Real-time Navigation, Guidance, and Control

of a UAV using Low-cost Sensors

Jong-Hyuk Kim, Stuart Wishart∗ and Salah Sukkarieh
Australian Centre for Field Robotics
University of Sydney, NSW 2006, Australia
Telephone: +61 (2) 9351 7154
Fax: +61 (2) 9351 7474
Email: {jhkim,salah}@acfr.usyd.edu.au
∗
BAE Systems, Australia
Email: {s.wishart}@acfr.usyd.edu.au

Abstract— Applying low-cost sensors for the Guidance, Navi-

gation and Control (GNC) of an autonomous Uninhibited Aerial
Vehicle (UAV) is an extremely challenging area. This paper
presents the real-time results of applying a low-cost Inertial
Measurement Unit (IMU) and Global Positioning System (GPS)
receiver for the GNC. The INS/GPS navigation loop provides
continuous and reliable navigation solutions to the guidance and
flight control loop for autonomous flight. With additional air data
and engine thrust data, the guidance loop computes the guidance
demands to follow way-point scenarios. The flight control loop
generates actuator signals for the control surfaces and thrust
vector. The whole GNC algorithm was implemented within an
embedded flight control computer. The real-time flight test results
show that the vehicle can perform the autonomous flight reliably Fig. 1. The overall structure of navigation, guidance and control loop in
even under high maneuvering scenarios. UAV.

