Autonomous Decisional High-level Planning for UAVs-based Forest-fire
Localization
Assia Belbachir and Juan-Antonio Escareno
Mechatronics Department, Polytechnic Institute of Advanced Sciences, IPSA, Ivry-sur-Seine, France
Keywords:
Autonomous Decision, High-level Planning, Control, UAVs.
Abstract:
This paper addresses the problem of forest-fire localization using unmanned aerial vehicles (UAVs). Due to
the fast deployment of UAVs, it is practical to use them. In forest fires, usually the area to explore is unknown.
Thus, existing studies use an automatic or semi-automatic exploration strategy following a zig-zag sweep pat-
tern or expanding spiral search pattern. However, such an approach is not optimal in terms of exploration time
since the mission execution and achievement in an unknown environment requires autonomous vehicle deci-
sion and control. This paper presents an enhanced approach for the fire localization mission via a decisional
strategy considering a probabilistic model that uses the temperature to estimate the distance towards the forest
fire. The UAV optimizes its trajectory according to the state of the forest-fire knowledge by using a map to
represent its knowledge and updates it at each exploration step. We show in this paper that our planning and
control methodology for forest-fire localization is efficient. Simulation results are carried out to evaluate the
feasibility of the generated paths by the proposed methodology.
1 INTRODUCTION
Applications of unmanned aerial vehicles (UAVs)
have evolved in a dramatic way. The development
of the technological aspect has enhanced the oper-
ational profile towards assigned tasks. The current
operational autonomy is translated into more opti-
mized data collection. The latest ecological trend has
also encompassed the inclusion of these aerial agents.
Applications such as weather forecasting, where the
aerial robot performs real-time information gathering
of different weather variables, or as animal preser-
vation, where vehicles are able to detect and track
hunting-targeted animals, or natural disaster assess-
ment such as floods, hurricanes, forest-fires.
Forest-fires represent a very large issue due to its
significant impact, not only environmental or eco-
nomic but also social. For this reason, efficient lo-
calization of forest-fires is a valuable feature during
the assessment of this disaster. Dealing with such an
event implies a huge mobilisation of resources, either
human or material. However, the main problem of
forest-fire localization is the lack of time to detect the
fire center and to predict its evolution in time. There-
fore, well-known capabilities of UAVs, such as their
fast deployment, make them a candidate solution for
forest-fire localization.
Several works demonstrate that localization of
forest-fire is possible (Ambrosia et al., 2003) (Maza
et al., 2011). However, existing solutions, UAVs ex-
ecute predefined tasks that are sequenced and mon-
itored by distant human operators. The communi-
cation between the human operator and the UAVs is
highly constrained and affect the efficiency of the mis-
sion since the human operators are in the control loop.
In this regard, in order to improve the localiza-
tion efficiency some researchers have embedded deci-
sional autonomy that automatically plans or organises
the robots activities, controls the execution of their
goals and monitors the state of the systems (Di Paola
et al., 2015) (Song et al., 2000) (Ingrand et al., 2007).
However, these approaches are used in terrestrial and
underwater robotics. Other decisional autonomy ap-
proaches are implemented in UAVs (Merino et al.,
2006). However, the decisional autonomy is related to
the low decisional level. High level decisional auton-
omy is a necessity for autonomous exploration within
unknown area.
In this paper, it is embedded a high-level au-
tonomous system adapted for UAVs to explore un-
known areas and to detect forest fires sources. In gen-
eral, exploration strategies deal with limited explo-
ration time and wide forests exploration areas. Since
UAVs features limited energy resources is a capital is-
Belbachir, A. and Escareno, J-A.
Autonomous Decisional High-level Planning for UAVs-based Forest-fire Localization.
DOI: 10.5220/0005972501530159
In Proceedings of the 13th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2016) - Volume 1, pages 153-159
ISBN: 978-989-758-198-4
Copyright
c
2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
153
sue during exploration operations, the paper details an
architecture that generates adaptive, cooperative and
autonomous decisions for each aerial robot.
Likewise, in order to judge the feasibility of the
proposed navigation strategy, a low-level controller
is integrated to track the generated paths by the au-
tonomous decisional guidance system.
The outline of the paper is organized as follow.
