AN EMBEDDED SYSTEM FOR SMALL-SCALED AUTONOMOUS
VEHICLES
David Vissi`ere and Nicolas Petit
D´el´egation G´en´erale pour l’Armement, France
´
Ecole Nationale Sup´erieure des Mines de Paris, France
Keywords:
Embedded systems, autonomous vehicles, UAVs.
Abstract:
We consider the problem of designing a modular real-time embedded system for control applications with
unmanned vehicles. We propose a simple and low-cost solution. Its computational power can be easily
improved, depending on application requirements. To illustrate its performance, we report the implementation
of a 75 Hz Extended Kalman Filter used for state estimation on a small-scaled helicopter.
1 INTRODUCTION
In this paper, we present some of our research effortin
a current program aiming at proposing control strate-
gies and a control architecture for a group of hetero-
geneous autonomous vehicles.
To conduct this research, we designed a versatile
and simple real time embedded system, which can be
easily used as real time guidance and navigation sys-
tem on various platforms. Our work focuses on het-
erogenous vehicles, including small-scaled (typically
less than 2 m wide) Vertical Take-Off and Landing
aerial vehicles (VTOL as in (Castillo et al., 2005)) or
fixed wing aircraft, and ground vehicles with tank like
dynamics (as in (Morin and Samson, 2006; Vissi`ere
et al., 2007)). In the future, these vehicles will be
asked to act cooperatively on the battlefield as pic-
tured in Figure 1 (see also (Kaminer et al., 2004;
Olfati-Saber, 2006) for other scenarios).
In practice, the aerial vehicles represent the most
challenging applications in terms of navigation and
guidance. The main reason for this, is that these ve-
hicles can not easily go into any safe mode, as op-
posed to the ground vehicles which are, in compar-
ison, slower and simpler. While it was proven that,
with lowered performance expectations, it is possi-
ble to stabilize a fixed-wing Unmanned Air Vehicles
(UAV) by directly closing the loop with signals from
well-chosen sensors (e.g. in (Lee et al., 2003), the
authors propose a solution to automatically control a
fixed-wing UAV using only a single-antenna GPS re-
ceiver), it is considered by the vast majority of the
UAV community that navigation systems require data
fusion (Cheng et al., 2006). In facts, each sensor
technology has its own flaws (among which are drift,
noises, and possibly low resolution or low update fre-
quency). Yet, large factors of accuracy can be gained
by reconciliating their data.
Example of on-board data fusion applications are
ubiquitousamong autonomous vehicle control experi-
ments. Reconciliating GPS and Inertial Measurement
Unit (IMU) measurements is a classic case-study.
In (Xiaokui and Jianping, 2002), results of data fusion
from a BeeLine GPS receiver from Novatel
TM
and a
miniaturized IMU are presented. In (Cremean et al.,
2005), high-speed data fusion systems have been de-
veloped in view of the DARPA Grand Challenge.
In this later experiment, several technological
breakthroughs are presented using a high-end and
powerful computer architecture. Software compo-
nents communicate in a machine-independentfashion
through a module management system.
Our experiments can not use such a high end
setup, because the typical payload of our aerial ve-
hicles does not exceed 5 kg.
Much smaller and lower-weighting systems can
be considered though. In (Jung and Tsiotras, 2007),
an embedded system is proposed which does not
incorporate any powerful calculation board. A
simple Rabbit Semiconductor RCM-3400
TM
micro-
controller is used to perform complementary filtering
data fusion using a limited computational power. In
the same spirit, in (Jung et al., 2005), a low-cost test-
bed for UAVs is presented. It is reported that the
main advantage of designing such an autopilot from
157
Vissière D. and Petit N. (2008).
AN EMBEDDED SYSTEM FOR SMALL-SCALED AUTONOMOUS VEHICLES.
