Transcript
Page 1: Applications of Single and Multiple UAV for Patrol and Target Search

Applications of Single and Multiple UAV for Patrol and Target Search.

Pinsky Simyon.

Supervisor: Dr. Mark Moulin

Page 2: Applications of Single and Multiple UAV for Patrol and Target Search

Summary of the Presentation.

Definition of UAV Patrol and Search (P&S) Problem.

Simplification and Explicit Definition of UAV Patrol and Search (P&S) Problem.

Dijkstra Shortest Path Algorithm and its application for P&S problem.

Single UAV P&S : Simulation Results. Multi UAV P&S : Simulation Results. Conclusions.

Page 3: Applications of Single and Multiple UAV for Patrol and Target Search

Background.

Today, UAV’s are mainly human operated. Area research task of today - Autonomous UAV. Area research task of tomorrow - Autonomous Team of UAVs. Applications: Search, Patrol, Mapping, Photographing etc. Common Problem of all the applications under autonomous

mode - Optimal Path Planning under constrains: terrain topography hazards fuel ,changing weather, mission updates, cooperative task etc.

Page 4: Applications of Single and Multiple UAV for Patrol and Target Search

Theoretical Solution Methods.

P&S automation problem is extremely complex. Numerous approaches to the problem definition and solution has

been investigated in resent years. We have examined the applicability of 3 methods:

Swarms Methodology : each individual UAV has little intelligence which contributes to the group intelligence. It is interesting to investigate the behavior of the group varying the intelligence of the individual UAV. The main problem is a lack of mathematical approach.

Q-learning – Very Large State Space. Dijkstra shortest path- is the applied method in this project.

Page 5: Applications of Single and Multiple UAV for Patrol and Target Search

Simplification of the original problem and its explicit definition.

Simplified Problem: A Patrol & target search under terrain constrainsThe Map is discrete, and includes regions containing Hazards. Preferred destinations. The target.

Solution: Find the Optimum Path between the bases. Applicability of the solution:

Patrol can be performed by single or multiple UAV.Nowadays performed by human operators on the US and Mexico Border by Hermes.

Page 6: Applications of Single and Multiple UAV for Patrol and Target Search

Discrete Model of Terrain.

G(N,A)- finite nonempty

set of N nodes and

collection A of arcs.

- sequence

of nodes.

- each pair

is an arc of G.

-the weight of the arc

represents the distance between the two nodes.

On the plotted map one can see:

Home Bases, Hazards,

Preferential Areas and a

Target.

),...,,( 21 lnnn

),(),...,,(),,( 13221 ll nnnnnn

),(),( jiji nndistnnw

Page 7: Applications of Single and Multiple UAV for Patrol and Target Search

Optimization Problem formulation.Given two nodes S and T (which represent the two bases) in a graph G=(N,A) , we

need to find the path P such that it minimizes the sum of the weights associated with the edges constituting the path.

Let , where and and

( denotes the neighboring nodes of )

Hence we need to find:

When is the weight between the two nodes.

Available Algorithms: Dijkstra, Floyd's(O(v^3)), Bellman-Ford(negative),…

),...,( 21 knnnP 1nS knT Nni )(1 ii nn

in

1

01),(min

k

iii

P

nnw

),( 1ii nnw

)(1 ii nn

Page 8: Applications of Single and Multiple UAV for Patrol and Target Search

Dijkstra Shortest Path Algorithm. Solves the single-source shortest-paths problem on a weighted, directed

graph G = (N, A) for the case in which all arcs weights are nonnegative. Maintains a set S of vertices whose final shortest-path weights from the

source s have already been determined. Repeatedly selects the vertex with the minimum shortest-path estimate,

adds n to S, and relaxes all edges leaving n.

DIJKSTRA(G, w, s) INITIALIZE-SINGLE-SOURCE(G, s) S ← Q ← V[G] while Q ≠ do n ← EXTRACT-MIN(Q) S ← S + {n} for each arc a Adj[n] do RELAX(n, v, w)

Page 9: Applications of Single and Multiple UAV for Patrol and Target Search

Application to the UAV case.

The field is being represented as an Adjacency Matrix.

For a grid of the dimensions nodes, create a

matrix of the dimensions where matrix value

in the row i and column j is:

2NNN

422 NNN

eapreferedArnndistm

hazardAreanndistM

odesconnectedNnndist

dNodesunconnecte

m

ji

ji

ji

ij

),(

),(

),(

Page 10: Applications of Single and Multiple UAV for Patrol and Target Search

Application to the UAV case.

For iteration number k

update the map :

while:

And update the visited preferred areas.

The running time for n nodes and m=8n edges in current configuration is O(n(n+8))=O(n^2).(n=resX*resY)

1]1[][ kmkm ijij

Page 11: Applications of Single and Multiple UAV for Patrol and Target Search

Simulation Results - The width of the path.