Index Terms— Autonomous UAV, low cost sensors, Kalman

filter, guidance and control, flight control system.
The Global Positioning System (GPS) can provide long-
I. I NTRODUCTION term stability with high accuracy. It also provides worldwide
coverage in any weather condition. As a result lots of research
Over the recent years the use of low cost Uninhibited
have been done to optimally blend the GPS and INS [6][7][8].
Aerial Vehicle (UAV) for civilian applications has evolved
Since the performance of the low cost GPS receiver can be
from imagination to actual implementation. Systems have been
easily degraded in high maneuvering environments, the quality
designed for fire monitoring, search and rescue, agriculture
and integrity of the GPS system becomes also a crucial factor.
and mining. In order to become successful the cost of these
In case of GPS outage or fault conditions, the stand-alone
systems has to be affordable to the civilian market, and
INS quality then becomes the dominate factor. If the cost is
although the cost/benefit ratio is still high, there have been
a prohibitive factor in developing or buying an IMU, then
significant strides in reducing this, mainly in the form of
improvements in algorithms, and/or fusing the navigation data
platform and sensor cost.
However, reduction in sensor cost also generally brings with other sensors such as a barometer is required.
about a reduction in sensor accuracy and reliability. Coupled In this paper the authors present a low-cost navigation
with the generally high mission dynamics that vehicles un- system which is successfully applied to the GNC of a UAV.
dertake within civilian aerospace due to the restricted mission Figure 1 depicts the overall GNC structure implemented in
areas, ensures that the design and implementation of these the Flight Control System (FCS). In remote operation mode,
sensors is an extremely challenging area. the remote pilot on the ground sends the control signals to
More importantly, the implementation of low cost sensors the actuator via wireless uplink channel. The INS/GPS/Baro
which are used for the Guidance, Navigation and Control navigation loop downlinks the vehicle states to the ground
(GNC) of the aerial vehicle is where most interest lies although station for vehicle state monitoring. When the autonomous
little research is done. When applying a low cost Inertial mode is activated, the navigation solution is fed into the
Measurement Unit (IMU) there are still a number of challenges guidance and control loop and the onboard Flight Mode Switch
which the designer has to face. The main restrictions are the (FMS) redirects the computed control outputs to the actuators.
stability of the Inertial Navigation System (INS) degraded by The INS/GPS/Baro navigation loop makes use of a four-
the inertial sensor drifts. The quality and integrity of aiding sample quaternion algorithm for the attitude update [2]. A
sensors is also the crucial factor for the integrated system. complementary Kalman filter is designed with the errors in po-
sition, velocity and attitude being the filter states. It estimates
the low-frequency errors of the INS by observing the GPS
data with noises. In actual implementation, a U/D factorised
filter is used in order to improve the numerical stability and
computational efficiency [3][13]. Under high maneuverability,
part of the GPS antenna can be blocked from the satellite
signals which cause the receiver to operate in 2D height-fixed
mode, hence to maximise the satellite visibility under these
conditions, a second redundant receiver is installed and used.
The guidance loop generates the guidance commands from
the vehicle states and the desired waypoint information. It
computes required vehicle speed with respect to the air, height
and bank angle. Then the flight control loop (or autopilot)
generates actuator control signals to make the vehicle follow
the guidance demands as well as to stabilise the vehicle. The
control outputs are fed to the control surfaces, or aileron,
elevator and rudder, and thrust vector [4][5].
The whole navigation, guidance and control software pack- Fig. 2. The location of onboard sensors in the Brumby MK-3 UAV. It is
equipped with IMU, GPS, vision camera, radar scanner, air data system, and
age was developed using the C++ class methodology. They flight electronics
are implemented and installed within a FCS which has a
multitasking capability. The developed flight vehicle, MK3-
Brumby, is a delta fixed wing platform with a pusher prop
configuration. It can carry an extra sensor payload for different
mission objectives. These sensors include radar, vision and
laser; and hence the accuracy of the navigation loop is not
only for control purposes but also for the accurate registration
of landmarks which these mission sensors pick up. The real- Fig. 3. The four flight vehicles: MK3-Brumby. Four sets of identical GNC
time flight tests show that the navigation system can provide modules are installed on the vehicles with different mission load.
accurate and reliable 3D navigation solutions as well as to
perform the guidance and control task reliable.
Section II will briefly describe the aircraft system including The ground station consists of a DGPS base station, weather
the flight platform, onboard systems, and ground system. station, hand-held controller, and monitoring computer. Ad-
Section III will provide the details of the low-cost sensors ditional mission computers performs the monitoring of the
used in this work. Section IV will describe the navigation mission objectives.
loop. Section V will detail the structure of the guidance and
control loop. Section VI will present the result of a real- III. L OW- COST S ENSORS
time autonomous flight test. Conclusions are then provided The Inertial Measurement Unit (IMU) is an Inertial Science
in Section VII. ISIS unit. Figure 4 shows the IMU installed in the fuselage
of the vehicle. It is a short-range tactical grade IMU with
II. A IRCRAFT S YSTEMS three voltage resonant gyros and three accelerometers. It has a
The physical aircraft system comprises of the flight plat- gyro bias repeatability within 0.01 ◦/s and accelerometer bias
form, on-board systems, communication links, and ground repeatability within 1mg. The IMU has very small physical
station. dimension and consumes only 5 watts of power. The unit
The flight platform, MK3-Brumby, is shown in figure 2. It can provide inertial data from 100Hz to 400Hz through an
is a delta fixed wing platform with a pusher prop configuration RS422 serial interface. The resolution of the gyro is too low
and is capable of flying at 100kts and up to 500m. The to detect the earth rotation rate and the self-azimuth alignment
platform can carry up to 11kg of additional mission sensors can not be performed for initialisation. In this work, the initial
(the payload sensors are part of the flight platform). Figure 3 heading is uplinked from the ground station. The IMU makes
shows four UAVs built for the demonstration of cooperative use of an internal temperature lookup table for the temperature
data fusion amongst multiple vehicles [12]. compensation. However, in field tests, the turn-on bias was not
The on-board systems consist of an FCS, FMS, vision removed effectively so the calibration process was performed
system, scanning radar and/or laser system, and an air data before starting the navigation. This was performed using a tilt
system. The vision and radar systems are mission specific sensor as shown in the upper right part on figure 4.
nodes and perform multi-target tracking, target registration and The GPS receiver is the BAE Systems Allstar T M receiver.
decentralised data fusion. Two GPS cards are stacked on the FCC as shown in figure 4.
The air-to-air/air-to-ground communication links are estab- Their antennas are installed on each of the wings as shown
lished for the decentralised fusion purposes, remote control in figure 5. The wing frame was made of fiber glass and
operation, differential GPS data uplink, and telemetry data. the GPS signal loss was negligible. The master GPS receiver
provides navigation solutions to the Kalman filter. The slave
receiver provides redundant navigation solution as well as
the raw measurement data for the purpose of GPS attitude
determination which is not implemented yet.

