Anda di halaman 1dari 6

2015 7th International Conference on Intelligent Human-Machine Systems and Cybernetics

UAV Cooperative Multiple Task Assignment Based on


Discrete Particle Swarm Optimization

Shaolei Zhou, Gaoyang Yin, Qingpo Wu


Department of Control Engineering
Naval Aeronautical and Astronautical University, 264001
Yantai, China
e-mail: zhouslsd@sina.com, ygy3632@163.com, wuqingpo@hotmail.com

Abstract—The combinatorial optimization problem of into the algorithm. The iterative CTAP strategy enables
assigning cooperating unmanned aerial vehicles to multiple quicker prosecution of the targets. However, the CTAP
tasks is posed by a shared model. A discrete particle swarm algorithm is only optimal for the current tasks and does not
optimization algorithm for solving such a problem is proposed. take into account tasks that will be required when the current
A matrix representation of the algorithm’s position vector tasks are completed. Mixed integer linear programming
spaces is employed. A new update strategy for the position and (MILP) [6] and tree generation algorithm [7] can find the
speed of particle is applied in this algorithm. The performance optimal assignment, but both of them can only be applied to
of the algorithm is compared to that of particle swarm online small-sized problems, and provide the best piecewise
optimization algorithm and genetic algorithm for two different
optimal solution.
cost functions. Simulation results demonstrated the
To avoid the computational complex of the above
effectiveness and feasibility of the proposed approach.
combinatorial optimization algorithms, nature inspired
Keywords-multiple task assignment; unmanned aerial methods such as genetic algorithm (GA) [8], ant colony
vehicle; matrix representation; update strategy algorithm (ACA) [9], and particle swarm optimization (PSO)
[10] are considered. The intelligent optimization methods
can obtain good feasible solutions for computationally
I. INTRODUCTION intensive large-sized problems, which has been a hot
The applications of unmanned aerial vehicles (UAVs) in research area. In [8], GA methodology is used to solve this
both military and civilian, such as mapping, forest fire NP-hard problem; a matrix representation of the GA
monitoring, suppression of enemy air defense, wide area chromosomes simplifies the encoding process and the
search and destroy, have received a growing attention by application of the genetic operators. However, it may not
various countries in the world [1]. UAVs not only have the yield the optimal solution and has low coverage speed. In a
obvious advantage of not putting human life in harm’s way, recent paper [11], AIS, PSO and VBA are used to obtain the
but also enable significant weight savings, lower costs, and optimal sequence and distance traveled for models that
long endurance. With the development of computer UAVs can cooperate to accomplish multi-task for the given
technology, sensor technology and wireless communication target. For optimization purpose, the velocity and the
network technology, low-cost autonomous attack munitions position are calculated using equations specified by PSO, but
have developed rapidly [2]. It equips with intelligent seeker then round them to nearest integer to get valid points in the
and data link, has the advantages of both UAV and search space.
monitions. Their mission for cooperative wide area research In this paper, based on the shared model for UAV
and destroy has been a heavily research point. WASD need cooperative multiple task assignment problem (CMTAP) [8],
perform four distinct tasks: search, classification, attack, and a UAV cooperative task assignments model for wide area
verification. The four tasks needed to prosecute each target search and destroy (WASD) mission is established. A novel
must be performed in the specified order. discrete particle swarm optimization (DPSO) algorithm is
In recent years, extensive work has been devoted to this applied to solve it. The performance of the algorithm is
problem and emerging cooperative decision and control compared to that of particle swarm optimization algorithm
algorithms of different classes have been proposed, such as and genetic algorithm. The remainder of this paper is
customized combinatorial optimization methods and nature organized as follows. In section 2, the cooperative task
inspired techniques. assignments model for WASD is established. In section 3,
In earlier studies [3,4], a linear programming formulation the application of DPSO algorithms to the problem is
known as a capacitated transshipment assignment problem presented. Simulation results are given in Section 4. Finally,
(CTAP) to represent multiple task assignment problem, and the paper is concluded in Section 5.
establishing CTAP network optimization model to solve it.
In [5], by running the CTAP algorithm iteratively, known
information about required future task can be incorporated

