gyros with cardanic suspension (Bohnenberger, 1817). As F.
Bohnenberger could not yet know the term gyroscope, he called the device simply Machine. There-
fore, the Machine of Bohnenberger was the basis for Foucault’s epochal work on gyros.1
The initial intention of using a gyro as a sensor was the provision of an instrument offering a self-
contained direction reference for surveying and navigation. This idea was born even before Boh-
1 Following today’s terminology, F. Bohnenberger was geodesist. His scientific work was dedicated mainly to survey-
ing and cartography, and he is regarded as the founder of the Surveying Authority of Württemberg in the southwest of
Germany, too. This fact illustrates how closely geodesy and gyro technology are traditionally connected.
'Photogrammetric Week 05'
Dieter Fritsch, Ed.
Wichmann Verlag, Heidelberg 2005.
124 Wagner
nenberger’s invention: In 1742 or 1743, the Englishman John Serson presented a fast spinning top,
whose upper surface perpendicular to the axis of its rotation was a circular plate of polished metal.
When the top was set in motion, the plain part of its surface became horizontal while maintaining
this behaviour even in a disturbing environment like on swaying ships. With this, J. Serson pro-
posed a solution of the problem of finding a satisfactory horizon for use in sextant observations at
sea when there was fog around the sea horizon. Herefrom the artificial horizon originates.
Unfortunately, J. Serson lost his life in 1744 during a test campaign at sea. It took about 140 years
before his invention was revived in France by Admiral G. Fleuriais. In the meantime, not only L.
Foucault made important contributions to gyro technology. The problem of keeping the rotor at an
approximately constant speed became better controllable by using electrically or pneumatically
driven motors. When people began furthermore to construct and to fly airplanes during the follow-
ing decades, it became necessary also for the pilots to have an artificial horizon in clouds and during
the night. The First World War stimulated particularly the technical progress in this area, and in
1930 the American Elmer Sperry succeeded in designing an especially reliable instrument. It was an
air-driven artificial horizon based on a gyro with cardanic suspension and with controlling the verti-
cal position of the gyro axis by pendulums and air jets (Sorg, 1976).
Besides the artificial horizon, the next important basic type of gyroscopic instruments is the direc-
tional gyro. Referring to Figure 1, F. Bohnenberger (1817) described already its main property
(translated analogously): “While the rotor is spinning around its axis, this axis will maintain perma-
nently that direction which was given to it at the beginning. This will happen well then if one takes
hold of the pedestal of the apparatus and starts moving it. While carrying around the Machine, one
can move in arbitrary directions and with arbitrary velocities, and the axis of the sphere will perma-
nently remain parallel to itself and will permanently stay aligned to north like a magnetic needle if
one has, for example, orientated it at the beginning to north.”
Due to technical limitations like friction in the gimbal bearings, the preservation of the rotor axis
alignment is possible only for a limited period of time. To enable a long-term usage, the orientation
of the gyro has to be aided suitably. Besides the use of a magnetic compass as described below, an
unbalanced inner gimbal of the cardanic suspension proved to be helpful in this connection (Mag-
nus, 1971). The latter case led to the gyrocompass and gyro theodolite. These instruments, called
north-seeking gyros, sense the direction of the earth rotation vector and align themselves thereupon
to north. Following a first attempt of L. Foucault to build such a device, many scientists participated
during the following decades in the development of the gyrocompass. (Hermann Anschütz-
Figure 1: Bohnenberger’s original drawing of a gyro with cardanic suspension (1817).
Wagner 125
Kaempfe and Max Schuler are known in particular.) An electrically driven rotor, a floated suspen-
sion, and the combination of two or three jointly used rotors were important contributions leading to
reliably usable models at the beginning of the 20th century.
Reconceived by means of modern control theory, the artificial horizon, the gyrocompass or gyro
theodolite, and the directional gyro aided by a magnetic compass represent special realizations of
observers. This aspect enables to classify these instruments in a more general way leading directly
to inertial navigation systems and their aiding. For this, section 2 illustrates next the observer prin-
ciple by the example of integrating a directional gyro and a magnetic compass. Section 3 contains
subsequently the transition from single gyro instruments to inertial navigation systems (INS), and
section 4 addresses the special coupling of these systems with GPS satellite navigation receivers.