Section 2 presents the previous woks and describe
the classical challenges of UAVs associated to a mis-
sions achievement. Section 3 details the proposed
methodology for autonomous UAVs, UAVs map up-
dates, the exploration strategy and the kynodynamics
model. Section 4 shows our architecture implementa-
tion used in each UAV. Section 5 presents the effec-
tiveness of our methodology at simulation level.
2 PREVIOUS WORKS AND
MISSION CONTEXT
The field of UAVs corresponds to a growing and
challenging topic. The application of such UAVs
are diverse (Valavanis and Valavanis, 2007) (Chen
et al., 2013), they can be used for exploration and
data collection, for rescue, fire detection, multi-UAV
grasping (Parra-Vega et al., 2012), synchronized fly-
ing (Schoellig et al., 2012) and even pole throw-
ing and catching (Brescianini et al., 2013). UAVs
can also cooperate to accomplish a common mis-
sion (G. Loianno and Kumar, 2015), such as the for-
mation flight (Merino et al., 2010) in which each
UAV is keeping a specific distance throw its neigh-
bor, coordinated rendezvous (Beard et al., 2002) to
localize fixed targets and avoid threads, coordinated
path planning (Nikolos and Brintaki, 2005) and task
coordination (Sujit et al., 2005). However, most of
the research mainly rely on operational UAV auton-
omy. This means that UAVs receive a pre-planned
tasks without high-level decisional reasoning. A few
works intend to introduce the decisional autonomy for
UAVs (Merino et al., 2006) such as COMETS (Ollero
et al., 2005).This project uses several UAVs, where
some of them are directly controlled by an operator
and others only rely on their operational autonomy.
As a result, these approaches are implemented at low-
decisional autonomy level such as perception opera-
tions. We consider that decisional task planning has
a significant potential to develop novel smart UAVs
capable to plan and to organize autonomously while
considering distance or time constraints, goals execu-
tion and system state verification. In order to provide
the aforementioned autonomy, we have developed the
proposed architecture using an existing system (T-
ReX), which provides an embedded planning and ex-
ecution control framework (McGann et al., 2007). Fi-
nally, a low-level is described considering a rotary-
wing configuration UAV. In this sense, a simulation
study was carried out regarding the validation of the
high-level trajectory planning during the forest-fire
localization.
3 METHODOLOGY
This section presents the way information is perceived
by UAVs used as a mapping platform in which the
probabilities of target presence at a given locations are
encoded. Thus, Firstly it is described the forest-fire
model. Secondly, we describe the approach to repre-
sent the collected data. Thirdly, we explain the pro-
posed exploration strategy for every UAV. Finally, we
provided a slight description of the kinematic model
of the UAV used to verify the decisional guidance ap-
proach.
3.1 Forest-fire Model
Since the goal is the forest-fire detection using dif-
ferent UAVs, let us consider a stationary (slow time-
varying) fire model. Considering that the forest fire
emission expands proportionally to the altitude, in our
model the target is defined according to the maximal
known temperature. The density of the temperature
T within the forest-fire is a decreasing function of
the horizontal distance ρ with respect to the forest-
fire center and the elevation z above the ground (Fig-
ure 1). This function is the model of the forest on fire,
Figure 1: Illustration of the temperature evolution within a
forest-fire in stationary weather: the temperature, here rep-
resented in red, decreases with the elevation z1, z2 and with
the distance to the vertical of the emitting fire.
which is an approximation of the actual diffusion phe-
nomenon: the model is probabilistic and expresses the
probability density function (pdf) of the temperature
T as a function of the distance ρ and the elevation z:
P(T = t|ρ,z) (1)
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics
154
The temperature dispersion is also an increasing func-
tion of the distance and the altitude. It is important
to notice that in our case several targets can interfere
with each other.
3.2 Map Updates
A common representation to map the environment is
the occupancy grids (Thrun, 2003). The environment
is subdivided into a grid. For each grid square, a prob-
ability that the cell contains an obstacle (Moorehead
et al., 2001) or a target (Low et al., 2009) is associ-
ated. Our choice is to discretize the operational envi-
ronment and use a grid map of N × M. For each grid
{x
i, j
}, i [0,N[, j [0, M[ cell we associate probabil-
ity value P
k
(x
i, j
). This represents the probability that
a fire is located within the cell at time k. Addition-
ally, each cell contains a boolean value initialized to
zero. This value is changed to one, if a UAV already
visited the cell. This condition allows to avoid fus-
ing twice data acquired from the same position. The
cell probabilities are updated incrementally according
to a classical Bayesian paradigm under a Markovian
assumption as follow:
P
k
(x
i, j
) =
P(T
k
|x
i, j
= fire)P
k1
(x
i, j
)
P(T
k
)
(2)
where P(T
k
|x
i, j
= fire) is the sensor model and P
k1
is the probability value of the fire existence at the time
k 1. We can observe that the probability P
k
(x
i, j
)
implicitly represents the precision of the fire-center
location. When the probability is equal to 1, means
that the fire is perfectly localized.
