You are on page 1of 10

DGPS/INS/VVheelSensor Integration for High

Accuracy Land-Vehicle Positioning


Yukihiro Kubo, Tsuyoshi Kindo, Akihiko Ito and Sueo Sugimoto,
Department of Electrical and Electronic Engineering,
Ritsumeikan University

substitute for DGPS, thus the wheel velocity is used as the


measurement corresponding to DGPS/INS mode. When the
INS/Wheel Sensor mode is utilized, the accuracy of wheel
velocity is important. The estimation problem of the error in
wheel sensors scale factor is also discussed,

BIOGRAPHY
Yukihiro Kubo received the B.S. and M.S. degrees in electrical and electronic engineering from Ritsumeikan University
in 1997 and 1999 respectively. He is a Ph.D. student at Ritsumeikan University, Shiga, Japan.

Finally, the results by using an experimental car for a test


run course is shown. Throughout the experiments, DGPS
correction information provided by the civil DGPS service
in Japan was obtained on telecommunications link with a
cellular phone.

Tsuyoshi Kindo and Akihiko Ito received the B.S. degree in


electrical and electronic engineering from Ritsumeikan University in 1998 and 1999 respectively. They are now studying for the master degree.
Sueo Sugimoto received the B.S. (1969), M.S. (197 1) degree
form Kyoto Inst. of Technology and Ph.D. degree from Polytechnic Institute of New York (presently, Polytechnic University), New York, in 1974. He is Professor, Department of
Electrical and Electronic Engineering at Ritsumeikan University.

With the increasing demand for the mobile surveying for


road profile and geographic information systems (GIS) applications, some integrated GPS/INS (Inertial Navigation
System) navigation systems for land-vehicles have been investigated [1-4]. In those reports, the positioning accuracy
is obtained by merging the precise DGPS information based
on the carrier phase measurement with INS via the Kalman
filter.

ABSTRACT
This paper describes the study on land-vehicle positioning
by DGPS/INS/Wheel Sensor integration. Our purpose is to
integrate the advantages of these systems and to develop a
high precision positioning system that can determine vehicles trajectory and provide the road profile,

Needless to say, the carrier-based DGPS is more complex


than the code-based DGPS and standalone GPS. Moreover
the GPS positioning suffers from various problems such that
mountains or many structures in an urban area shade the
satellites. In those areas, the interference such as cycle slip
and multipath may occur. In these cases the accuracy of the
GPS/INS system will be degraded. This is why the independent wheel sensor is added to the GPS/INS system in this
work.

In this work, we consider two navigation modes such as


DGPS/INS mode and INS/Wheel Sensor mode. When the
DGPS signals are available, thus DGPS/INS mode, the accurate DGPS solutions are used as measurements in the
Kalman filter to calibrate the wheel sensor and to estimate
inertial errors (INS errors in position, velocity and attitude)
and instruments (accelerometers and gyros) biases or drifts.
Then the estimated errors are fed back to INS solutions (position, velocity and attitude) and measured values (acceleration and angular rate).

On the other hand, recently, it has been very easy to implement the code-based DGPS with a low cost receiver and
convenient communications link, In this work, therefore,
we consider two integrated navigation modes by applying
the Kalman filter such as code-based DGPS/INS mode and
INS/Wheel Sensor mode . And the performance of positioning by using the both navigation modes is also studied.

If it becomes that DGPS is not available in an urban


area or near mountain, the navigation mode is switched to
INS/Wheel Sensor mode. The independent wheel sensor is
ION GPS '99, 14-17 September 1999, Nashville, TN

_..

--

INTRODUCTION

555

-~

orthogonal triad accelerometers and gyroscopes, or simply


gyros. Thus the INS information such as position, velocity
and attitude, which is given by integration of acceleration
and angular rate, is very accurate for short time intervals.
However, there are growing errors in INS information due
to the slowly changing errors of accelerometers and gyros.
In order to obtain accurate positioning, those errors must be
removed by using external measurement with long-term accuracy.