Motivated by still existing performance limitations of the INS/GPS integration, section 5 outlines
further development work for such systems. Section 6 finishes the paper with a general assessment.
2. SOME BASICS ON INTEGRATED NAVIGATION
2.1. Integrating a Directional Gyro and a Magnetic Compass
Figure 2 illustrates the principle of aiding a directional gyro with a magnetic compass: The fast
spinning rotor with cardanic suspension shows, in principle, an inertially constant orientation. How-
ever, if a torquer appropriately readjusts the outer gimbal of the mechanism, the axis can be coupled
to the mean motion of a magnetic needle. With this, the integrated instrument has the long-term
accuracy like a magnetic compass but shows the much smoother reading of a gyroscopic indicator.
–
+
pick-off
pick-off
torquer
display
vehicle structure
amplifier
αψ
ˆˆ =
α
local vertical
Figure 2: Principle of a directional gyro aided by a magnetic compass.
Figure 3 is a schematic representation of the signal flow of this device (it is very abstract but leads
directly to the more general observer principle): A vehicle equipped with such an instrument shall
be subject to the vertical angular rate
ψ
& leading to a time varying heading
ψ
. Based on
ψ
and the
swaying of the magnetic compass, the needle pick-off generates the signal α. In parallel to the vehi-
cle, the gyro is also subject to
ψ
&. It is now possible to deem the rotor and the gimbals as a mechani-
cal computer integrating
ψ
& numerically with respect to time. The result of this procedure is the dis-
played estimate
ψ
ˆ of the true heading
ψ
. It is identical to the estimate
α
ˆ generated by the gimbal
pick-off. The difference between
α
ˆ and the “true” value α forms the input of a series connection of
an amplifier and a torquer having the property of a low-pass filter and controlling the alignment of
the outer gimbal. With this, the difference between
ψ
and its estimate remains normally small.
126 Wagner
+
α
vehicle
amplifier
and torquer
outer gimbal
pick-off
compass
with pick-off
gyro –
ψ
&
ψ
α
ˆ
ψ
ˆ
Figure 3: Block diagram for the signal flow of Figure 2.
2.2. The Observer Principle
Despite its simple design, the gyro aided by a magnetic compass is representative for the general
properties and system structure used for an extensive class of integrated navigation systems:
• functional combination of dissimilar signal sources with complementary properties (i.e. in par-
ticular various measurement principles),
• increased performance compared with the abilities of the components separately,
• observer principle as authoritative integration scheme.
The last of these three points follows from the fact that Figure 3 is simply a special case from Fig-
ure 4 representing the general case of a multivariable observer (Levine, 1996): The angular rate
ψ
&
corresponds to the input u,
ψ
matches the motion state x and
α
the aiding vector y; the compass
acts as the aiding sub-system.
+
aiding
sub-system
vehicle
–
input vector u motion state xaiding vector y
estimate
u
uy
ˆ
),
ˆ
(
ˆ
model aiding
uxhy =
x
ˆ
estimate
),
ˆ
(
ˆuxfx =
&
)
ˆ
-(
ˆ
oncompensati
yyKx =Δ
model of vehicle
kinematics
Figure 4: Observer principle employed for integrated navigation systems.
The middle branch in the diagram of Figure 4 is the central point of the observer. It functions as a
device reproducing the initially unknown vehicle motion x (lower left block) and the operation of
the aiding sub-system employed (lower right block): A set of ordinary differential equations with
vector function f describes the vehicle kinematics considered by the integrated system, and a set of
algebraic equations with vector function h models the aiding part. To adapt furthermore the esti-
mate x
ˆ to the real state x, a reasonable design of the compensation matrix K is essential. Integrated
navigation systems employ in particular the Kalman filter theory (Gelb, 1989) for this purpose.
In principle, it is possible in Figure 2 to replace the gyro mounted on gimbals by a vertically aligned
rate gyro together with a numerical integrator calculating digitally
ψ
from the measured rate
ψ
&.
Wagner 127
This simple example shows that the realisation of an observer is not bound to a certain technology.
Generally, neither the sensor principles employed nor special mechanical, electrical or software
elements realizing the system of Figure 4 make up the character of an integrated navigation system.
The main distinguishing feature consists in the vehicle and aiding model used (i.e. x, f and h).