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

Path Width=1 Path Width=2

The width of the path is a factor of an efficiency of the patrol.The wider path has smaller probability to be a ’useless’ sortie.

Page 12: Applications of Single and Multiple UAV for Patrol and Target Search

Simulation Results-The size of the preferred area.

Radius = 2 Radius = 4 0 200 400 600 800 1000

0

100

200

300

400

500

600

700

800

900

1000

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

• The vertices leading toward the preferred area has relatively law weight. • The radius around this area reflects the relative priority of this area.• The bigger radius causes higher priority of the region which defines how fast the UAV will choose to visit this area.

Page 13: Applications of Single and Multiple UAV for Patrol and Target Search

Simulation Results - The target location relative to the shortest path.

“Shortest Path” definition: the shortest route connecting the initial and the final points (bases) on a map with no hazards and no preferred areas.

Intuition: if the distance of the target from the shortest path is bigger, it is harder to find it -more sorties are needed.

Results of two simulations for two different distances are presented.

Page 14: Applications of Single and Multiple UAV for Patrol and Target Search

Simulation Results - The target location relative to the shortest path.

Two results are not sufficient for final conclusions.

Two different aspects are found to be interesting for dipper investigation:

1. The correlation between the

distance of the target from the ‘main’ (shortest) route, and the number of sorties needed to reveal the target.

2. The connection between the different Hazards densities, and the average number of sorties needed for detection.

Page 15: Applications of Single and Multiple UAV for Patrol and Target Search

Simulation Results - The target location relative to the shortest path.

0 100 200 300 400 500 6000

2

4

6

8

10

12

14Number of Sorties via Target Distance from center-10 Hazards

Distance from the shortest route

Num

ber

of S

ortie

s

real Values

3rd Degree polinom

• The graph presents the number of sorties needed to reveal the target via the distance of the target from the “shortest path”. (for fixed hazards density-10 hazards)• The more distant the target is located, the higher the number of the sorties which are needed for detection.

Page 16: Applications of Single and Multiple UAV for Patrol and Target Search

Simulation Results - The target location relative to the shortest path.

0 100 200 300 400 500 6000

2

4

6

8

10

12Number of Sorties via Target Distance from center

Distance[]

Num

ber

of S

ortie

s

10 Hazards20 Hazards

30 Hazards

40 Hazards

50 Hazards60 Hazards

• The graph presents the number of needed sorties via distance from the “shortest path” for different hazard density levels:10-60Haz/Field. • The same behavior takes place for the fields with different hazard density.

Page 17: Applications of Single and Multiple UAV for Patrol and Target Search

Simulation Results - The target location relative to the shortest path.

10 15 20 25 30 35 40 45 50 55 601.6

1.8

2

2.2

2.4

2.6

2.8

3

3.2Average Number of Sorties via Increasing Hazard Number

Number of Hazards

Ave

rage

Num

ber

of S

ortie

s

real Values

3rd Degree polynom

FieldAreaHaz %4040

• The graph presents the average number of sorties via increasing Hazard Density.• An increase of the density yields an easier reveal of the target, until the optimum is reached.• After the optimum point, the task is becoming harder. Why?• An increase of low density means searching in fewer regions.• An increase of high density means searching inside a labyrinth.

Page 18: Applications of Single and Multiple UAV for Patrol and Target Search

Application for Multi UAV search.

Simulation scenario : Number of UAVs=2 4 bases50 hazards

18 Preferred Areas 1 Target The UAV Group is provided an

updated map of hazards and preferred areas.

Each UAV updates the map according to its planned sortie.

Serial planning: each UAV plans its route taking into account the

plans of other group members.

Page 19: Applications of Single and Multiple UAV for Patrol and Target Search

Animation-Multi UAV search.

Legend :

Blue Six Pointed Stars-Bases 1-4.

Yellow Dots-Preferred areas.

Red Dot-The Target. Red Trace-UAV1. Blue Trace-UAV2. The Hazards represent the terrain.

Result: The patrol area is being divided between two UAVs into the inner and the outer parts.

Page 20: Applications of Single and Multiple UAV for Patrol and Target Search

Conclusions:

We have examined Several methods for solving the Patrol & Search problem.

We have found Dijkstra algorithm to be the most promising one. For each application the right parameters have to be found and

tuned. For the simulated conditions:

The path width = 2 nodes. The preferred area radius = 2 nodes. 40% Hazards density has been found to be the optimum.

The project results show the practical applicability of the “shortest path” problem solved using the Dijkstra algorithm for: Sorties planning for multiple/single UAV. Search/Patrol performed by multiple/single UAV.

Page 21: Applications of Single and Multiple UAV for Patrol and Target Search

Thanks…

To Dr. Mark Moulin for the guidance. To the rest of you for your attention.


Top Related