3.3 Exploration Strategy
The current paper presents the development of a dis-
tributed reactive and cooperative exploration strategy
based on a probability threshold P
con f
. The objective
of this decisional strategy is to improve the efficiency
of the exploration mission, i.e. the detection of forest
fire.
In forest-fire detection operations, based on clas-
sical approaches, the location of targets are unknown
and thus predefined exploration strategies can some-
times be inefficient in terms of time-to-detect or dis-
tance covered. In this solution the UAVs can modify
their navigation according to their current perception,
for this reason such approaches are called reactive ap-
proaches. However, most of them do not take into
account the collected data during the mission execu-
tion (Zhang and Sukhatme., 2008) (Popa et al., 2004).
Thus, in order to counteract the drawbacks asso-
ciated to reactive approaches the first stage aims at
confirming the existence and localization of the clos-
est forest-fire hypothesis, until its probability exceeds
a threshold P
con f
. The UAV heads towards ”proba-
ble” target while collecting data about the fire hypoth-
esis until its probability exceeds P
con f
. Doing so, it
is reduced the number of vehicle actions, at the cost
of less precisely localized center forest fire. Algo-
rithm 1 shows the exploration strategy for each UAV.
Four actions are accepted: a
le f t
,a
right
,a
f ront
,a
behind
The UAV choose next cell to explore in order to max-
imize P(x
i, j
/
x
i, j
=NE
). This means that the exploration
strategy chooses actions that reaches cells that are not
explored and that maximize the expected probability
of the target.
Max
a
i
∈{le f t,right,behind, f ront}
P(x
i, j
/
x
i, j
=NE
) (3)
In equation 3, the maximization function is evaluated
by one action look ahead. We can generalize the func-
tion as follow:
Max
a
i
∈{le f t,right,behind, f ront}
f (z
τ(x
i
,a
i
)
) (4)
Algorithm 1: Next Cell Exploration Algorithm.
Require: D: the diameter of the target ;
Target
x,y
: the coordinate of the target ;
P(x
i, j
): the probability that x
i, j
is the target;
a: an action ;
τ: transition function.
1: Update grid using equation (2).
2: if (i, j : i ]0,N], j ]0, M],P(x
i, j
) P
con f
) then
3: <Target Found>
4: for (i, j : i ]0,N], j ]0, M]) do
5: if (0 < |x
i, j
Target
x,y
| < D) then
6: P(x
i, j
) 0;
7: P(Target
x,y
) 1;
8: end if
9: end for
10: end if
11: <Choose Next Cell>
12: a using the equation (3) or (4).
13: x
0
i, j
τ(a,x
i, j
)
14: return x
0
i, j
where, f (z
τ(x
i
,a
i
)
) is a function that predicts the
future values in the Map or GRID, when action a
i
is
chosen.
3.4 Vehicle Model
In this part, we explain the low-level layer that drives
the aerial robot while detecting the forest-fire. In this
case, for the purpose of clarity, we restrict the motion
to the horizontal plane, following a non holonomic-
like motion. For this reason, it is more convenient
to address vehicle’s translational motion from a kine-
matic perspective, i.e. focusing on motion resulting
Autonomous Decisional High-level Planning for UAVs-based Forest-fire Localization
155
y
x
f
V

+
Forest-fire
center
t0
t1
Figure 2: Illustration of a 2D UAV Navigation.
from heading the forward velocity V
f
(see Figure 2),
i.e.
(Σ
ξ
) :
~
˙
ξ = R(ψ)~v
f
˙x = v
f
cosψ
˙y = v
f
sinψ
(5)
whereas the rotational motion within a dynamic
framework. The latter configures a 3DOF (three de-
grees of freedom) Kynodynamic model which is ex-
pressed by the following equations:
(Σ
ψ
) :
¨
ψ =
1
I
z
τ
ψ
(6)
where τ
ψ
is the yaw control input.