The navigation mode is selected appropriately for the situations. Namely, when the DGPS signals are available,
DGPS/INS mode is utilized. And the navigation mode is
switched to INS/Wheel Sensor mode during the periods that
DGPS is not available.
When INS/Wheel Sensor mode is utilized, there are some
problems that must be considered. The wheel sensor (equivalent to odometer in this paper) measures revolution of the
wheel. Then its measurement is multiplied by the scale factor that converts wheel revolution to velocity. Although the
wheel sensor is autonomous, it can become unreliable for
various reasons such as wheel slipping, skidding and error
in the scale factor [3,5]. In order to keep the positioning accurate, in INS/Wheel Sensor mode, the short-term stability
of INS is utilized to detect the wheel slips etc. Thus the estimation problem of the error in wheel sensors scale factor
is aIso discussed.

CONCEPTION

DGPS is the suitable method for this purpose. When the


DGPS signals are available, DGPS/INS mode is utilized.
Namely, the accurate DGPS position and velocity are used
as measurement in the Kalman filter to estimate the INS errors. To say more exactly the measurements are difference
between DGPS position and INS position and difference between DGPS velocity and INS velocity. The INS errors to
be estimated are the errors in the INS solutions (position, velocity and attitude) and the errors in the accelerometers and
gyros.

OF THE SYSTEM
Usually wheel sensors scale factor that converts wheel revolution to velocity (or distance traveled in unit time) is calibrated at the first installation or after changing the wheel
size by driving a known distance. In DGPS/INS mode, the
calibration of the scale factor is also implemented by the
comparison of wheel sensor velocity and DGPS velocity. In
this scheme the Kalman filter is also used to estimate the
scale factor error.

The conception of DGPS/INS/Wheel Sensor integration that


we adopt in this work falls into the category that is well
known as the closed loop mechanization. Fig. 1 shows the
architecture of the integration with chief data paths between
the system components. The components of the system are
strapdown INS, DGPS receiver and wheel sensor. In this architecture there are two navigation modes, DGPS/INS mode
and INS/Wheel Sensor mode.
2.1

2.2

DGPWINS Mode

INS/Wheel Sensor Mode

The main drawback of DGPS positioning is the degradation of the accuracy due to time-lag for the vehicles motion,
satellite obstruction, poor satellite geometry and

The INS is able to measure vehicles accelerations and angular rates precisely with very short time delay by an

Strapdown
~

rrw

m
Errors

Position -1

J
I

Fig. 1: Integration architecture

556

..

....

Position
Velocity
Attitude

i) The E frame (XE, YE, ZE) is the right-handed earth


fixed coordinate frame. It has origin at the center of
the earth; the ZE-axis is directed toward the North
Pole; the XE- and Y~-axes are located in the equatorial plane, whereby the XE-axis is directed toward the
Greenwich Meridian. It is used for the definition of position location such as latitude and longitude.

disconnection of the communications link. These situations


are usual in an urban area, under the highway overpass, in
tunnels, and nearby mountains. If it becomes that the DGPS
signals are not available, the other information is required to
keep the positioning accurate.
The wheel sensor is a natural choice to obtain velocity information. During the period that the DGPS accuracy is
degraded or there is no DGPS information, therefore, the
navigation mode is switched to INS/Wheel Sensor mode.
The autonomous wheel sensor is substitute for DGPS, thus
the wheel sensor velocity is used as the measurement corresponding to DGPS.

ii) The L frame (XL, YL, ZL) is the right-handed locally


level coordinate frame. The XL- and YL-axis is directed toward local north and east respectively; ZL -axis
is downward vertical at the local earth surface referenced position location. It is used for defining the angular orientation of the local vertical in the E frame.

When INS/Wheel Sensor mode is utilized, there are some


problems that must be considered. Namely, the accuracy of
wheel sensor velocity is degraded by wheel slips and changing in the wheel radius due to wearing down of tires and
variations in tire pressure etc. Various error sources in wheel
sensor measurements are discussed in [6]. Moreover, since
the direction of the tire is not always parallel to the direction
of the body by a small angle due to a shake of the body and
steering, the difference between them also degrades the accuracy when the wheel sensor velocity and the INS velocity
are compared. In INS/Wheel Sensor mode, the short-term
stability of INS is also utilized to detect these errors. It results in an integrated navigation system with performance
superior to either one acting alone such as only INS or simply integrating the wheel velocity (so-called dead reckoning
system).

iii) The C frame (Xc, Yc, Zc) is the right-handed computer frame that is defined by rotating the L frame
about negative ZL-axis through the wander angle a,
the Ye-axis is directed toward the negative YL-axis and
the Zc-axis is directed toward the negative ZL-axis
(upward vertical). It is used for integrating acceleration
into velocity, and used as the reference for describing
the strapdown sensor coordinate frame orientation.
iv) The 1? (XE3,YB, ZB ) frame is the strapdown inertial
sensor coordinate frame (body frame). The XB -axis is
directed toward the head of the vehicle; the YB-axis is
the right-hand of the vehicle; the ZB-axis is downward
vertical about the XB-YB plane. The frame is fixed on
the vehicle and rotates with the motion of the vehicle
(Fig. 3).