2.3. Motional Degrees of Freedom
To illustrate the last statement, Table 1 contains several traditional variants of integrated navigation
systems. They differ primarily in the degrees of freedom considered from the vehicle motion. With
this, it is then possible to regard the various sensor types as being simply a result of this classifica-
tion. (“Which sensor principles are able to detect certain degrees of freedom?”) Furthermore, the
vector x describes plainly the degrees of freedom, whereas f and h assign all measurements to x.
(Wagner (2003) has compiled the most important versions of f and h.)
Degrees of freedom
considered Typical sensor types for
Category
rotational translational u y
directional gyro 1
(yaw) – gyro magnetic compass
artificial horizon 2
(pitch, roll) – gyro
levelling sensors
or accelerometer
attitude and heading
reference systems 3
(pitch, roll, yaw) – gyro
accelerometer,
magnetic compass
classical dead reckoning – 2
(latitude, longitude)
log / wheel sensor,
magnetic compass GPS receiver
extended dead reckoning 1
(yaw) 2
(latitude, longitude)
log / wheel sensor,
gyro
GPS receiver,
magnetic compass
INS (inertial platform,
strapdown system) 3
(pitch, roll, yaw) 3
(lat., long., altitude)
gyro,
accelerometer
GPS receiver,
radar, barometer
Table 1: Motional Degrees of Freedom considered in traditional integrated navigation systems.
3. DEVELOPMENT OF INERTIAL NAVIGATION SYSTEMS
In navigation, each vehicle is traditionally being regarded as a solitary rigid body. Assuming the
general case of a free spatial motion, it has consequently three rotational and three translational de-
grees of freedom. An aided INS is therefore the most extensive integrated navigation system (Table 1).
Historically, inertial navigation systems emerged during the Second World War, when the flight
control for the launch vehicle A4 (the missile V2) required to consider the rocket motion in full
(Wrigley, 1977): The determination of the attitude and heading by means of gyros (Table 1) had to
be completed by calculating the vehicle velocity and position through numerically integrating ac-
celerometer signals.
In older, early INS, all sensors are attached to a gimballed platform, which is typically kept hori-
zontal and kept orientated to north during the flight. The attitude angles of a vehicle moving
“around” the platform correspond therefore to the gimbal angles. As the classical platform align-
ment is based furthermore on mechanical gyros, it is possible to interpret the gimbals as the output
unit of a mechanical computer determining now the complete attitude by suitably integrating the
128 Wagner
angular velocity detected through the gyros.
Showing suspensions with one or two gimbals, the gyros of the early inertial navigation systems
resemble directional gyros. Considering also the gimbals of the platform itself, these mechanical
systems are considerably subject to environmental influences and wearout effects. Although well
proven, they require a high level of maintenance; they are costly and have a high weight. Optical
gyroscopes (laser gyros, fibre optical gyros) with nearly no moving parts represent here an advanta-
geous alternative (Lawrence, 1993). Their use spread about three decades ago. It was accompanied
by the gradual transition from gimballed platforms to strapdown systems: Omitting any articulated
suspension, this newer variant of inertial navigation systems has the essential feature that all inertial
sensors (gyros and accelerometers) are attached directly to the vehicle. In parallel, the position, the
velocity, and the attitude of the vehicle are determined now all by numerically integrating a set of
ordinary differential equations. This procedure can be directly attributed to the lower left block of
Figure 4. Here, the input u is composed of the angular rate vector and the acceleration vector of the
vehicle as measured by (at least) three rate gyros and three accelerometers. (These six sensors form
the inertial measurement unit IMU of the INS.)
Mathematically, the numerical treatment of ),
ˆ
(
ˆuxfx =
& is the solution of an initial value problem.
Independently from using a mechanical platform or a digital computer, it includes inevitably the
generation of numerical errors, which increase steadily with time. Therefore, an inertial navigation
system can only be operated for a restricted period of time unless an additional procedure limits the
long-term errors to a reasonable size. This is the task of the aiding.
Interestingly, the historical development of aided inertial systems did not refer to Figure 4 directly.