In Proceedings of the Fifth International Conference on Informatics in Control, Automation and Robotics - RA, pages 157-164
DOI: 10.5220/0001497701570164
Copyright
c
SciTePress
Figure 1: Cooperative autonomous vehicles in a future bat-
tlefield.
scratch is that, by contrast to commercially available
products (Cloud Cap Technologies, 2004; Micro Pi-
lot, 2004), it provides full access to the internal con-
trol structures. We totally agree with this point of
view.
In this paper, we present a solution lying in the
middle of the two previously mentioned categories.
Our system uses two processors. One processor is
used to gather data from the sensors and to control
the actuators. The other processor is used to perform
the data fusion calculations (and possibly the control
algorithms). The advantages of this structure are as
follows: i) task scheduling is easily programmed, be-
cause only one of the two processors is in charge of
handling the numerous devices and I/O; ii) the com-
putations are performedas one single thread on a ded-
icated board (PC type); iii) depending on the com-
putational requirements, the computation board can
be easily upgraded without requiring any software
changes or rising any concern about task scheduling;
iv) finally, the overall system is quite low-cost, since
it relies on off-the-shelf components and can be easily
maintained.
The paper is organized as follows. In Section 2,
we present our system architecture. We detail our
hardware components and comment on their choices.
In Section 3, we present as a test-case the embedding
of our system into a small-scaled helicopter. Numer-
ous details of implementation are provided. Finally,
we conclude and give directions of future work.
2 SYSTEM ARCHITECTURE
Our primary goalwas to develop an embedded system
to test algorithms of various complexity on-board var-
ious small-scaled platforms. Early in the design pro-
cess, one first constraint which appeared to us was the
payload limitations of the considered flying machines.
This lead us to focus on designing a low weight em-
bedded system.
A second issue that was also raised early in the
design stage was that the real-time requirements of a
control system for such small UAVs are very strong.
This is mostly due to the short time horizons instabil-
ities. Yet, in the context of embedded systems, real-
time scheduling of a number of sensing and compu-
tation tasks is known to be a difficult problem. More
precisely, as exposed in (Caccamo et al., 2005), the
problem of determining the feasibility of a periodic
sequence of prioritized tasks is often (NP)-hard. Suf-
ficient, but not necessary tests are pessimistic. Pop-
ular strategies such as the Rate-Monotonic policy
(see again (Caccamo et al., 2005)), which consists of
putting the highest priority on the shortest task can be
proven to be unfeasible is the CPU load is too large
1
.
While being troublesome on ground vehicles, such in-
feasibilities (and the induced inconsistencies in the
embedded calculations) would represent a cause of
potential major failure for our aerial platforms.
Keeping these two considerations in mind, we de-
cided to develop a robust two-processors embedded
system, running two distinct softwares and communi-
cating through a simple two-ways protocol. The sys-
tem specifications are as follows.
1. It performs the sensing and calculation tasks sep-
arately.
2. It is fast enough to run a typical 15 to 30 dimen-
sional states EKF algorithm with a low latency
(to eventually produce satisfactory closed-loop re-
sults).
3. Itis easy to upgrade.
4. It is versatile enough to handle various type of
sensors and communication protocols.
As exposed in Figure 2 and Figure 3, this (mod-
ular) embedded system is composed of a micro-
controller,which is in chargeof gathering information
from all the sensors, and a calculation board. These
two elements are connected by a serial interface. The
micro-controller also has a downlink to a ground sta-
tion. We now present the details of the hardware com-
ponents of our system.
2.1 Sensors
Considering both ground and aerial vehicles control
applications, we listed a series of useful sensors that
needed to be incorporated into our embedded system.
Among these are: an IMU, a GPS receiver, a pres-
sure sensor, an anemometer, magnetometers, and var-
ious switches. Other possibilities include odometers,
1
the upper limit on admissible load is 69%, approxi-
mately.
ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics
158
LADARs (as used in (Cremean et al., 2005)), and
sonars (as used in (Vissi`ere et al., 2007)), or cameras
(as used in (Hamel and Mahony, 2007)). In the con-
text of our study, we only considered low-cost sen-
sors. We now detail these. In each case, we specify
the weight (in g), the cost (in USD), the dimensions
(in mm), the update rate (f in Hz), and the protocol of
communication (Comm.).