COORDINATE SYSTEMS

To integrate the navigation systems, it is important to consider the coordinate systems that the navigation systems or
included sensors refer to, This section defines the coordinate
frames used in this paper and represents the angular relationship between them [7, 8].

Jy

The coordinate frames are defined as follows:

Wheel

sensor

gyroscope

YB

accelerometer

ZB

Fig. 3: The B frame and sensors

x ,~
North

Pole

Fig. 2 shows the spatial image of the E, L and C frames,


where A represents the geodetic longitude and p represents
the latitude.

Ecruator

ial

In inertial computations, three components of acceleration


sensed with respect to the B frame have to be transformed
onto the C frame. The velocity and position of the vehicle are then computed with respect to the C frame. Such a
transformation is known as an Euler angle transformation,
A similar transformation will take us from the B frame to C
frame by a series of four rotations.

olane

XE

Fig. 2: The E, L and C frames

557

.
_.

...

Beginning with the B frame, the first rotation is about the


negative XB -axis through the roll angle @, then the new
coordinate axes (X:, Y:, Z~ ) is derived, whereas the X~axis still remains coincident with the XB -axis. The second
rotation is about negative Y&-axis through the pitch angle (I,
then new (X;, Y;, Z;) frame is derived. The third rotation
is about negative ZB-axis through the yaw angle W, then
new (XB, Y;, Z; ) frame is derived. The fourth and final
rotation is about negative X; -axis through 180 deg. We
define the product of direction cosine matrices for a chain
of these rotations as T~. Then the coordinates (zE, y~, .z~)
in the B frame are transformed into (zC, g~, ZC)inthe C
frame as follows:

Table 1: List of states


_

No.
1
2
3
4
5
6
7
8
9
1o*
11*
12*
13
14
15
16
17
_ 18

E1=TW!

(1)

The angles Q, (3 and W used in this transformation are generated by integrating the gyroscopic angular rates, but we do
not pursue this subject here.

Xc-axis error in relative position


Ye-axis error in relative position
Xc-axis velocity error
Ye-axis velocity error
Xc-axis tilt error
YC-axis tilt error
ZC-axis tile error
Zc-axis altitude error
ZC-axis velocity error
ZB-axis wheel sensor azimuth angle
wheel sensor scale factor variation
YB-axis wheel sensor pitch angle
XB-axis accelerometer bias
YB-axis accelerometer bias
ZB-axis accelerometer bias
XB-axis gyro bias
YB-~is gyro bias
ZB-axis gyro bias

* appearonly in INS/Wheel Sensormode

STATE MODEL

The first step towards implementing the navigation filter is to


define a system model which describes adequately the system dynamics. Using the four coordinate systems defined
in section 3, we can now develop a set of linear differential
equations describing the propagation of INS errors for the
locally level C frame navigation equations.

processes, generally written as


i(t) =

+) +w(t),

(5)

where T is the correlation time constant and tu(t ) is zero


mean Gaussian white noise process [12].

The states of the system to be estimated are thus described


by the vector z as follows:

By discretizing the state equation (3), we get


x(k + 1) = A(k)z(k) +w(k),
Ea E.q Ep

bz by bz dz dy

dz

(6)

(2)

where

where e., E. and eP appear only in INS/Wheel Sensor mode


and the description of the components are listed in Table 1.

z(k) = [&X(kAt), Jr-Z(kAt),. . . , dz(kAt)]T,

(7)

A(k) = eAIkAtJAt
The following dynamic model is then utilized
~(t)= A(t)z(t)

+ q(t),

(3)

E[w(k) wT(j)] = Q(k)L$M,

(9)

w(k) =

(4)

(k+l)Ai
~{(~+l)At&}q(f)

df.
(lo)

kAt

The INS error model in the state equation (3), namely, the
equations for the states excepting Nos. 10-12 components
of z and also the related components of A(t) are derived
in, for example, [91 1]. For the components of Nos. 1012, in this work, they are modeled as the first order Markov

4.1

Measurement Equation in DGPWINS Mode

In DGPS/INS mode, the measurements are the differences


between DGPS and INS positions, and velocities as well.