Fig. 6. A view from the rear hatch of the vehicle. The air data system can
be seen next to the fuel tank. It is implemented using a FlexIO card which
can be configured in multi-purpose data sampling.

Fig. 4. A view from the front hatch of the vehicle. A low-cost IMU is
reliable position, velocity and attitude with sufficiently high
installed (lower right) next to the vision camera pointing downwards and a rates. The Kalman filter estimates the navigation errors by
tilt sensor (upper right). Two GPS receivers are stacked on the flight control blending the GPS observation or baro-altimeter data running
computer. The sensors are connected to the flight control computer via serial
and parallel lines.
as a background task.

A. Inertial Navigation
The INS is mechanised in an earth-fixed tangent frame
as shown in figure 7. It computes position, velocity and
attitude of the vehicle with respect to the reference frame by
numerical integration of the accelerations and angular rates.
In this mechanisation scheme, the reference frame is assumed
as a non-rotating inertial frame. The short mission flight time
and the frequent GPS corrections makes this assumption valid
without significant performance degradations in most of the
local terrestrial navigators. If the INS should perform long-
range missions without GPS corrections, the INS will require
a more precise mechanisation scheme to remove systematic
errors like frame rotation effect and coriolis force. In this earth-
fixed tangent frame mechanisation, the coriolis and transport
Fig. 5. A view of the GPS antenna installed inside the wing of the vehicle. rates term are not calculated. The navigation outputs are
Two antennas are installed on both of the wings to provide redundancy. in MGA (Map Grid Australia) coordinate format instead of
WGS-84 coordinate as it is convenient to exchange the vehicle
The air data system is implemented by using a F lexIO T M
states and relative target observations between multiple UAVs.
card developed by the University of Sydney as shown in figure
6. It is a multi-purpose A/D (Analog to Digital) and D/A
(Digital to Analog) card which can sample various analog
signals such as air pressure, temperature, engine RPM, battery
current and voltage. The sampled data are sent to the FCS via
RS232 serial interface and FCS downlinks them to the ground
station for vehicle monitoring. The measured air data is used
in the guidance and control loop and the baro-altimeter data
is used to stabilise the vertical axis of the INS.

IV. NAVIGATION L OOP

Fig. 7. The INS mechanisation in the earth-fixed tnagent frame.
The navigation loop plays a key role in aircraft system. Its
navigation outputs are used in guidance and control and affect
the performance of target registration and picture compilation
tasks. It also has to provide precise timing synchronisation to B. Fusion Kalman Filter
other sensor nodes. The fusion Kalman filter is the heart of the navigation
The core of the navigation loop is the strapdown INS and system. In a low-cost system the IMU errors like bias, scale
the Kalman filter. The strapdown INS provides continuous and factor error and random walk noise dominate the INS error
growth. These INS errors typically show low dynamics and its
models have been well developed [1]. The INS error model in
earth-fixed tangent frame is used in this paper [11].
Figure 8 shows the complementary filter structure that deals
with the INS error state instead of the total vehicle states using
the INS error model. Since the INS error has low frequency
dynamics, the filter can run in relatively low sample rates Fig. 9. The guidance structure.
with lower priority. When the external aiding information from
GPS or baro-altimeter is available, a measurement residual is
generated by subtracting the actual measurement from the INS B. Flight Control
predicted measurement. The resulting residual contains the The flight control loop controls the vehicle’s attitude and
INS error as well as the measurement error. The measurement attitude rates as well as the vehicle speed with respect to the
error typically has high-frequency noise and can be modelled air. The block diagram of the control loop is shown in figure
as white (or broadband white) noise process. The Kalman 10. It performs speed control, height and height rate control,
filter suppresses this high-frequency noise and estimates the bank angle control, heading control, turn compensation and
INS error by the low-pass filtering nature. If the measurement elevation control by using the guidance demands and measured
error contains additional dynamic error, this should be properly vehicle states. It generates actuator signals for the engine
modelled inside the filter [3]. throttle, rudder, elevator, and aileron. In this work, the control
loop generates the control signals every 50Hz which is limited
by the bandwidth of the electric servo actuators. The control
loop is the most time critical task in autonomous flight so it
is allocated to the highest priority inside the FCS. Most of the
guidance and control algorithms were verified and tested using
the Hardware-In-The-Loop (HWIL) simulator which uses the
vehicle model and simulated sensor data in the laboratory.
However the true vehicle model and control action to the
vehicle can only be verified through real-time flight tests.