978-1-4799-8646-0/15 $31.00 © 2015 IEEE 81


DOI 10.1109/IHMSC.2015.206
II. COOPERATIVE MULTIPLE TASK ASSIGNMENT MODEL whole task accomplish time are two main objective functions
for evaluating the operational effectiveness [8]. Because each
A. Problem Description UAV has a constant speed, the task accomplish time can be
We assume that there are a set of N v vehicles with a represented by traveling distance.
1) Minimum total travelling distance (fuel consumption)
constant speed simultaneous deployed by multiple launch NV
platform; the terrain has already been searched by them Min J1 = ¦ L ( Plani ) (2)
and Nt targets have been found. Let T = T1 , T2 ," , TNT be { } i =1

L ( Plani ) is the distance traveled by vehicle i from the


the set of targets found and let V = V1 , V2 ," , VNV { } be the set beginning of the mission until finishing its part in the group
of UAVs performing tasks on these targets. The set of tasks tasks plan.
that need be performed by the UAV team on each target 2) Minimum the whole task accomplish time
{ }
is M = M 1 , M 2 ," , M Ntype and we let N type be the number of Min J 2 = max ( L ( Plani ) )
i∈V
(3)
such tasks. For WASD mission, M = {Classify, Attack,
Verify}, N type = 3 . We let N c be the total number of tasks, III. MULTIPLE TASK ASSIGNMENT USING DPSO

{
then N c = NT N type . Task = Task1 , Task2 ," , Task Nc } is the A. Structure of particles
target set. Every particle in swarm represents a full multiple task
There are three constrains for cooperative multiple task assignment in search space. The population is denoted by a
matrix X = [ X 1 , X 2 ," , X m ] , where m is the size of the
T
assignment problems:
1) Task precedence and time constraints. The precedence
and time constraints states that three tasks needed to population. It is customary to use a string for the particle
prosecute each target must be performed in the specified X i encoding. Taking the task precedence and coordination
order. A target can be attacked only after it has been into account, for our problem we use a matrix. Each particle
classified, and a target can be verified only after it has been matrix is composed of two rows and N c columns,
attacked.
ª xi1 xi 2 " xiNc º
2) Task coordination. For group coordination each task Xi = « ' ' »
. The first row represents the
¬« xi1 xi 2 " xiNc ¼»
'
should be accomplished once. UAVs should not be assigned
to attack a target twice, unless the target is verified alive after assignment of vehicle i ∈ V to perform a task on
an attack or there is a predefined need for multiple attacks;
target j ∈ T appearing in the second row. The ordering of
each target exactly N type tasks are performed. We let
appearance of the target determines which task k ∈ M is
xi , j ∈ {0,1} be the decision variables that is 1 if vehicle being performed; the first time a target appears in the second
Vi performs Task j and is 0 otherwise. row (from left to right) means it has been assigned to be
NV
classified. Thus, we denoted a feasible particle as defining an
¦x
i =1
ij =1, ( j = 1, 2,", N c ) (1) assignment for the cooperating vehicles performing
exactly N type tasks on each of the Nt targets [8].
3) Maximum voyage constrains. The total voyage
requirement of the tasks for each vehicle does not exceed its B. Evaluation of Particles
maximum voyage. The performance of each particle is calculated by a
predefined fitness function based on Eqs.(2) and (3).
B. Combinatorial Optimization Problem N mi
L ( Plani ) = ¦ Routek( )
i
We assume task assignments to Vi are as follows: (4)
k =1

° {
­ Plani = stage1 , stage2 , ", stageN
(i )
mi
(i ) (i )
} This paper is focus on the performance of algorithms to
® find best task sequence and not on trajectory optimization. It
(i )
{ (i)
°̄ stagek = Task j , Route k , Task j ∈ Task } is assumed that the UAVs are directly overhead the target to
accomplish any task. Route(ki ) is the minimum distance to be
where Stagek(i ) is the k th task stage of Vi , N mi is the number
travelled by Vi to perform Task j at its k th stage, the
of total tasks assigned to Vi . Task j is the task be executed by
distance can be obtained through rapid route prediction
Vi in its k th task stage. Route(ki ) is the distance to be algorithm. Thus, we calculate each column from left to right
travelled by Vi to perform Task j at its k th stage, the distance and remember the distance each vehicle need to travel,
can be obtained through rapid route prediction algorithm. meanwhile, the sum of distance will be modify to meet task
The goal of cooperative multiple task assignment for precedence and time constraints by prolong some UAV’s
UAV group is to achieve the best overall effectiveness of voyage. If vehicle i performs two or three tasks consequently
them, while the total distance (or fuel consumption) and the to target j , it is assumed to hover about the target position,