558

..

(8)

and At is a sampling interval for updating the measurement


data. The system noises w(k) in (6) and q(t) in (3) are
related by

where the dot above a letter denotes differentiation with respect to time, A(t) is the dynamics matrix and the process
noise q(t) is assumed as mutually uncorrelated zero mean
Gaussian white noise with covariance matrix IV:
E[q(t)?-/T(t)]
= N(t).

R 118+ A(kAt)At,

4.2

By the definition of drz, Jrv and Jh, the INS-indicated positions at time k are modeled as

MeasurementEquationinINS/WheelSensorMode

The output of the wheel sensor is modeled as


?-;(k) = T-:(k) + J?-z(k)
hi(k) = ht(k) + 6h(k),

where T is the radius of the wheel (scale factor), w is the


angular rate of the wheel revolution, n is the measurement
noise, and the superscript m means measured value by the
wheel sensor [13, 14].

where the superscripts z and t mean inertial and true respectively. The DGPS position indications aremodeled as the
true positions corrupted by mutually uncorrelated Gaussian
white noises {n: (k) } with zero mean and covariance matrix
R9(k):

On theotherhand, thewheel radius slowly changesbecause


of wearing down of tires, and variations in tire pressure,
weight of the body as describedin section2. And there are
wheel slips that depend on vehicles speed and road situations such as wet or dry. Weassume that the slow changes
in wheel radius during a run is small, and the slips for the
stationmy speed are small and are in proportion to vehicles
speed. With above assumptions, they can be regarded as
changes in the wheel radius. Thus, the true velocity t? can
be written by

r:(k) = r;(k) n;(k)


(12)

~;(k) = rj(k) @(k)


W(k) = ht(k) nj(k),

(16)

vm=~w+n,

(11)

r$(k) = r;(k) + JrV(k)

where the superscript g means DGPS. Thus, subtracting (12)


from (11) yields the measurement models for position errors
as follows
z!(k) = r;(k) r:(k)

t?=(~+A~)u

= drz + n?(k)

=(l+&.)~L4/,

zl(k) - r)(k) rj(k)


= (hv + n;(k)

(13)

where AT is the change in wheel radius, and we define ES=


AT/~. Then (16) reduces to

zj(k)~hi(k)hg(k)
=dh+n~(k).

1
urn = Vt
l+&.

Similarly to the above derivation, the measurements for velocity errors can be modeled as
z~(k)~v;(k)
=

h.

zg(k)-vj(k)
= (hv

n!(k)

~;(k)
n:(k)

z~(k)-v;(k)
=
Thusthemeasurement
ten as

(h.

(14)

v:(k)
+

n:(k).

equations(l3)and

(18)

(14)canberewritv:

zg(k)=llgz(k)

+ n,.

(18) shows the vehicles velocity in the direction of the tire


measured by the wheel sensor. It thus have to be properly
transformed onto the C frame. However, since the direction
of the tire is not always parallel to the B frame due to a
shake of the body or steering, the transformation T~ cannot
be used directly. Therefore we assume that the direction of
the tire is slightly misaligned from the B frame via small
attitude angles such as Ca about ZB -axis and eP about YBaxis. Then the true velocity in the C frame and the true
wheel velocity are related by

v$(k)
+

(17)

(15)

+ng(k),

T1l

COS&@ . Vt

V; =

T21 COSEa

V:

T31 COS&p - Vt,

. #

(19)

where
zg(k)=[z~(k)

z;(k) . ..z~(k)]T.]T,

rtg(k)=[n;(k)

n;(k) . . . . ..n~(k)]T.

where the subscripts x, y and z represent C frame coordinate axes, and T~j is the element in row i, column j of T$.
Similarly, the output of the wheel sensor Vm is transformed
as follows:

10000000

01000000
H9

00
0
00
1
00010000
00001000

0
0

00
00

0
0

1
0

Vm
z

T1l

V~

T21 COS&.. Vm

V~

T31 CoS&p

6X7

COS&a . Vm

Then from (18), (19), (20), and

559

(20)

. Vm.

the

definition of 6vZ, 6V9

and JvZ, we have the measurement equation as follows:


g(k) = Tllv~(k)
..