4 IMPLEMENTATION
The integration of the exploration strategy and the
controller for each UAV is implemented using an
existing architecture CoT-ReX (Cooperative Teleo-
Reactive EXecutive) (Belbachir et al., 2012). This ar-
chitecture is a goal-oriented system, with embedded
automated planning and adaptive execution (McGann
et al., 2007). Such architecture provides, for each
UAV, the capacity to plan, re-plan and execute its mis-
sion. Instead of being reactive, this architecture incor-
porates planners that can cope with different planning
horizons
1
and deliberation
2
times. Goals-based mis-
sion management CoT-ReX allows at the same time
being reactive (short planning horizon and delibera-
tion) against new situation such as obstacle avoidance
and deliberative such as mission execution. A CoT-
ReX agent is divided into several layers called reac-
tors. Each reactor can be deliberative or reactive, de-
pending on the horizon and deliberation time. In our
implementation, we consider the mission as a maxi-
mization problem to locate the forest fire. We assume
that the UAV has predefined goals to achieve (e.g.
way points) that defines the exploration strategy. To
1
It is the prediction time.
2
It is the time that is given to the system to generate a plan.
Figure 3: Illustration of the deployed architecture for each
UAV.
improve this strategy, the UAV is able to reason on its
online perceived data. Additionally, another reactor
(MapReactor) in the CoT-ReX architecture which al-
lows getting up-to-date information on UAV state and
modifying its trajectory if necessary. The MapReac-
tor is the component that takes into account the UAV
perception generating new goals for the Mission Man-
ager reactor. We assume that the UAV, equipped with
the temperature sensors, is able to compute the prox-
imity probability to the forest-fire using equation (1).
Figure 3 shows the used architecture. Mission man-
ager is responsible of all the exploration area. Ex-
ecuter is sending action by action to be executed by
the Controller. MapReactor generates new area to ex-
plore.
The low-level layer consider a two-level hierarchi-
cal control commonly used for rotorcraft UAVs. In-
deed, we have considered a Kynodynamic model that
also can be adapted for aircrafts. The controller con-
siders a basic PI control for the translational kinemat-
Figure 4: Exemple of an executed UAV trajectory for forest-
fire localisation.
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics
156
Time(s)
0 100 200 300 400
A
-300
-200
-100
0
100
200
300
A[deg]
A
d
Time(s)
0 100 200 300 400
error
A
(deg)
-400
-200
0
200
400
Time(s)
0 100 200 300 400
A
dot
[deg]
-400
-200
0
200
400
_
A
_
A
d
Time(s)
0 100 200 300 400
Heading rate [deg/seg]
-400
-200
0
200
400
Figure 5: Angular behavior while tracking the 2D trajec-
tory.
Figure 6: Translational behavior while following the com-
manded trajectory.
ics defined through stable error polynomials
u
x
(e
x
) = k
p
x
e
x
k
i
x
R
t
0
e
x
dτ
u
y
(e
y
) = k
p
y
e
y
k
i
y
R
t
0
e
y
dτ
(7)
where e
x
= x x
d
and e
y
= y y
d
denotes the error
variables, and k
p
and k
i
stands for the stiffness and
damping control gains. For the rotational dynamics, a
PID control is used to deal with heading of the flying
robot.
u
ψ
= k
p
ψ
e
ψ
k
d
ψ
˙e
ψ
k
i
ψ
Z
t
0
e
ψ
dτ (8)
where k
p
and k
v
ψ
are positive constants. The control
objective is reached introducing the velocity
T =
u
2
x
+ (u
y
)
2
1
2
(9)
and tracking the desired heading
ψ
d
= arctan
r
2
r
1
(10)
Table 1: Statistical Results For target localization based on
simulation results.
Amount
of UAVs
Amount of
targets
Explor.
time(ut)
Found
Targets
%of found
targets
Same
depth
1 10 100 3 30% -
1 13 100 4 30% -
1 20 100 7 35% -
2 10 100 7 70% Yes
2 10 100 7 70% No
2 20 100 13 65% Yes
2 20 100 13 65% No
2 30 100 21 70% Yes
2 30 100 21 70% No
5 SIMULATION RESULTS
We setup one UAV to explore an area with several
forest-fire. A predefined plan is embedded in the
UAV. According to the UAV perception, it modifies
its plan using its constructed map.