Inertial Measurement Unit (IMU). Our IMU
is a 3DM-GX1 from Microstrain
TM
. It contains
three angular rate gyroscopes, three orthogonal
single-axis magnetometers, and three single-axis
accelerometers, along with 16 bits A/D convert-
ers and a micro-controller. This IMU can deliver
different messages, ranging from raw-data, to rec-
onciliated measurements. In our setup, we ask
the IMU to deliver only calibrated sensors data at
a 75Hz rate.
Weight Cost Size f Comm.
30 1450 39,54,18 75 RS232
Global Positioning System (GPS). Our GPS is
a TIM-LS from µblox
TM
. Through a proprietary
binary protocol, it provides position and velocity
information at a 4Hz rate. Position error is 2.5 m
(Circular Error Probability CEP) and velocity er-
ror is 2 m CEP.
Weight Cost Size f Comm.
23 100 32,47,9.5 4 RS232
The GPS receiver is not very tolerant against
power supply voltage ripples. These can be kept
below the 50 mV requirements thanks to a dedi-
cated power supply regulator from TRACO
TM
.
Barometer. Our barometer is the MS-5534 from
Intersema
TM
. Using a SPI-type protocol, it gives
calibrated digital pressure and temperature infor-
mation. This device requires a 3 V power supply
which is obtained through a fast response diode
from the main 5 V power supply of the micro-
controller.
Weight Cost Size f Comm.
2 14 5,4,2 20 SPI
Anemometer. A PXLA sensor from ASensTec
TM
delivers a differential pressure analog signal,
which can be read through a 10 bit A/D converter.
Weight Cost Size f Comm.
10 25 11,8,12 75 Analog
Magnetometer. We use a HMR2300 three axis
magnetometer from Honeywell
TM
. Its range is ±
2 gauss and it has a 70 µgauss resolution.
Weight Cost Size f Comm.
28 230 75,30,20 154 RS232
Take-off and Landing Detector. Being able to
detect take-off and landing instants is necessary to
properly initialize data fusion algorithms. In de-
tails, detection of the corresponding switches in
the dynamics defines when the controls actually
have an effect on the system. This is not the case
when an UAV is on the ground. This detection
is performed with on-off switches which can be
located, e.g., on the landing gear. They deliver a
logic signal which can be readily interpreted. To
prevent electric arcs which might cause trouble to
the connected micro-controller, we added a spe-
cific interfacing circuit. This switches can also be
replaced by active switches which can be used to
activate various devices such as digital cameras,
or parachutes.
Weight Cost Size f Comm.
10 6 25,10,5 75 Boolean
Our system is data-driven by the IMU. The main
reason behind this choice is that the IMU is consid-
ered as a critical sensor.
2.2 MPC555 Micro-controller
The micro-controller which serves as an interface
for the sensors and actuators is a MPC555 Power
PC from Motorola
TM
. It runs a specific software
we developed using the Phytec
TM
development kit.
The reason for choosing this micro-controller are as
follows. This device provides a double precision
floating-pointunit (64 bit) which is convenientfor po-
tential embedded algorithms (even if we do not use
this possibility here since all computations are per-
formed on the calculation board), it has a relatively
fast 40 MHz clock, it has 32 bit architecture and
448Kbyte of Flash memory and 4 MBytes of RAM.
Most importantly, among the family of 32 bits kits,
the MPC555 has substantial computational capabili-
ties and a large numberof versatile and programmable
Input/Output ports. In particular, we make an exten-
sive use of its TPUs (Time Process Units), UARTs,
A/D converters, SPIs, MPIOs (Modular I/O system)
(see (Motorola, 2000)). Finally, it is small (credit card
format) and has a low weight.
Weight Cost Size f Comm.
25 450 72,57,8 all all
No operating system is used on the micro-
controller. Rather the MPC555 runs a specific
interrupts-driven software written in C. Information
AN EMBEDDED SYSTEM FOR SMALL-SCALED AUTONOMOUS VEHICLES
159
Sensors
Ground
station
Remote
Control
Power PC
Vehicle
actuators
Computation
Board
Figure 2: Sensors and computation board connections to the
central micro-controller.
