Page 1
A Multi-Population Genetic Algorithm for UAVPath Re-Planning under Critical Situation
Jesimar S. Arantes (USP)Márcio S. Arantes (USP)
Claudio F. M. Toledo (USP)Brian C. Williams (MIT)
São Carlos, SP
November – 2015
Jesimar S. Arantes (USP)Márcio S. Arantes (USP)Claudio F. M. Toledo (USP)Brian C. Williams (MIT) (USP)IEEE ICTAI 2015 November – 2015 1 / 41
Page 2
Outline
1 Introduction
2 Problem Description
3 Methods
4 Computational Results
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 2 / 41
Page 3
IntroductionOverview
Figure 1: Illustrative scenario for mission planning.Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 3 / 41
Page 4
IntroductionOverview
Figure 2: Illustrative scenario for mission planning.Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 4 / 41
Page 5
IntroductionOverview
Figure 3: Illustrative scenario for mission planning.Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 5 / 41
Page 6
IntroductionOverview
Figure 4: Illustrative scenario for mission planning.Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 6 / 41
Page 7
IntroductionOverview
Figure 5: Illustrative scenario for mission planning.Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 7 / 41
Page 8
IntroductionOverview
Figure 6: Illustrative scenario for mission planning.Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 8 / 41
Page 9
Problem DescriptionTypes of Regions and Critical Situation
Regions1 No-Fly Zone (φn)2 Penalty Region (φp)3 Bonus Region (φb)4 Remainder Region (φr )
Critical Situation1 Motor Failure (ψm)2 Battery Failure (ψb)3 Aerodynamic Surfaces Failure
type 1 (ψs1)4 Aerodynamic Surfaces Failure
type 2 (ψs2)
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 9 / 41
Page 10
Problem DescriptionTypes of Regions and Critical Situation
Regions1 No-Fly Zone (φn)2 Penalty Region (φp)3 Bonus Region (φb)4 Remainder Region (φr )
Critical Situation1 Motor Failure (ψm)2 Battery Failure (ψb)3 Aerodynamic Surfaces Failure
type 1 (ψs1)4 Aerodynamic Surfaces Failure
type 2 (ψs2)
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 9 / 41
Page 11
MethodsCodification, Decodification and Solution
Codification ut :
Decodification FΨ:xt+1 = FΨ(xt , ut) pxt+1
pyt+1vt+1αt+1
=
pxt + vt · cos(αt) ·∆T + at · cos(αt) · (∆T )2/2pyt + vt · sen(αt) ·∆T + at · sen(αt) · (∆T )2/2
vt + at ·∆T − F dt
αt + εt ·∆T
Solution xt :
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 10 / 41
Page 12
MethodsCodification, Decodification and Solution
Codification ut :
Decodification FΨ:xt+1 = FΨ(xt , ut) pxt+1
pyt+1vt+1αt+1
=
pxt + vt · cos(αt) ·∆T + at · cos(αt) · (∆T )2/2pyt + vt · sen(αt) ·∆T + at · sen(αt) · (∆T )2/2
vt + at ·∆T − F dt
αt + εt ·∆T
Solution xt :
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 10 / 41
Page 13
MethodsCodification, Decodification and Solution
Codification ut :
Decodification FΨ:xt+1 = FΨ(xt , ut) pxt+1
pyt+1vt+1αt+1
=
pxt + vt · cos(αt) ·∆T + at · cos(αt) · (∆T )2/2pyt + vt · sen(αt) ·∆T + at · sen(αt) · (∆T )2/2
vt + at ·∆T − F dt
αt + εt ·∆T
Solution xt :
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 10 / 41
Page 14
MethodsGreedy Heuristic
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 11 / 41
Page 15
MethodsGreedy Heuristic
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 12 / 41
Page 16
MethodsGreedy Heuristic
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 13 / 41
Page 17
MethodsGreedy Heuristic
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 14 / 41
Page 18
MethodsGreedy Heuristic
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 15 / 41
Page 19
MethodsGreedy Heuristic
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 16 / 41
Page 20
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 17 / 41
Page 21
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 18 / 41
Page 22
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 19 / 41
Page 23
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 20 / 41
Page 24
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 21 / 41
Page 25
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 22 / 41
Page 26
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 23 / 41
Page 27
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 24 / 41
Page 28
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 25 / 41
Page 29
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 26 / 41
Page 30
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 27 / 41
Page 31
MethodsMulti-Population Genetic Algorithm
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 28 / 41
Page 32
MethodsObjective Function
minimize fitness = −Cφb ·|φb |∑i=1
(P(xK ∈ Z iφb
)) + Cφp ·|φp |∑i=1
(P(xK ∈ Z iφp
)) +
Cφn ·max(0, 1−∆− P(∧K
t=0
∧|φn|i=1 xt /∈ Z i
φn
)) + 1
|εmax | ·K∑t=0‖ut‖ · |εt | +
shortestDist(xK ,Zφb ) +{
Cφb , vK − vmin > 00 , otherwise +
{Cφb · 2
(K−T )10 , ψ = ψb
0 , otherwise
Landing on φbLanding on φpLanding and fly on φnCurves of the UAVDistance to φbTime violationBattery failure
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 29 / 41
Page 33
MethodsMethods Used
In this work, the following methods were used.GH: Greedy HeuristicMPGA1(–GH): Multi-Population Genetic Algorithm 1
Without greedy operatorMPGA2(+GH): Multi-Population Genetic Algorithm 2
With greedy operator
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 30 / 41
Page 34
Computational ResultsAutomatically Generated Maps
Level of Difficulty1 ME : a), b)2 MN : c), d)3 MH : e), f)
Level of Coverage1 C25%: a), c), e)2 C50%: b), d), f)
Legend Colors1 φb
2 φp
3 φn
4 φr
(a) (b) (c)
(d) (e) (f)
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 31 / 41
Page 35
Computational ResultsParameters and Settings used in the Experiments
Model Parameters Value
Map Dimension X [m] 1000Dimension Y [m] 1000
UAV
Initial Position (px0 , py
0 ) [m] (0; 0)Initial Velocity (v0) [m/s] 24Initial Angle (α0) [o] 90Linear Velocity (vmin; vmax ) [m/s] [11; 30]Angular Variation (εmin; εmax ) [o/s] [−3; 3]Acceleration (amin; amax ) [m/s2] [0; 2]Number of time steps to land (T ) [s] 60Time Discretization (∆T ) [s] 1Probability of failure (∆) 0.001
MPGA
Populations 3Individuals/Pop 13Individuals Total 39Mutation Rate 0.5Crossover Rate 0.75Stop Criterion 10000
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 32 / 41
Page 36
Computational ResultsExperiments: Result obtained for the GH, MPGA1(–GH) and MPGA2(+GH)
GH MPGA1(–GH) MPGA2(+GH)Ψ Instance φb φr Inf. φb φr Inf. φb φr Inf.