Figure 4 is an example of an executed UAV tra-
jectory, using the developed exploration strategy dis-
cussed in section 3.3. Figure 4 represents an hori-
zontal view of the forest-fire. We can see that thir-
teen forest-fire exists in this example. However, even
if the UAV modifies its trajectory it detects four tar-
gets. First part of Table 1, represents a statistical ex-
periment with different environment using one and
several UAVs. We used different exploration depths
for UAVs to evaluate the decisional and cooperative
exploration strategy. Each UAV communicates and
Figure 7: 2D commanded trajectory while identifying forest
fires.
Figure 8: Illustration of two trajectories and communication
points of two UAVs.
Autonomous Decisional High-level Planning for UAVs-based Forest-fire Localization
157
sends its explored data to the other UAVs using pre-
defined communication points (see an example in Fig-
ure 8). From our experimental results there is an im-
provement of the number of targets localization when
UAVs are working together. Additionally, the local-
ization precision of the targets in different depths are
better than the onces with the same depth.
The effective performance of the low-level con-
troller while tracking the commanded trajectory is de-
picted on figures 5, 6 and 7. The simulation study,
during the forest fire identification, reveals that head-
ing behavior is significantly aggressive. Thus, it
suggests that the proposed decision-based planner is
more adapte to the navigation profile of rotorcraft
aerial robots. The simulation also shows that a Kin-
odynamic model is adequate to meet the trajectories
provided by the high-level planner.
6 CONCLUSIONS
In this paper we implemented a high-level decision-
based planning to localize forest fires using a rotor-
craft UAV within an unknown exploration area. The
effectiveness of forest fire-detection missions based
on UAVs are constrained by the flight endurance of
the vehicle. Thus, it is proposed a methodology to
avoid exhaustive exploration of the fire zone. Instead,
the UAV explores the area based on the perceived data
while optimizing the decision related to the explo-
ration. Previous works on high-level task planning are
mostly used on terrestrial and underwater robotics. To
the best of our knowledge, there are no high level
task-planning used on unmanned aerial robotics. In
the herein presented proposal, we have implemented
a high decisional task-planning combined with a two-
level controller (low-level) using an UAV featuring
a kinodynamic model. The latter considers a forest
fire model to represent its evolution and also incor-
porates a map containing the perceived and the pre-
dicted data of the forest fire. The conducted simula-
tion study exhibit satisfactory performance of the pro-
posed approach applied for rotorcraft UAVs for dif-
ferent depths and in cooperative mode. The obtained
results show that the proposed methodology provides
encouraging results. Future works include the experi-
mental implementation using a quadrotor vehicle de-
veloped locally.
REFERENCES
Ambrosia, V., Wegener, S., Sullivan, D., Buechel, S., Duna-
gan, S. E., Brass, J. A., and Stoneburner, J. (2003).
Demonstrating uav-acquired real-time thermal data
over fires. Photogrammetric Engineering and Remote
Sensing,, 69:391–402. Full text available.
Beard, R. W., McLain, T. W., Goodrich, M. A., and Ander-
son, E. P. (2002). Coordinated target assignment and
intercept for unmanned air vehicles. IEEE Transac-
tions on Robotics and Automation, 18(6):911–922.
Belbachir, A., Ingrand, F., and Lacroix, S. (2012). A coop-
erative architecture for target localization using multi-
ple auvs. Intelligent Service Robotics, 5(2):119–132.
Brescianini, D., Hehn, M., and D’Andrea, R. (2013).
Quadrocopter pole acrobatics. In Intelligent Robots
and Systems (IROS), 2013 IEEE/RSJ International
Conference on, pages 3472–3479.
Chen, X., Gao, W., and Wang, J. (2013). Robust all-
source positioning of uavs based on belief propaga-
tion. EURASIP Journal on Advances in Signal Pro-
cessing, 2013(1).
Di Paola, D., Gasparri, A., Naso, D., and Lewis, F.
(2015). Decentralized dynamic task planning for het-
erogeneous robotic networks. Autonomous Robots,
38(1):31–48.
G. Loianno, J. T. and Kumar, V. (2015). Cooperative lo-
calization and mapping of mavs using rgb-d sensors.
In IEEE International Conference on Robotics and
Automation (ICRA), pages 4021 4028, Seattle, WA
USA.
Ingrand, F., Lacroix, S., Lemai-Chenevier, S., and Py, F.
(2007). Decisional autonomy of planetary rovers.