TPU
TPU
SCI2 MPWM
QSPI
IMU
QADC
GPS
BAROMETER
PITOT
COMPASS
Ground
Station
DATALINK
Remote
Control
Receiver
SWITCH
actuators
MDASM
QSCI1
Computation Board
TPU
T/L interrupt
MIOS
VEHICLE
+3V
+5V
+12V
main.c
Figure 3: Embedded system internal connections.
from each sensors is transferred using a dedicated in-
terrupt handler routine. Each external source or group
of sources has its own interrupt level which avoids po-
tential conflicts. Each data link is associated with a
checksum to validate reception.
The data acquisition software running on the
micro-controlleris event-drivenby the IMU messages
which periodically sends 31 bytes of data. Once the
message of the IMU is received and validated by the
micro-controller, others sensors information are ei-
ther directly read or picked in data buffers which are
constantly fed with serial messages from the sensors
through hardware interrupts. Information is gathered
in a 116 bytes message containing all the onboard
measurements. This message is sent to the calculation
board through a high-speed serial port. Once the mes-
sage is received and validated, the calculation board
carries out one navigation loop consisting of a predic-
tion equation and an estimation equation of a Kalman
filter.
2.3 PC Computation Board
The computing board is a PC running the Knop-
pix 3.8.1 Linux distribution. The PC board was se-
lected among numerous models (mostly mini-ITX,
and PC104) based on computational power, energy
consumption, toughness, and price. A fan-less board
was considered as the most relevant choice, due to the
often observed failures of fans in mechanically dis-
turbed environments.
The chosen fan-less calculation board has a stan-
dard mini-ITX PC architecture. Its processor is a
1.2Ghz C7-M from VIA
TM
designed for embedded
applications. It can perform 1500 MIPS and has clas-
sic PC Input/Output ports such as a UART serial port
(used as main data link with the micro-controller), an
ethernet board (not used here), a VGA screen output
(which can be used to monitor the system during de-
bugging phases of the software and hardware devel-
opment), a keyboard, and 4 USB ports (which can be
considered for plugging future devices such has con-
trollable cameras).
Experimental preliminary tests have shown us that
multi-threading(one threadfor messagedecodingand
one thread for the main algorithm) presents two major
drawbacks: some data can be lost, and the calculation
cycle may end unexpectedly (slightly) late. For this
reason, we decided to write our own UART driver us-
ing an interrupt handler in the kernel space. Further,
we de-activated all hardware interruptions associated
to unnecessary devices. As a result, only interrupts
from the UART are enabled. Finally, we used a single
computation thread.
The operating system is installed on a
bootable 1 Gbyte Disk-On-Chip system which
prevents all possible mechanical failure associated
to hard-drives. This flash memory device is directly
connected to the IDE port of the mother-board. The
board is powered by a pico-PSU
TM
power supply
which provides various voltages ranging from 5V to
18V. The computation software are written in C and
can be either be updated directly on the board via a
ssh
connection, or transferred, in a compiled form,
from a remote PC. Custom scripts for compiling and
distributing our executable code and configuration
files are an efficient way to upgrade the software
during development and testing.
Weight Cost Size f Comm.
800 350 170,180,40 1.5e6 RS232
3 FIELD EXPERIMENT
We now report some field experiment using our em-
bedded system. To perform model identification
tasks and design a real-time state observer in view of
closed-loop control, we decided to load our embed-
ded system into a small-scaled helicopter. In this sec-
tion, we expose this work, and give numerous details
about solutions to specific interfacing and vibrational
issues.
ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics
160
3.1 Experimental Setup
We conducted all the experiments with the help of a
human pilot which could, at any time remotely con-
nect the inputs of the actuators of the helicopter to the
outputs of our embedded system or directly to the out-
puts of the radio receiver (and thus take direct control
of the helicopter). This is achieved thanks to a re-
motely controlled switch board” which we designed.