82
the Euclidean distance between two tasks is 2 × π × Rmin , 2)
Rmin is the minimum turn radius of vehicle i . ­ F2 (φi (t ), Pi (t )), rand () < c1
ψ i (t ) = c1 ⊗ F2 (φi (t ), Pi (t )) = ®
¯ φi (t ), rand () ≥ c1
C. Position Updating of Particles
The second part is the cognition part of the particle,
Each particle adjusts its position in the search space which represents particle i inherit partial information from
according to its own flying experience and companions’ its individual optimal value. It defined as a cross operation
flying experience. The traditional update strategy for the
with a probability of c1 . As shown in figure 2,
position and speed of particle in PSO algorithms is used for
solving continuous problems, while the optimization rand () generates a uniformly distributed pseudorandom
problem for cooperative multiple task assignment is a number on the unit interval, if rand () < c1 , two random
discrete problem. Here, a new update strategy for the numbers a and b are generated, a ∈ [1, N c ] , b ∈ [1, N c ] ,
position and speed of particle was applied [12, 13,14].
X i (t + 1) = c2 ⊗ F3 (c1 ⊗ F2 (ω ⊗ F1 ( X i (t )), Pi (t )), Pg (t )) (5) a > b . Assignments of vehicles to targets from task a to
task b in particle i generated by the first part are replaced
ª xi1 xi 2 " xiNc º by its individual optimal assignments of vehicles to targets
where X i = « ' ' »
is the position of particle i
«¬ xi1 xi 2 " xiNc »¼
'
from task a to task b .
a b a b
during the iteration procedure, the search space is N c
x x x " x x x p p p " pi ( N −2 ) pi ( N −1) piN
ª pi1 pi 2 "
piNc º i ( Nc −2 ) i ( N c −1 ) iN c i1 i2 i3
i1 i2 i3 c c c

dimension, Pi = « ' " x " pi ( N −2 ) pi ( N −1) piN


' ' '

» is the personal best x '


x '
x ' '
i ( N c −2 )
x '
i ( N c −1 ) x '
p i1
p i2
p i3
' ' '

pi 2 " piN' c »¼
i1 i2 i3 iN c c
' c c

«¬ pi1 φ (t )
i
P (t ) i

ª pg1 pg 2 " pgNc º x p " p x x


position of particle i , Pg = « ' » is the i1
x i2 i3 i ( Nc −2 ) i ( N c −1 ) iN c

¬« pg1 pg 2 " pgNc ¼»


' '
" p
' '
x '
i1
x '
i2
p i3 i ( Nc −2 ) x '
i ( N c −1 ) x '
iN c

global best position, ω is the inertial weight, c1 and c2 are ψ (t ) = F (φ (t ), P (t ))


i 2 i i

the acceleration coefficients which are non-negative constant. Figure 2. Cross operation to individual optimal value.
F1 ( X i (t )) is the function of particle X i (t ) , which represents
influence the velocity of particle itself has on its position 3)
change. F2 ( X i (t ), Pi (t )) represents learn operation of particle ­ F (ψ (t ), Pg (t )), rand () < c 2
X i (t + 1) = c 2 ⊗ F3 (ψ i (t ), Pg (t )) = ® 3 i
X i (t ) to its personal best position Pi (t ) , while ¯ ψ i (t ), rand () ≥ c 2
F3 ( X i (t ), Pg (t )) represents learn operation of particle X i (t ) The third part is the social learning part of the particle,
which represents particle i inherit partial information from
to global best position Pg (t ) . current global optimal value. It defined as a cross operation
Position updating equation is made up of three parts: with a probability of c 2 . As shown in figure 3, rand ()
­ F ( X (t ), rand () < ω generates a uniformly distributed pseudorandom number on
1) φi (t ) = ω ⊗ F1 ( X i (t )) = ® 1 i
¯ X i (t ), rand () ≥ ω the unit interval, if rand () < c 2 , two random numbers a
The first part is the inertia part of the particle, which and b are generated, a ∈ [1, N c ] , b ∈ [1, N c ] , a > b .
represents the position change due to the velocity of itself. It
Assignments of vehicles to targets from task a to task b in
defined as a target substitution operation with a probability
particle i generated by the second part are replaced by current
of ω . As shown in figure 1, rand () generates a uniformly
global optimal assignments of vehicles to targets from task a
distributed pseudorandom number on the unit interval, if
to task b .
rand () < ω , two random numbers a and b are generated, a b a b
a ∈ [1, NT ] , b ∈ [1, NT ] , a ≠ b . Target a and target b in the x x x " xi ( N −2 ) xi ( N −1) xiN c
p g1
p g2
p g3 " p g ( N c −2 )
p g ( N c −1 )
p gN c
i1 i2 i3 c c