The functional separation between stabilized platforms and early on-board computers as well as the
limited digital precision of early microprocessors lead intuitively to a segmentation of the two
model blocks in the middle. Figure 5 shows this partition. The upper two model blocks contain the
computation of the main part of x
ˆ and y
ˆ, i.e. the more significant digits. The lower two blocks
calculate the less significant digits using basically the Jacobians F and H of f and h. As usual done,
these second parts of x
ˆ and y
ˆ can be interpreted at the same time as the errors x
ˆ
δ and y
ˆ
δ caused
by the limited precision of the upper model blocks and of the sensors measuring u. In case the
vector function f is nonlinear, as usual in navigation, the avoidance of linearization effects requires
from time to time to an additional update of x
ˆ by x
ˆ
δ (dotted line in Figure 5).
+
+
aiding
sub-system
vehicle
input vector u motion state xaiding vector y
u
–
–
estimate
u
xuxFx ˆ
δ),
ˆ
(
.ˆ
δ=
y
ˆ
x
ˆ
estimate
),
ˆ
(
ˆ
model aiding
uxhy =
xuxHy ˆ
δ),
ˆ
(
ˆ
δ
aiding
of modelerror
=
x
ˆ
δ
estimate
)
ˆ
δ
ˆ
(
ˆ
δ
oncompensati
yyyK
x
−−=
Δ
y
ˆ
δ
estimate
),
ˆ
(
ˆuxfx =
&
model of vehicle
kinematics
error model of
vehicle kinemat.
Figure 5: Modified observer principle employed classically for aiding inertial navigation systems.
Considering INS technology during the last two decades, three trends are noteworthy in particular:
• increasingly accurate aiding by satellite navigation receivers (especially GPS devices),
• development of low-cost microelectromechanical gyros and accelerometers (MEMS technol-
ogy) with moderate accuracy (Maluf, 2000),
• availability of very powerful microprocessors.
Wagner 129
Besides supporting the first and the second trend, the last one promotes specially the system inte-
gration and allows in particular using the simpler structure of Figure 4 instead of Figure 5. This is
important as the utilization of low-cost sensors causes a raised overall error level questioning the
linearization approach of Figure 5. Figure 4 offers here an alternative. It is directly designed to han-
dle nonlinear functions f and h if for example the theory of the Extended Kalman Filter provides the
matrix K. Indeed, the evaluation of flight tests with a MEMS IMU (see section 4 below) shows im-
proved system stability for the simpler system structure compared to the classical one.
The first point, the good aiding accuracy of GPS, favours also the use of low-cost inertial sensors
because it is especially able to dampen the mentioned increase of the system error level. This is the
reason why during the last years low-cost systems based on a MEMS IMU and a satellite navigation
receiver became more and more interesting for demanding navigation applications.
4. INTEGRATING INERTIAL AND SATELLITE NAVIGATION
Integrated navigation systems based on inertial sensors and GPS combine normally the strapdown
technology with a one-antenna satellite navigation receiver. Due to their importance obtained dur-
ing the last years, there are a number of books describing the details of the system design (e.g. Far-
rell & Barth, 1999; Jekely, 2001). These publications reflect that customary equipment is still based
on the classical structure of Figure 5. Moreover, they show that the schemes of Figure 4 and 5 allow
a certain freedom in defining u, x, and y. Wagner (2003) has therefore written a survey comparing
several system variants. Concerning accuracy, system stability, and computational effort, one de-
sign proved to be especially favourable (see also Figure 6 below). It utilizes, as mentioned above,
the simpler structure of Figure 4 as well as the following composition of the vectors u, x, and y:
element) (1
element)(1
element) (1
element) (1
satellite 2 rate, range pseudo
satellite 2 range, pseudo
satellite 1 rate, range pseudo
satellite 1 range, pseudo
elements) (2
elements) (3
elements) (3
system)coordinatefixedearthcentered,earth.,comp(4
system)coordinatefixedearthcentered,earth.,comp(3
system)coordinatefixedearthcentered,earth.,comp(3
drift and biasclock receiver GPS
tersaccelerome all of biases
gyrosallofbiases
quaternion
attitude
ectorvelocity v
ectorposition v
system)coordinatefixedvehicle,components(3
system)coordinatefixedvehicle,components(3
vectorrateangularvehicle
vectoronaccelerativehicle
4
nd 3
nd 2
st 1
st
⎥
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎢
⎣
⎡
=
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
⎡
=
⎥
⎦
⎤
⎢
⎣
⎡
=
MM
y
y
y
y
c
a
q
v
p
b
b
y
x
x
x
x
x
x
x
ω
a
u
--
--
--
-
-
ω
As the associated functions f and h are repeatedly documented (e.g. Wagner, 2003), it is not essen-
tial to repeat them here. Nevertheless, two points concerning h are noteworthy. The first one relates
to the location of the GPS receiver antenna relative to the IMU. Antenna and IMU are two separate
components. They cannot be attached to the vehicle at exactly the same point, and the lever arm
between them must be included into h. The second point deals also with this position difference.