..
z:(k)

{l+&..@); cosEa(k)

{l+&?.@); cos&a(k)

= T31W(IC)

{v:(k) -h.(k)}

{V$(k)-dv,(k)}+

+n~(k)

ny(k)

{l+ E.@); cos&~k)

{V;(k) -c$vz(k)}+ny(k),
(21)

where the superscript w means wheel sensor, and we assume


that {n~ (k)} is the zero mean white Gaussian noise processes with covariance matrix Rw (k). Thus the measurement equations (21) can be rewritten as
ZW(k)

h(z(k)) + rtw(k),

Fig. 4 Image of tire size


(Fig. 4), namely, AT = ~ ~t. Thus equation (23) can be
rewritten as

(22)

where
(24)

zw(k) = [.z~(k) ,z~(k) .z~(k)]T,


nw(k) = [n~(k) n;(k)

n~(k)]T,

where a a A~/-yt and Vt is true speed.

and the components of the vector function h are described


in equation (21).

The DGPS speed W is modeled as the true speed Vt corrupted by the measurement noise ftz:

The above measurement equation (22) is nonlinear. Therefore we apply the extended Kalman filter, namely, (22) is
linearized by applying first order Taylor series approximation around the one-step predictor [12].
4.3

v=

(Vg)

vt +ti.

(25)

To calibrate the wheel sensor, it is necessary to estimate ~.


Thus the Kalman filter is formulated to the model of a and
true speed Vt. The state vectors of variables to be estimated
comprises:

Wheel Sensor Calibration with DGPS

The scale factor of the wheel sensor is basically the wheel


radius -y in equation (16). It must be initially calibrated to
remove the deterministic error at the first installation or after changing the car condition such as weight of the body,
tire pressure and size of the tires. The conventional way to
calibrate is driving a known distance.

t?(t) = [v~(t) a(t)]T.

(26)

And Vt and @ are modeled by a first order Markov and a


Brownian motion process respectively. Therefore the state
space representation is

In order to keep the system autonomous, in this work, the


calibration is performed by comparing wheel sensor speed
with DGPS speed, where the scheme is conceptually introduced in [3] for GPS/Dead Reckoning
system. Also
for this calibration purpose, we apply the Kalman filter to a
simple model formulated as follows.

i?(t) = Is(t) + q(t),

(27)

where
F=

Let ~ be the nominal radius of the wheel. Then the wheel


sensor output tim is modeled as
w=~w+fil,

J(vw+(w+
=

.10
T,,
00

[1

(28)

-rVis the correlation time constant and fj(t) is mutually uncorrelated zero mean Gaussian white noise with covariance
matrix lV(t).

(23)

By discretizing the state equation (27), we get

where w is the angular rate of the wheel revolution and ii.1 is


the measurement noise. Now we assume that the difference
between the nominal radius ~ and the true radius Tt is AT

s(k + 1) = rs(k)

560

+ m(k),

(29)

where
35.51

r = eFAt
cx 12 + FAt,
E [m(k)] = O,

35.508 -

E [tIJ(k)t.iJT(j)] = @(k)6/cj.

E
&l
~ 35.506 n
x
u

Also, the measurement equations (24) and (25) can be reformed:

.g

35.504 -

z(k) = Tt(s(k)) + ii(k),

(30)
35.502
t

where

z(k) = [r(k)
n(k) = [ill(k)
E [~(k)] = O,

139.3695

vg(k)]T
,
ii2(k)]T
,
E [@k) fiT(j)]

Longitude(Deg.

EXPERIMENTAL

E)

Fig. 5: The test run course measured by KGPS

= ~(k)dkj ,

and the components of the vector function ~ are described


in (24) and (25). Since the measurement equation (30) is
nonlinear, the extended Kalman filter is utilized.

The DGPS correction information, provided by Nippon Senpakutsushin, Inc. in RTCM SC-104 format, was obtained on
telecommunications link with a cellular phone. The DGPS
reference station is located in Yokohama City where the distance is about 30 km from the test course. Also, the carrierbased kinematic GPS (KGPS) method with approximately
8 km baseline to the surveyed station in Atsugi City was implemented simultaneously. The coordinates of this KGPS
method were regarded as true for evaluating the system performance. Fig. 5 shows the trajectory of the car measured
by KGPS.