Journal of Field Robotics, 24(7):559–580.
Low, K. H., Dolan, J. M., and Khosla, P. (2009).
Information-theoretic approach to efficient adaptive
path planning for mobile robotic environmental sens-
ing. In Proceedings of the 19th International Confer-
ence on Automated Planning and Scheduling (ICAPS-
09).
Maza, J. I., Caballero, F., Capit
´
an, J., Martinez de Dios,
J. R., and Ollero, A. (2011). Experimental results in
multi-uav coordination for disaster management and
civil security applications. Journal of Intelligent and
Robotic Systems, 61(1-4):563–585.
McGann, C., Py, F., Rajan, K., Thomas, H., et Henthorn,
R., and et McEwen, R. (2007). T-rex: A deliberative
system for auv control. ICAPS.
Merino, L., and. J.R. Martinez-de Dios, F. C., Ferruz, J., and
Ollero, A. (2006). A cooperative perception system
for multiple UAVs: Application to automatic detec-
tion of forest fires. Journal of Field Robotics, 23:165–
184.
Merino, L., Caballero, F., de Dios, J. M., , Maza, I., and
Ollero, A. (2010). Automatic forest fire monitoring
and measurement using unmanned aerial vehicles. In
Proc. of the VI Intl. Congress on Forest Fire Research
ICFFR.
Moorehead, S., Simmons, R., and Whittaker, W. (2001).
Autonomous exploration using multiple sources of in-
formation. In Robotics and Automation, 2001. Pro-
ceedings 2001 ICRA. IEEE International Conference
on, volume 3, pages 3098–3103 vol.3.
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics
158
Nikolos, I. and Brintaki, A. (2005). Coordinated uav path
planning using differential evolution. In Intelligent
Control, 2005. Proceedings of the 2005 IEEE Inter-
national Symposium on, Mediterrean Conference on
Control and Automation, pages 549–556.
Ollero, A., Lacroix, S., Merino, L., Gancet, J., Wiklund, J.,
Remuss, V., Veiga, I., Gutierrez, L. G., Viegas, D. X.,
A.Gonzalez, M., Mallet, A., Alami, R., Chatila, R.,
Hommel, G., Colmenero, F. J., Arrue, B., Ferruz, J.,
Martinez, J. R., and Caballero, F. (2005). Multiple
eyes in the sky: Architecture and perception issues in
the COMETS unmanned air vehicles project. IEEE
Robotics and Automation Magazine, 12(2):46–57.
Parra-Vega, V., Sanchez, A., Izaguirre, C., Garcia, O., and
Ruiz-Sanchez, F. (2012). Toward aerial grasping and
manipulation with multiple uavs. Journal of Intelli-
gent & Robotic Systems, 70(1):575–593.
Popa, D., Sanderson, A., Komerska, R., Mupparapu, S.,
Blidberg, R., and Chappel, S. (2004). Adaptive
sampling algorithms for multiple autonomous under-
water vehicles. Autonomous Underwater Vehicles
IEEE/OES, pages 108–118.
Schoellig, A., Wiltsche, C., and D’Andrea, R. (2012). Feed-
forward parameter identification for precise periodic
quadrocopter motions. In American Control Confer-
ence (ACC), 2012, pages 4313–4318.
Song, M., Tarn, T., and Xi, N. (2000). Integration of task
scheduling, action planning, and control in robotic
manufacturing systems. Proceedings of the IEEE,
88(7):1097–1107.
Sujit, P., Sinha, A., and Ghose, D. (2005). Multi-uav task
allocation using team theory. In Decision and Control,
2005 and 2005 European Control Conference. CDC-
ECC ’05. 44th IEEE Conference on, pages 1497–
1502.
Thrun, S. (2003). Exploring artificial intelligence in the
new millennium. chapter Robotic Mapping: A Sur-
vey, pages 1–35. Morgan Kaufmann Publishers Inc.,
San Francisco, CA, USA.
Valavanis, K. P. and Valavanis, K. P. (2007). Advances in
Unmanned Aerial Vehicles: State of the Art and the
Road to Autonomy. Springer Publishing Company, In-
corporated, 1st edition.
Zhang, B. and Sukhatme., G. S. (2008). Adaptive sampling
with multiple mobile robots. In IEEE International
Conference on Robotics and Automation.
Autonomous Decisional High-level Planning for UAVs-based Forest-fire Localization
159