This switch is actuated from our Graupner
TM
MC24
remote controller using one of the 12 available radio
channels. This system provides a rapid transition be-
tween manual and autonomous flight.
During all flights, the pilot could see images from
an on-board camera and read measurements from the
embedded system received for the down datalink. A
convenient Head-Up display system on the ground
station was designed for that purpose. Our data link
is a MaxStream
TM
module operating at 2.4 GHz. It
provides a RS232 serial data link at 9600 baud send-
ing the information from the embedded system to the
ground station.
Weight Cost Size f Comm.
250 230 40,68,9 30 RS232
3.2 Small-scaled Helicopter
We use a Benzin acrobatic helicopter from Vario
TM
,
with Graupner
TM
electronics (C4441 servomotors
which have high speed and high torque (8.5 Nm), a 16
channels receiver, and a yaw hobby gyroscope). It is a
very reliable helicopter platform (we never observed
any mechanical issue over mode than 100 flights). Its
payload capacity (4Kg) is large enough for numerous
reconnaissance applications. Its rotor is 1.9 m wide.
3.3 Interfacing and Vibrational Issues
Wiring the embedded system to the existing heli-
copter circuitry was achieved using some specific ad-
ditional boards and connectors. To measure the pilot’s
orders in real-time, we used a 6 channels voltage fol-
lower circuit. Numerous LEDs were added to check
the status of our system.
A central problem observed on-board helicopters
is the 25 Hz vibrations induced by the main rotor
blades. These vibrations generate a large amount of
noise on the inertial sensors. In practice, these noises
totally overwhelm the useful signals. Fortunately, it
is possible to solve this issue by using well-chosen
noise dampers. On our helicopter, we decided that
the micro-controller and the sensors would all be lo-
cated on a board which would be physically con-
nected to the frame of the helicopter through four
Figure 4: Our embedded system fitted into the (custom-
built) landing gear of a small-scaled Vario Benzin heli-
copter. Springs and dampers are used to lter out vibrations
from the main rotor blades.
spring-damper systems (see Figure 4). Experiments
conducted on a vibrating table have shown that it was
advantageous to keep the embedded system as com-
pact and as rigid as possible. The total weight of the
subsystem is about 600 g. We decided to attach some
of the batteries to it to bring the weightclose to 1.8 kg.
This enabled us to use off-the-shelf dampers yield-
ing appropriate cut-off frequencies. MV801-5CC
dampers from Paulstra
TM
were chosen for their abil-
ity to work with low masses vibrating at low frequen-
cies. With these, we obtained a satisfactory vibration
damping, with a cut-off frequency around 9 Hz. This
is represented in Figure 5. Further, resonant frequen-
cies due to the engine frequency (around 160 Hz),
the tail rotor frequency (around 115 Hz), and the tail
boom were removed using a digital notch filter. The
presented solution attenuates high frequency vibra-
tion inputs down to negligible levels.
3.4 Experimental Identification
Preliminary model identification experiments need
to be conducted before state estimation can be per-
formed.
In particular, using our embedded system, we
studied the actuators transfer functions and the yaw
rate gyroscope.
In some works (see (Mettler, 2003) or (Mettler
et al., 2000)), actuators (servomotors) are considered
as first order systems with dead band. We identified
such transfer functions for various Graupner
TM
and
Futaba
TM
servomotors. Results of various tip-in and
tip-out in reference signals were recorded to compute
the time constant of the first order model.
On board our helicopter, a hobby gyroscope from
AN EMBEDDED SYSTEM FOR SMALL-SCALED AUTONOMOUS VEHICLES
161
Figure 5: Bode diagram showing the resonance peak and the cut-off frequency of the mechanical structure equiped with the
sensors, the micro-controller, and the spring-dampers suspension. The various plots are obtained on varying locations on the
vibrating structure and show a good spatial uniformity of the vibration damping.
Graupner
TM
is used to help the human pilot keeping
the yaw rate as small as possible. Pilot orders are
transferred from the R/C receiver to the tail actuator
through this gyroscope.