RESULTS

The experiment of the DGPS/INS/Wheel Sensor positioning architecture described above were carried out. The measurement data were obtained from DGPS, INS, wheel sensor
mounted in an experimental car.
The navigation system components of the car consists of
Ashtech G-12 DGPS receiver, Tamagawa Seiki TA7421
strapdown INS and the wheel sensor using Tamagawa Seiki
TS5304N510 optical encoder. The INS is within an accuracy of 2 nmi/h (1a), and some sensor parameters of the
INS are briefly shown in Table 2. The optical encoder outputs 2000 pulses per one revolution of the wheel.

Wheel Sensor Calibration Results


Firstly, the conventional calibration method was performed.
Namely, the scale factor was calibrated by mean of five
times, straightforward and 100 m running. We define ~ as
the wheel radius evaluated by this conventional method.
Next, in order to test the method described in section 4.3 for
various nominal values of ~, the calibration was performed
in each situation such that ~ = 0.87, 1.O~, 1.057 and 1.57.
Fig. 6 shows the measured DGPS speed vg in (25) and wheel

A rectangular course round approximately 2.2 km in suburban area of Atsugi City (Kanagawa Prefecture, Japan) was
selected as a test run course. The car run about 4.5 laps for
15 minutes. Throughout the test, the INS data were collected
at 51.2 Hz rate, and the DGPS and the wheel sensor at 1 Hz
rate. Thus the measurement interval for the Kalman filter
update At was 1 sec.

Table 2: Sensor error parameters of INS

200

400

600

Time (sec.)
Random error

I 0.008

Fig. 6: Measured DGPS and wheel sensor speed

deg/fi

561

. .

_..

7=0.8-? ---7=1.0+ ..

Positioning Results

7=1.05+
-7=1.5-7

The positioning tests were performed in four DGPS/INS/


Wheel Sensor based methods such that

06!"
.... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ..~

i)

Stand alone INS;

ii) Integrated DGPS/INS (DGP!YINS mode);


t

600

400

200

iii) Positioning by wheel sensor speed and INS attitude information


(i, e. This method is a kind of dead reckoning);

Time (sec.)
Fig. 7: Estimetes of d (A7/FYt)

iv) Integrated INS/Wheel


mode).

sensor speed W in (24) for the period of 600 sec. The value
of system parameters have been given as

Sensor (INS/Wheel

Sensor

The positioning results are shown in Figs. 811.


rV= 70.

Fig. 8 shows the result of stand alone INS. It can be seen


that the positioning errors are growing (totally about 1000
m for the period of the test). Fig. 9 shows the result of the
DGPS/INS mode. The position errors of the four methods
show that the DGP!YINS mode has the best performance.
Fig. 10 shows the result of position determination by using
wheel sensor speed and INS attitude information, i. e. there
were no INS correction or scale factor correction. There
is about 25 m error in position for 15 minutes run (about
0.25 % of distance traveled). Fig. 11 shows the result of
INS/Wheel Sensor mode. We can see that the performance
in position is degree of slightly degraded DGPS/INS mode.

The estimates of a based on those speed measurements are


shown in Fig. 7. The scale factor calibration can be performed by dividing the nominal radius ~ by 1 + ii(k), where
d(k) is the filtered estimate of a(k).
From Fig. 7, it can be seen that the calibration method described in this paper can exert an almost similar influence to
the conventional method. The results also suggest that, in
practice, the implementation of the method for a period of
3W60 seconds is enough to complete the calibration.

32~

r
,
......................................

-1oo-200 -

I
g

,
...........................................................................................................

Star

35.504

-500 35.502

till

,lJLJ

. I 000 -

II
139.37

139.375

139.38

139

200

Longitude (Deg. E)

,
400

800

600

Time (sec.)

Fig. 8: Positioning by stand alone INS (Trajectory and difference between the results and KGPS)

562

.-

r
35.51

35.508
E
~

35.506
start

i
%
d

35.504

35.502
t

L,

Longitude

139.37 139.371139.372139.3:

200

400
Time

(Deg. E)

600

800

600

800

600

800

(sec.)