second row of particle i are replaced by each other, which x '


i1
x '
i2
x '
i3
" xi ( N −2 ) xi ( N −1) xiN'
' '

c c c
p '
g1
p '
g2
p '
g3 " p
'
g ( N c −2 ) p '
g ( N c −1 )
p '
gN c

means assignments of vehicles to perform tasks on target a ψ (t )


i P (t )
g

and target b are exchanged.


x p p " pg ( N −2 ) pg ( N −1) xiN
" xi ( Nc −3) xi ( Nc −2) xi ( N c −1) i1 g2 g3 c
xi1 xi 2 xi 3 xi 4 xi 5 xiN c c c

" pg ( N −2 ) p g ( N −1) xiN'


' ' '
X i (t ) x '
p p '

xi'1 a b xi' 4 a " a b b xiN' c i1 g2 g3 c c c

X i (t +1) = F3 (ψ i (t ), Pg (t ))
a b

xi1 xi 2 xi 3 xi 4 xi 5 " xi ( Nc −3) xi ( Nc −2) xi ( Nc −1) xiNc Figure 3. Cross operation to current global optimal value.
φi (t ) = F1 ( X i (t ))
xi'1 b a xi' 4 b " b a a xiN' c

Figure 1. Target substitution operation.

83
D. Local Search and Variation Strategy of Personal and 6) compute the personal best position Pi ( k ) of each
Global Best Value
particle and global best position Pg ( k ) of the population
In each generation cycle, the position of personal best Pi
using Eqs.(8) and (9);
and global best Pg are computed from Eqs.(6) and (7) 7) repeat Steps 3) to 6) until the maximum number of
respectively, iteration allowed has been reached;
Pi k +1 = arg min fitness ( X i j ) 8) output the Pg as the optimal task assignment when the
1≤ j ≤ k +1 loop ends.
k +1 (6)
°­ P , if fitness( Pi ) ≤ fitness( X i )
k k
= ® ik +1 k +1
IV. EXAMPLES
°̄ X i , if fitness ( Pi ) > fitness ( X i )
k
The performance of the proposed DPSO is analyzed in
Pgk +1 = arg min fitness ( Pi k +1 ) this section using simulation in a Matlab programming
1≤ i ≤ m environment. We assumed that there are a set of four
­° P k , if fitness ( Pgk ) ≤ fitness( Pi k +1 ) (7) vehicles with constant speed simultaneous deployed by
= ® kg+1 k +1 multiple launch platform; the terrain has already been
°̄ Pi , if fitness( Pg ) > fitness( Pi )
k
searched by them and seven targets have been found. The
To accelerate the optimization algorithm, local search initial situation map is shown in figure 4. The initial
and variation strategy of personal best Pi and global best Pg positions of targets and UAVs are as Table 1.
are applied in each generation cycle. The local search 80
strategy is that two random numbers a and b are generated, Target 1 Target 6

a ∈ [1, NT ] , b ∈ [1, NT ] , a ≠ b . Target a and target b in 70

the second row of particle i are replaced by each other, and UAV 2
a random number c is generated, c ∈ [1, N c ] , the value pick 60 UAV1
Target4
in position c of first row of particle i is replaced by a Target7
Y/km

random number d , d ∈ [1, N v ] , d ≠ pick . The position of


50
Target 3
UAV3
personal best Pi and global best Pg are computed from 40
Target 2

Eqs.(8) and (9) respectively,


Target 5
­° P ' (k ) if fitness ( Pi ' (k )) ≤ fitness ( Pi (k )) 30
Pi (k ) = ® i (8)
°̄ Pi (k ) if fitness ( Pi (k )) > fitness ( Pi (k ))
'
UAV 4
20
Pg' (k ) = arg min fitness( Pi (k ))
20 30 40 50 60 70 80
X/km

­° P ' (k ) if fitness( Pg' (k )) ≤ fitness( Pg (k )) (9) Figure 4. Initial situation map.