To validate simple models of this transfer, we put
our IMU under this gyroscope to measure the angu-
lar velocity. Simultaneously, we connected the gyro-
scope and recorded the gyroscope signals sent to the
tail actuator. Surprisingly, it was discovered that the
gyroscope feedback behaves as a 2 Hz low-pass filter
on the pilot orders, and directly transmits the opposite
of the received angular rate, without filtering it. This
can be summarized under the form
δ
gyro
= N
r
r
m
+
δ
pilot
1+ τ
gyro
s
3.5 State Estimation
The helicopter is a 6 degrees of freedom mechanical
system with high bandwidth dynamic. Reconstruct-
ing the full state of this system from low-cost sensors
only is quite a challenge. We use an ExtendedKalman
Filter (EKF, see e.g. (Simon, 2005)) to estimate the
state of this system. In practice, the state of our EKF
is composed of 23 variables including configuration
states (13 variables using quaternions instead of Euler
angles), and the aerodynamic and external forces and
torques (10 variables including the harmonic expan-
sion of theflapping phenomenaas exposed in (Mettler
et al., 2001)). We used equally valued tuning parame-
ters for the 3 axis. These are chosen to capture fast
dynamics (statistics data are σ
acceleration
= 8 m.s
2
,
State EstimationReceive data
Send Data
Figure 6: Succession of steps of data transfer and computa-
tion.
and σ
torque
= 4 rad.s
2
, respectively). Classically,
discrete-time updates are implemented. As already
discussed, updates are synchronized with the 75Hz
measurements from the IMU.
The succession of tasks performed by the compu-
tation board of our embedded system is described in
Figure 6. As can be seen, it is necessary, due to a re-
ception time being superior to the available time be-
tween two calculation cycles to simultaneously read
or send data and perform computations. Some ex-
perimental results are presented in Figure 7. Position
estimates around hovering are reported. In practice,
they appear to be in great accordance with recorded
videos.
4 CONCLUSIONS
Designing an embedded system which can qualify
as a control system for various small-scaled air and
ground vehicles is the subject of the research project
presented in this paper. The embedded system we
propose here has some interesting features. It is sim-
ple, low cost and, most of all, easy to upgrade. Its two
processors architecture can incorporate various new
ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics
162
−2
0
2
−4
−2
0
2
4
−1
0
1
2
x(m)
y(m)
z(m)
Figure 7: Position estimates during a hovering flight.
processors and sensors.
In our laboratory, this embedded system has been
successfully used on a fixed wing aircraft (Ras-
cal 110), a small-scaled helicopter (Vario
TM
Benzin
trainer), and ground vehicles (Pioneer 4 from Mobile
Robots
TM
).
Further developments focus on giving more au-
tonomy, including path planning algorithm, follow-
ing (Kogan and Murray, 2006; Murray et al., 2003), to
respond to high-levels orders from a remote user. We
will surely need more computational power, which
can be obtained by simply upgrading the computa-
tional board. In parallel, we develop a new aerial ve-
hicle for urban area applications. For this last project
we consider using other sensors. In particular, for
obstacle avoidance and navigation, we will connect
LADARS, and ultrasonic sensors to our embedded
system.
ACKNOWLEDGEMENTS
The authors are indebted to the numerous students,
technicians and engineers who have been collaborat-
ing to the development of the presented technology
and have brought their support during the conducted
experiments.
REFERENCES
Caccamo, M., Baker, T., Burns, A., Buttazzo, G., and Sha,
L. (2005). Real-time scheduling for embedded sys-
tems. In Hristu-Varsakelis, D. and Levine, W. S., ed-
itors, Handbook of networked and embedded control
systems. Birkh¨auser.
Castillo, P., Lozano, R., and Dzul, A. E. (2005). Modelling
and control of mini-flying machines. Advances in in-
dustrial control. Springer.
Cheng, J., Lu, Y., Thomas, E. R., and Farrell, J. A. (2006).