Fig. 8. The complementary INS/GPS/Baro Kalman filter.

V. G UIDANCE AND F LIGHT C ONTROL L OOP

The guidance loop forms an outer control loop in au-
tonomous mode. It computes guidance demands to force the
vehicle to follow the desired way-point. The flight control (or
autopilot) loop forms the inner control loop and it generates
the actual control signals to follow the guidance objectives as Fig. 10. The control structure.
well as to stabilise the vehicle attitude and its rate.

A. Guidance VI. R EAL - TIME R ESULTS

The guidance loop generates the guidance demmands from Several real-time autonomous flight tests were performed
the current vehicle states and the next waypoint information. at the test site of the University of Sydney. To evaluate the
The guidance demands are desired vehicle speed with respect navigation performance, a more precise navigation system
to the air, desired height and bank angle as shown in figure 9. which usually has an order of accuracy is required as a
If the autonomous mode is activated, it selects the appropriate reference system. However, due to the limited payload capacity
next waypoint depending upon the guidance state. Then it of the UAV, this was not adopted. Instead, the covariance and
decides if the waypoint has been intercepted or missed. If it is innovation data from the fusion filter were used to evaluate
not intercepted it determines the Line Of Sight (LOS) angles the performance and the filter consistency.
and LOS rates to the next waypoint. Based on this information A flight test was performed on April 17, 2003. The purpose
it computes the lateral acceleration required to intercept the of this test was to verify the real-time autonomous flight with
next waypoint and converts this acceleration to the desired different sets of waypoint scenarios. The total flight time was
bank angle with a set of additional guidance demands: airspeed 25min and the desired flight height in autonomous mode was
and height [4]. The guidance loop implemented in this work set to 328f t above the ground and the desired air speed was
updates its guidance demands every 10Hz. set to 80kts.
 The full flight trajectory is shown in figure 11. The blue
solid line is the real-time estimated positions and the red dotted
line is the GPS indicated position. Except take-off and landing
the vehicle flied in autonomous mode. Figure 12 depicts the
vehicle trajectory from the take-off in remote control mode
until the activation of autonomous mode. Figure 13 shows the
trajectory in autonomous mode with 6 waypoints. During the
first six rounds, the waypoints are activated in CCW sequences
with some transition sequences. After the seventh round the
way points are changed to CW sequences. The former scenario
requires higher vehicle maneuvers than the latter. This can
be seen in the guidance response in figure 15 which show
larger overshoots in measured vehicle speed and height as
well as in bank angle. After performing the last scenario, the
vehicle mode is changed to remote mode to prepare the landing
approach and is returned to the ground station as shown in
figure 14. Fig. 12. The vehicle takes off in remote control mode.
During the test the number of GPS satellites changed
from eight to four during banking. The estimated vehicle 1σ
uncertainties in position were maintained within 2m in three
axes as shown in figure 16. The attitude uncertainties are
shown in figure 17 where the roll and pitch uncertainties are
maintained under 0.4 ◦ and the heading is under 0.8 ◦ .

Fig. 13. The autonomous mode is activated after the take-off. The guidance
loop uses the uplinked waypoints information to find out the next destination.
Various combinations of waypoint sequences are applied which requires
different dynamics on the vehicle.