Pg (k ) = ® g
°̄ Pg (k ) if fitness( Pg (k )) > fitness( Pg (k ))
'

The variation strategy is that if the fitness of particle i is TABLE I. INITIAL POSITIONS
unchanged in continuous six iterations, the particle i is Target 1 2 3 4 5 6 7
replaced by a random new particle. Coordinate x 50 30 45 54 30 74 70
Coordinate y 74 40 45 54 30 74 50
E. Description of DPSO Algorithm UAV 1 2 3 4
The DPSO algorithm is implemented in detail as follows: Coordinate x 25 65 38 32
Coordinate y 63 65 47 25
1) initialize population size, particle dimensions and
maximum iteration;
2) calculate fitness of each particle i , set current position
of each particle as personal best position Pi (0) , find the The population size and maximum number of iterations
are set to 1000 and 200 respectively. Dimension is equal to
global best position Pg (0) ; the total number of tasks N c = 21 . For DPSO, the inertial
3) renew the position of each particle in the swarm from weight coefficient ω and acceleration coefficients c1 are
Eq.(5); decreasing from 0.9 to 0.4 linearly, while acceleration
4) compute the personal best position of each particle and coefficients c 2 is on the contrary, Rmin =2 km. The simulation
global best position of the population using Eqs.(6) and (7);
5) apply local search and variation strategy to personal parameters of GA and PSO are respectively the same as
and global best value; reference [8] and [11]. The performance of the DPSO is
compared to the PSO and GA algorithms for the two
different cost functions of Eqs.(2) and (3).

84
A. Cost Function of Eqs.(2) during the process of iteration is presented in Fig.8. The
The performance of the DPSO compared to the PSO and optimal multiple task assignment plan is as table 3.
GA algorithms for the cost functions of Eqs.(2) is shown in 150
PSO
Fig.5. The change of mission distance cost to each vehicle 140 GA
during the process of iteration is presented in Fig.6. The DPSO

optimal multiple task assignment plan is as table 2. 130

Mission distance cost


120
500
PSO
GA 110
DPSO
450
100
Mission distance cost

90
400
80

350 70
0 20 40 60 80 100 120 140 160 180 200
Number of iterations

300 Figure 7. The convergence curve of each algorithm.

250 150
0 20 40 60 80 100 120 140 160 180 200
UAV1
Number of iterations
140 UAV2
UAV3
Figure 5. The convergence curve of each algorithm. 130 UAV4

120

Mission distance cost


250
110
UAV1
UAV2
100
UAV3
200
UAV4
90
Mission distance cost

80
150

70

100 60
0 20 40 60 80 100 120 140 160 180 200
Number of Iterations

50 Figure 8. Mission distance cost to each UAV.

0
0 20 40 60 80 100 120 140 160 180 200
Number of Iterations TABLE III. MULTIPLE TASK ASSIGNMENT PLAN

Figure 6. Mission distance cost to each UAV. UAV L Task assignment plan
T1(C,27.313)T4(A,47.709)T4(V,60.275)
V1 73.003
T3(V,73.003)
TABLE II. MULTIPLE TASK ASSIGNMENT PLAN T6(C,12.727)T6(A,25.294)T6(V,37.861)
V2 74.427
T1(A,61.861) T1 (V,74.427)
UAV L Task assignment plan
V1 0 T3(C,7.280)T3(A,19.846)T4(C,32.574)
V3 74.199
T6(C,12.727)T6(A,25.294)T6(V,37.861) T7(C,49.067) T7 (A,61.633) T7(V,74.199)
V2 86.993
T1(C,61.861)T1 (A,74.427)T1 (V,86.993) T5(C,5.385) T2(C,15.385) T2(A,27.951)
V4 60.518
T3(C,7.280)T3(A,19.846)T3(V,32.413) T5(A,37.951) T5 (V,50.518) T2(V,60.518)
V3 111.898 T4(C,45.141) T4 (A,57.707) T4(V,70.273) Min J=74.427 C-classify A-attack V-verify
T7(C,86.766)T7 (A,99.332) T7(V,111.898)
T5(C,5.385) T2(C,15.385) T5(A,25.385)
V4 55.385
T2(A,35.385)T5 (V,45.385) T2(V,55.385) From the convergence curve of each algorithm shown in
Fig.7, it is apparent that, for the cost function of Eqs.(3), the
Min J=254.276 C-classify A-attack V-verify
proposed DPSO algorithm outperforms the other methods.
From the convergence curve of each algorithm shown in C. Sample Runs Comparison
Fig.5, it is apparent that, for the cost function of Eqs.(2), the From the data in table 2 and table 3, we can see that
proposed DPSO algorithm outperforms the other methods. using the cost function of Eqs.(2)ˈthe total mission distance
B. Cost Function of Eqs.(3) cost is 254.276, less than the value of using the cost function
of Eqs.(3), which value is 283.174. However, the
The performance of the DPSO compared to the PSO and cooperation is enhanced by using the cost function of Eqs.(3),
GA algorithms for the cost functions of Eqs.(3) is shown in all four vehicles participate in executing the group tasks, and
Fig.7. The change of mission distance cost to each vehicle the mission cost for each vehicle is more balanced.