Data fusion via Kalman lter: GPS and INS. In Ge,
S. S. and Lewis, F. L., editors, Autonomous mobile
robots, Control engineering series. Taylor and Francis.
Cloud Cap Technologies (2004). http://
www.cloudcaptech.com. Hood River, OR, USA.
Cremean, L. B., Foote, T. B., Gillula, J. H., Hines, G. H.,
Kogan, D., Kriechbaum, K. L., Lamb, J. C., Leibs, J.,
Lindzey, L., Rasmussen, C. E., Stewart, A. D., Bur-
dick, J. W., and Murray, R. M. (2005). Alice: An
information-rich autonomous vehicle for high-speed
desert navigation. Journal of Field Robotics, (9).
Hamel, T. and Mahony, R. (2007). Image based visual
servo-control for a class of aerial robotic systems.
43(11):1975–1983.
Jung, D., Levy, E. J., Zhou, D., Fink, R., Moshe, J., Earl,
A., and Tsiotras, P. (2005). Design and development
of a low-cost test-bed for undergraduate education in
UAVs. In Proc. of the 44th IEEE Conf. on Decision
and Control, and the European Control Conference
2005.
Jung, D. and Tsiotras, P. (2007). Inertial attitude and posi-
tion reference system development for a small UAV.
In AIAA Infotech at Aerospace.
Kaminer, I., Yakimenko, O., Dobrokhodov, V., Lizaraga,
M., and Pascoal, A. (2004). Cooperative control of
small uavs for naval applications. In Proc. of the 43rd
IEEE Conf. on Decision and Control.
Kogan, D. and Murray, R. M. (2006). Optimization-based
navigation for the DARPA Grand Challenge. In Proc.
of the 45th IEEE Conf. on Decision and Control.
Lee, S., Lee, T., Park, S., and Kee, C. (2003). Flight test re-
sults of UAV automatic control using a single-antenna
GPS receiver. In AIAA Guidance, Navigation, and
Control Conference and Exhibit.
Mettler, B. (2003). Identification Modeling and Character-
istics of Miniature Rotorcraft. Boston : Kluwer Aca-
demic Publishers.
Mettler, B., Tischler, M. B., and Kanade, T. (2000). System
identification of a model-scale helicopter. Technical
Report CMU-RI-TR-00-03, Carnegie Mellon Univer-
sity.
Mettler, B., Tischler, M. B., and Kanade, T. (2001). System
identification modeling of a small-scale unmanned he-
licopter. Journal of the American Helicopter Society.
Micro Pilot (2004). http://www.micropilot.com. Stony
Mountain, Canada.
Morin, P. and Samson, C. (2006). Trajectory tracking
for nonholonomic vehicles. In Kozlowski, K., ed-
itor, Robot Motion Control: Recent Developments.
Springer.
Motorola (2000). MPC555 / MPC556 user’s manual. User’s
manual, Motorola.
Murray, R. M., Hauser, J., Jadbabaie, A., Milam, M. B., Pe-
tit, N., Dunbar, W. B., and Franz, R. (2003). Online
control customization via optimization-based control.
In Samad, T. and Balas, G., editors, Software-Enabled
Control, Information technology for dynamical sys-
tems, pages 149–174. Wiley-Interscience.
AN EMBEDDED SYSTEM FOR SMALL-SCALED AUTONOMOUS VEHICLES
163
Olfati-Saber, R. (2006). Flocking for multi-agent dynamic
systems: Algorithms and theory. IEEE Trans. Au-
tomat. Control, 51(3):401–420.
Simon, D. (2005). Optimal state estimation. Wiley.
Vissi`ere, D., Chang, D. E., and Petit, N. (2007). Experi-
ments of trajectory generation and obstacle avoidance
for a UGV. In Proc. of the 2007 American Control
Conference.
Xiaokui, Y. and Jianping, Y. (2002). Study on low-
cost GPS/DMU integrated navigation system. In
AIAA/AAS Astrodynamics Specialist Conference and
Exhibit.
ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics
164