However, it pertains primarily to the stability of the feedback loop of in the lower part of Figure 4
and 5. To get a reliable system performance, it is indispensable that the correction output of the
compensation block affects all elements of x
ˆ (or x
ˆ
δ respectively). This necessary property is firstly
determined by the information introduced through the aiding into the systems. (Especially, it is a
well-known property that at least the range measurements for four GPS satellites are required.) On
130 Wagner
the other hand, it depends also on )
ˆ
(xf causing a time-variable system stability. If in particular a
single GPS antenna is used for aiding (Hong et al., 2000), the system stability deteriorates signifi-
cantly when the vehicle is at rest or moves uniformly. Unfortunately, such conditions occur often on
critical occasions like observation periods in airborne photogrammetry or automatic landings in
flight guidance. For this, Figure 6 gives an impression. It shows the estimation error variance of the
attitude of an aircraft performing a test flight with two approaches for landing and with an aero-
drome circuit between them. Being representative for a system destabilisation, the variance in-
creases strongly in phases of straight flights and stabilizes during turns. (Wagner (2003) gives addi-
tional details about this flight test and the equipment used.) In addition, the diagram illustrates also
the better performance of the system structure of Figure 4.
0
3*10-4
6*10-
4
462500 462750 463000
estimation error covariance [rad2]
GPS time [s]
system structure of Figure 4
system structure of Figure 5
phases of 1st and 2n
d
approach
phase of a constant straight flight
Figure 6: Total attitude estimation error covariance during a flight test.
5. AIDING WITH DISTRIBUTED GPS ANTENNAS
An increasing covariance as illustrated in Figure 6 does not lead necessarily to an extensive reduc-
tion of estimation accuracy. Rather, the problem is a reduced stability margin of the feedback loop
in Figure 4 and 5, which affects the system reliability as well as the aptitude to handle temporary
errors and outages of the sensor signals. A sustainable solution for this inconvenience can be pro-
vided by a multi antenna GPS receiver. To exemplify the last statement, Figure 7 shows three IMU-
antenna configurations used for a simulated flight test (Wagner, 2003). The ground path of the
flight is given in Figure 8. The IMU properties assumed corresponded to a MEMS unit.
60°
antenna
IMU
|l| =1m
No. 1-1 No. 3-1
|l| = 1m
No. 3-25
|l| = 25m
60°
direction
of flight
Figure 7: IMU-antenna configurations of a simulated flight test.
Taking a flight time of 30 minutes, the flight track was passed once. Near the crossing point in the
centre of Figure 8, the simulated aircraft experienced phases of comparatively uniform motion.
These took place at the beginning, in the middle, and the end of the flight. During those periods, the
Wagner 131
estimation error variance of the attitude showed again a significantly high level. Corresponding to
Figure 6, Figure 9 reveals clearly this effect. However, it contains furthermore two diagrams, the
left one for an aiding of an accuracy of 1 m (corresponding to differential GPS), and the right one
for an aiding of an accuracy of 4 cm (GPS carrier phase measurements). Interestingly, the rise of the
covariance does disappear neither if the aiding accuracy is improved nor if a small multi antenna
array is employed. The only way to remove this undesired effect is the use of a widely distributed
multi antenna array. On the other hand, this approach causes problems due to vehicle flexibilities.
0
1*10-4
2*10-4
0 500 1000 1500
time [s]
1-1
3-1
3-25
estimation error covariance [rad2]
0
2*10-5
4*10-5
0500 1000 1500
time [s]
3-25
1-1
3-1
estimation error covariance [rad2]
Figure 9: Total attitude estimation error covariance during a simulated flight test;
left: aiding accuracy of 1 m; right: aiding accuracy of 0.04 m.