ψm
ME and C25% 79 21 0 81 19 0 90 10 0ME and C50% 92 6 2 92 7 1 96 3 1MN and C25% 58 39 3 60 39 1 71 28 1MN and C50% 86 12 2 84 16 0 96 4 0MH and C25% 30 52 18 36 64 0 40 60 0MH and C50% 62 28 10 60 33 7 82 15 3
Avg 67.8 26.3 5.8 68.8 29.7 1.5 79.2 20.00 0.83
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 33 / 41
Page 37
Computational ResultsExperiments: Result obtained for the GH, MPGA1(–GH) and MPGA2(+GH)
GH MPGA1(–GH) MPGA2(+GH)Ψ Instance φb φr Inf. φb φr Inf. φb φr Inf.
ψb
ME and C25% 99 0 1 100 0 0 100 0 0ME and C50% 97 0 3 99 0 1 99 0 1MN and C25% 93 3 4 94 5 1 99 0 1MN and C50% 98 0 2 99 0 1 100 0 0MH and C25% 67 5 28 73 27 0 94 6 0MH and C50% 83 0 17 68 17 15 95 2 3
Avg 89.5 1.3 9.2 88.8 8.2 3.0 97.8 1.3 0.8
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 34 / 41
Page 38
Computational ResultsExperiments: Result obtained for the GH, MPGA1(–GH) and MPGA2(+GH)
GH MPGA1(–GH) MPGA2(+GH)Ψ Instance φb φr Inf. φb φr Inf. φb φr Inf.
ψs1
ME and C25% 81 8 11 90 8 2 91 7 2ME and C50% 88 0 12 89 0 11 93 0 7MN and C25% 68 16 16 76 18 6 86 8 6MN and C50% 82 1 17 84 3 13 89 0 11MH and C25% 41 23 36 49 46 5 67 28 5MH and C50% 56 0 44 46 23 31 78 4 18
Avg 69.3 8.0 22.7 72.3 16.3 11.3 84.0 7.8 8.2
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 35 / 41
Page 39
Computational ResultsExperiments: Result obtained for the GH, MPGA1(–GH) and MPGA2(+GH)
GH MPGA1(–GH) MPGA2(+GH)Ψ Instance φb φr Inf. φb φr Inf. φb φr Inf.
ψs2
ME and C25% 90 4 6 94 4 2 99 0 1ME and C50% 90 0 10 95 1 4 95 1 4MN and C25% 70 20 10 79 16 5 92 5 3MN and C50% 87 1 12 83 8 9 94 0 6MH and C25% 40 17 43 62 35 3 74 24 2MH and C50% 61 3 36 57 13 30 76 4 20
Avg 73.0 7.5 19.5 78.3 12.8 8.8 88.3 5.7 6.0Avg Final 74.9 10.8 14.3 77.1 16.7 6.2 87.3 8.7 4.0
Time (Sec)GH MPGA1 MPGA20.07 1.017 0.874
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 36 / 41
Page 40
Computational ResultsExperiments: Example of Routes
SE
(a)
S
E
(b)
Figure 7: Routes determined by the planner MPGA2(+GH) in a map MN withcoverage C25%: (a) ψm. (b) ψb.
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 37 / 41
Page 41
Computational ResultsExperiments: Example of Routes
S
E
(c)
S
E
(d)
Figure 8: Routes determined by the planner MPGA2(+GH) in a map MN withcoverage C25%: (c) ψs1 . (d) ψs2 .
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 38 / 41
Page 42
Computational ResultsExperiments: Example of Routes
(a) Wind Velocity 10 Knots - MN with C25% (b) Wind Velocity 50 Knots - MN with C25%
Figure 9: (a), (b) FG simulation with winds 10 and 50 knots. Wind direction:west.
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 39 / 41
Page 43
Computational ResultsVideo FlightGear Simulator
Figure 10: Video FlightGear Simulator.
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 40 / 41
Page 44
Acknowledgements
Questions send email to:[email protected]
marcio, [email protected] @mit.edu
Thank You!
Jesimar S. Arantes (USP) IEEE ICTAI 2015 November – 2015 41 / 41