Fig. 11. Real-time estimated 3D trajectory. the vehicle took off and landed in
remote control mode. The autonomous mode was activated after the first round
autonomous system. This system is planned to be used for
and performed several scenarios which has different sequences of waypoints. decentralised data fusion between multiple UAVs. The future
work is to make this system to be fully autonomous which
does not rely on the external GPS aiding signal by exploring
VII. C ONCLUSIONS the unknown environments and map building process.
This paper presents the real-time implementation and result
of the navigation, guidance, and control of UAV using low- ACKNOWLEDGEMENTS
cost sensors. A low tactical grade IMU is used for the inertial The authors wish to thank BAE Systems for their support
navigation and two low-cost commercial GPS receivers are in providing funding and systems engineering towards this
used to estimate and correct the INS error. The air data project.
system is built using a customised analog/digital conversion
card. It provides air pressures, temperature and engine control R EFERENCES
signal for the guidance and control loop. The GNC algorithm
[1] D.G. Meskin and I.Y. Bar-Itzhack, “Unified Approach to Inertial Nav-
was developed in C++ methodology and implemented in igation System Error Modeling,” Journal of Guidance, Control, and
an embedded computer. Real-time autonomous flight tests Dynamics, vol. 15, no. 3, pp. 648–653, May-June 1992.
shows reliable and accurate GNC performances with several [2] P. G. Savage, “Strapdown Inertial Navigation Integration Algorithm
Desogn Part I: Attitude Algorithms,” Journal of Dynamic Systems,
way-point scenarios. With the low-cost aircraft platform and Measurement and Control, vol. 21, no. 1, pp. 19–28, January-February
GNC module, the FCS becomes a modular and powerful 1998.
Fig. 14. The vehicle is back to remote control mode to prepare the landing Fig. 16. The evolution of the vehicle 1σ uncertainty in position.
approach.

Fig. 17. The evolution of the vehicle 1σ uncertainty in attitude.

Fig. 15. The vehicle responses to the guidance demands in autonomous
mode. The desired air velocity was set to 80 knts and desired altitude above
the ground was set to 328 ft. The desired bank angle was calculated from
In IEEE Transactions on Automatic Control, 15:572–578, June 1999.
the current vehicle state and the next waypoint. During the first six rounds,
[11] S. Sukkarieh, “Aided Inertial Navigation Systems for Autonomous
the vehicle undergoes more severe dynamics by controlling the sequence of
Land Vehicles,” PhD thesis‘, Australian Centre for Field Robotics, The
waypoints.
University of Sydney, 1999.
[12] S. Sukkarieh, A. Goktogan, J. Kim, E. Nettleton, J. Randle, M. Ridley,
S. Wishart, and H. Durrant-Whyte, “Cooperative Data Fusion Amongst
Multiple Uninhabited Air Vehicles,” in International Conference on
[3] M. Grewal and A. Andrew, Kalman Filtering - Theory and Practice. Service and Experimental Robotics, Italy, 2002.
Prentice-Hall, Inc., 1993. [13] J.H. Kim and S. Sukkarieh. Flight Test Results of a GPS/INS Navi-
[4] B.L. Stevens, F.L. Lewis, Aircraft Control and Simulation, John Wiley gation Loop for an Autonomous Unmanned Aerial Vehicle (UAV). In
and Sons, Inc., 1992. Proceedings of the 15th International Technical Meeting of the Satellite
[5] Skogestad, Postlethwaite, Multivariable Feedback Design, John Wiley Division of the Institute of Navigation, pages 510–517, September, OR,
and Sons, Inc., 1997. USA, 2002.
[6] A. H. Lewantowicz. Architectures and GPS/INS Integration: Impact on
Mission Accomplishment. In IEEE Position, Location and Navigation
Symposium, pages 284–289, 1992.
[7] S. Snyder, B. Schipper, L. Vallot, N. Parker, and C. Spitzer. Differential
GPS/Inertial Navigation Approach/Landing: Flight Test Results. In IEEE
Transactions on Aerospace and Electronic Systems Magazine, pages 3–
11, 1992.
[8] R.L. Greenspan. GPS and Inertial Navigation. In American Institude of
Aeronautics and Astronautics, 1996.
[9] R.E. Phillips and G.T. Schmidt. GPS/INS Integration. In Advisory
Group for Aerospace Research and Development (AGARD), Lecture
series 207:9.1–9.10, 1996.
[10] S. Sukkarieh, E.M. Nebot, and H. Durrant-Whyte. A High Integrity
INS/GPS Navigation Loop for Autonomous Land Vehicle Applications.