The classical approach to circumvent the influence of vehicle distortions is to attach all navigation
sensor elements close together on a relatively rigid part of the structure. The increasing use of multi
antenna systems has nevertheless initiated to consider also relative movements of the aiding com-
ponents. Besides filtering techniques, which only average structural vibrations, pure GPS proce-
dures exist, which consider explicitly flexibilities, but their sample rate and real time capability is
limited. (Wagner (2003) has compiled a corresponding literature survey.) It is therefore better to
detect time-varying shapes of vehicles also with inertial sensors by distributing all signal sources
explicitly over the structure. To this, Stieler (1999) published a first approach for the corresponding
problem regarding robots, which was generalized for arbitrary multibody systems by Wagner
(2004). An equivalent theory exists also for elastically deformed vehicles as outlined in brief.
Forming the starting point, Figure 10 shows a fuselage with a distorted wing half attached; the
original wing shape is indicated by the dashed line. An aircraft-fixed coordinate system serves for
describing the time-variant structural geometry. An IMU with three gyros and three accelerometers
is in the origin of the coordinate system. Distributed over the structure, there are satellite navigation
antennas at positions j. Due to structural deformations, these elements undergo time-variant dis-
-10000
0
10000
-20000 020000
η [m]
ξ
[m]
start and stop
Figure 8: Ground path of a simulated test flight.
132 Wagner
placements Δrj (Figure 10). Each vector rj΄ forming now an antenna lever arm consists accordingly
of the constant original part rj and the additional part Δrj. To describe the latter one, it is helpful to
employ an approximation by the main vibration modes of the vehicle structure. In detail, every
mode κ can be expressed individually by a spatial deformation function sκ(r) being modulated by
the time-variant “amplitude” bκ(t):
.)()(
∑
+≈Δ+=
′
κ
κκ
jjjjj tb rsrrrr
The modal functions sκ result from a vibration analysis of the structure and shall be taken for being
known. The amplitudes bκ inhere the property of additional, elastic degrees of freedom. They ex-
tend the motion state vector x and have to be estimated by the Kalman filter, too. However, as a
single IMU is only able to measure the filter input u for the motional degrees of freedom of a rigid
body, extra inertial sensors become necessary: Additional accelerometers and gyros have to be dis-
tributed over the vehicle at certain positions j. They are subject to the local acceleration vector aj
and the local angular rate vector ωj (ab and ωb are the vectors measured by the IMU, see section 4):
.)(curl)(
2
1
,)()( )()(with2)( ,
∑
∑
∑
+≈
≈
′
≈
′′
+
′
×+
′
×+
′
××+=
κ
κκ
κ
κκ
κ
κκ
jbj
jjjjjjbjbjbbbj
tb
tbtb
rsωω
rsrrsrrrωrωrωωaa
&
&&
&&
&
&&&&&
Corresponding to their measurement axes, the sensors detect components of aj and ωj respectively,
which in turn reflect
κ
b
& and
κ
b
&& . Assuming now a sufficient number of aptly attached inertial sen-
sors, an ordinary differential equation for each bκ can be derived from the last relations. They are of
the types
∑
∑
−=
−=
j
jbjj
j
jjbbbjj
gb
bbgb
))(curl,(
or),),(,,,,(
rsωω
rsrωωaa
κκ
κκκκ
&
&
&
&&
with appropriate functions j
g
and j
g
. The differential equations altogether form a set completing
the vector function f (Figure 4) for the additional degrees of freedom.
The expansion of the aiding function h with respect to all bκ is simple. Instead of using rigid an-
tenna lever arms, the formula for rj΄ has to be employed. Concerning h, it has furthermore to be
mentioned that deformations of vehicle structures can also be measured directly using e.g. strain
gauges. These additional, simple sensors offer therefore extra aiding possibilities and can help to
reduce the number of GPS antennas required. (Further details about this theory as well as an
example based on a simulated flight test are to be found in a thesis of the author (Wagner, 2003).)
η
Δr
j
IMU
ζ
r
j
r
j
΄
Figure 10: Fuselage cross-section with distorted wing half and peripheral sensors.