Fig. 9: Positioning by DGPS/INS mode (Trajectory and difference between the results and KGPS)

35.51

20 -

10 35.508

start

-8
.fj

35.504

A
-lo -

35.50:

139.37 139.371139.372139.373

-20 0

200

400

Longitude (Deg. E)
Fig. 10 Positioning by wheel speed and INS heading(Trajectory and
difference between the results and KGPS)

10

35.51

35.508
E
-lo start

100
35.502

-10 I

139.37 139.371 139.372139.3;


Longitude

(Deg.

200

400

Time (sec.)

E)

Fig. 11: Positioning by INS/Wheel Sensor mode (Trajectory and difference between the results and KGPS)

563

[4] Global Positioning System: Theory and Applications


U (Chap. 7), ed. by Parkinson, B. W., AlAA, Washington, DC, 1997.

CONCLUSION

Two integrated navigation modes, namely DGPS/INS and


INS/Wheel Sensor mode, have been developed by applying
the Kalman filter.

[5] Abbott, E. and Powell, D., Land-Vehicle Navigation


Using GPS, Proceedings of The IEEE, Vol. 87, No. 1,
January, 1999.

In this system, when the DGPS signals are available, the


long-term stability of DGPS in used to not only correct the
INS but also calibrate the wheel sensor scale factor. And
when the DGPS cannot be utilized, the accuracy of positioning is kept by INS/Wheel Sensor integration.

[6] Zickel, R. and Nehemia, N., GPS Aided Dead Reckoning Navigation, Proceedings of ION National Technical Meeting, pp. 577586, 1994.
[7] Lin, C-F., Modern Navigation, Guidance, and Control
Processing, Vol. II, Prentice Hall, New Jersey, 1991.

The experiments were carried out for each of DGP!YINS,


INS/Wheel Sensor mode and scale factor calibration. To
summarize the results, this system can continuously provide
accurate positioning even while losing DGPS.

[8] Farrell, J. L., Integrated Aircrajl Navigation, Academic Press, San Diego, 1976.
[9] Huddle, J. R., Inertial Navigation System Error-Model
Considerations in Kalman Filtering Applications, in
Control and Dynamic Systems Vol. 38, ed. by Leondes, C. T., pp. 293339, Academic Press, San Diego,
1990.

ACKNOWLEDGEMENT
The present work comes out with the collaboration of
Messrs. H. Kumagai, K. Kokue, and O. Sakamoto, Tarnagawa Seiki Co., Ltd. The authors would like to express their
appreciation to them for their valuable supports.

[10] Leach, B. W., A Kalman Filter Integrated Navigation


Design for The IAR llvin Otter Atmospheric Research
Aircraft, Report No. : NRC-32148; IAR-AN-72; CTN92-60369, Institute for Aerospace Research, Ottawa
(Ontario), Flight Research Lab. April, 1991.
[11] Diesel, J. W., GPS/INS Integration for Civil Aviation,
NTC 91, National Telesystems Conference Proceedings, Vol. 1, 1991.

REFERENCES
[1] Lapucha, D., Schwarz, K. P., Cannon, M. E. and Martel, H., The Use of INS/GPS in a Highway Survey System, Proc. of IEEE PLANS, 1990.

[12] Maybeck, P. S., Stochastic Models, Estimation, and


Control, Vol. 1, Academic Press, San Diego, 1979.
[13] K. Watanabe, Y. Ieiri, Y. Okada and A. Kuramoti, Absolute Speed Estimation of Automobiles from Noisy
Acceleration and Erroneous wheel Speed Measurements, Trans. of SICE, Vol. 26, No. 5, pp. 544-549,
1990 (in Japanese).

[2] EL-Sheimy, N. and Schwarz, K. P., Navigating Urban Areas by VISATA Mobile Mapping System Integrating GPS/INS/Digital Cameras for GIS Applications, Navigation, Journal of The Institute of Navigation, Vol. 45, No. 4, Winter 1998-1999, pp. 275285.

[14] Y. Kubo et al., Recursive Filters and DGPS/INS/Wheel


Sensor Integration, Proc. IMEKOXV World Congress,
Osaka, June, 1999.

[3] Understanding GPS: Principles and Applications,


ed. by Kaplan, E. D., Artech House, Boston, 1996.

564

You might also like