85
V. CONCLUSIONS programming,” Proc. AIAA Unmanned Unlimited Conference,
Chicago, IL,2004.
In this paper, UAV cooperative task assignments [7] S. J. Rasmussen and T. Shima, “Branch and Bound Tree Search for
problem for wide area search and destroy (WASD) mission Assigning Cooperating UAVs to Multiple Tasks,” Proc. American
is established by a shared model. A novel discrete particle Control Conference(ACC 06), Minnesota, pp.2171-2176,2006.
swarm optimization (DPSO) algorithm is applied to solve it. [8] T. Shima, S. J. Rasmussen, A. Sparks, and K. Passino, “Mutiple task
The effectiveness and performance of GA, PSO and DPSO assignments for cooperating uninhabited aerial vehicles using genetic
was compared by simulation for two different cost functions. algorithms,” Computers and Operations Research, vol. 33, pp.3252-
3269, 2005.
Simulation results showed that the proposed DPSO
[9] F. Su, C.Yan and L. C. Sheng, “UAV Cooperative Multi-task
algorithm outperforms the other methods. However, to Assignment Based on Ant Colony Algorithm,” ACTA
simple the performance comparison of these algorithms, AERONAUTICA ET ASTRONAUTICA SINICA, Vol.29, pp.184-
trajectory optimization is not considered. 191, 2008.
[10] W. Li and W. Zhang, “Method of tasks allocation of multi-UAVs
REFERENCES based on particles swarm optimization,” Control and Decision,
[1] Jeff S. Shamma. “Cooperative Control of Dynamical Systems,” John Vol.25, pp. 1359-1363, 2010.
Wiley&Sons, Ltd, 2007. [11] J.Senthilnath, S.N. Omkar, and V.Mani. “Cooperative
[2] Tal Shima and Steven Rasmussen. “UAV Cooperative Decision and Communication of UAV to Perform Multi-task Using Nature Inspired
Control: Challenges and practical Approaches,” Society for industrial Techniques,” Proc. IEEE Symposium on Computational Intelligence
and applied mathematics, Philadelphia, 2009. for Security and Defense Applications, pp.45-49, 2013.
[3] Corey Schumacher and Phillip R. Chandler. “Task Allocation for [12] Pan, Q. K., Tasgetiren M. F., and Liang Y. C. “Minimizing total
wide area search munitions,” Proc. American Control Conference, earliness and tardiness penalties with a common due date on a single-
Anchorage, AK, pp.1917-1922, 2002. mashine using a discrete particle swarm optimization algorithm,”
Lecture Notes in Computer Science, vol.4150, pp.460-467, 2006.
[4] C. Schumacher and P. R. Chandler. “Task Allocation for wide area
search munitions via network flow optimazation,” AIAA Guidance, [13] Pan, Q.K., Wang W.H., and Zhu J.Y. “Modified discrete particle
Navigation, and Control Conference and Exhibit, Montreal, Canada, swarm optimization algorithm for no-wait flow shop problem,”
pp.1-8,2001. Computer Integrated Manufacturing Systems, Vol.6, pp.1127-1132,
2007.
[5] C. Schumacher and P. R. Chandler. “Task Allocation for wide area
search munitions via iterative network flow,” AIAA Guidance, [14] Ye, W, Zhu A. H. and Fan H. D., “Multi-UCAV cooperation mission
Navigation, and Control Conference and Exhibit, Monterey, assignment based on hybrid discrete particle swarm optimization
California, pp.1-8,2002. algorithm,” ACTA ARMAMENTAR, Vol.31, pp.331-336, 2010.
[6] C. Schumacher, P. R. Chandler, M. Pachter and L. Pachter. “UAV
task Allocation with time constraints via mixed-integer linear

86

Anda mungkin juga menyukai