Wagner 133
6. CONCLUSIONS
Integrated navigation systems based on inertial sensors show a considerable variety reaching from
simple instruments like the directional gyro with magnetic aiding to extensive motion measurement
networks for flexible vehicles. Furthermore, this diversity suggests
• that there is still a significant development potential for aided inertial navigation systems and
• that the technological basis concerning sensors and signal fusion may also vary in the future.
Nevertheless, integrated systems with inertial components as a central element show some general
properties that determine their applicability. These characteristics shall finally be ordered as fol-
lows:
• Strengths:
- high resolution with respect to time,
- good short-term and good long-term accuracy,
- no jamming of inertial measurements,
- identification and elimination of poor aiding measurements,
- adaptable level of accuracy.
• Weaknesses:
- high complexity with respect to the sensor equipment and the mathematical algorithms,
- insufficient aiding (e.g. only one GPS antenna) leads to system instability.
• Opportunities:
- extremely flexible and powerful navigation systems,
- potential for applications with a high level of automation/autonomy,
- considerable variety regarding new applications.
• Threats:
- high-level development work necessary,
- low-cost systems require a mass market.
Increasingly powerful microprocessors and software tools have the potential to reduce the develop-
ment costs and development risk for integrated navigation systems. The long-term decrease of the
price/performance ratio for inertial and aiding sensors allows furthermore lowering the production
costs. Therefore, there is a realistic chance that integrated navigation systems with inertial sensors
will become less exclusive but much more common than today.
7. REFERENCES
Bohnenberger, J.G.F. (1817): Beschreibung einer Maschine zur Erläuterung der Geseze der Umdre-
hung der Erde um ihre Axe, und der Veränderung der Lage der letzteren. Tübinger Blätter für
Naturwissenschaften und Arzneikunde, Vol. 3, pp. 72-83
Broelmann, J. (2002): Intuition und Wissenschaft in der Kreiseltechnik. München: Deutsches Mu-
seum
Farrell, J.A. and Barth, M. (1999): The Global Positioning System and inertial navigation. New
York/...: McGraw-Hill
Foucault, L. (1852): Sur les phénomènes d’orientation des corps tournants entraînés par un axe fixe
à la surfaces de la terre. Comptes rendus hebdomadaires des séances de l’Académie des Scien-
ces, Vol. 35, pp. 424-427
Gelb, A. (Ed.) (1989): Applied optimal estimation (11th printing). Cambridge, MA, London: The
M.I.T. Press
134 Wagner
Hong, S., Lee, M.-H., Rios, J. and Speyer, J.L. (2000): Observability analysis of GPS aided INS. In:
Proceedings of ION GPS 2000. Alexandria, VA: The Institute of Navigation, pp. 2618-2624
Jekeli, C. (2001): Inertial navigation systems with geodetic applications. Berlin, New York: Walter
de Gruyter
Lawrence, A. (1993): Modern inertial technology. New York/...: Springer
Levine, W. (Ed.) (1996). The control handbook. Boca Raton, FL: CRC Press
Magnus, K. (1971): Kreisel, Theorie und Anwendungen. Berlin/…: Springer
Maluf, N. (2000): An introduction to microelectromechanical systems engineering. Boston, MA/…:
Artech House
Poisson, D. (1813): Mémoire sur un cas particulier du mouvement de rotation des corps pesans.
Journal de l'École Polytechnique, Vol. 9, pp. 247-262
Sorg, H.W. (1976): From Serson to Draper – two centuries of gyroscopic development. Navigation,
Vol. 23, pp. 313-324
Stieler, B. (1999): Gyros for Robots. In: Sorg, H. (Ed.): Symposium Gyro Technology 1999. Stutt-
gart: Universität / Deutsche Gesellschaft für Ortung und Navigation, pp. 9.1-9.13
Wagner, J. (2003): Zur Verallgemeinerung integrierter Navigationssysteme auf räumlich verteilte
Sensoren und flexible Fahrzeugstrukturen. Düsseldorf: VDI-Verlag
Wagner, J. (2004): Adapting the principle of integrated navigation systems to measuring the motion
of rigid multibody systems. Multibody System Dynamics, Vol. 11, pp. 87-110
Wrigley, W. (1977): History of inertial navigation. Navigation, Vol. 24, pp. 1-6
