24
Multiple UAV Waypoint Ordering with Time Windows Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Embed Size (px)

Citation preview

Page 1: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Multiple UAV Waypoint Ordering with Time Windows

Project Presentation by Eddie Smolyansky & Shilo AbramovitchSupervisor: David Erdos

Page 2: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Presentation Structure

Project definition Previous solutions Work environment & interface Our solution

Finding shortest paths Building initial solution Moving in solution space Taboo search

Results & Discussion Summary Improvements & Future work

Page 3: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Project Definition

Vehicle Routing Problem

With Time Windows

Assumptions

Complication: No Fly Zones

Complex combinatorial optimization problem.

Page 4: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Previous Solutions

Background

Greedy

Genetic Algorithms

Simulated Annealing

Taboo Search

combinations

Page 5: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Work Environment & Interface

Main code written in C++

Graphical output using MATLAB

Input, output and interface between programs in form of text files

Page 6: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Why we chose Taboo search

It has been proven reliable

Simple and understandable concept

Easy to modify and improve

Page 7: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Stages of the algorithm

Finding the shortest paths between points and their “costs”

Finding an initial solution to the problem

Trying to improve that solution

Page 8: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Finding the shortest paths using the Floyd–Warshall algorithm

Finding the cost of going directly between all two points (including NFZ polygon points)

Allowing to pass through one more NFZ polygon points in each iteration

Along the way saving all the minimum costs (time/distance) and the shortest paths in a matrix

Page 9: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Initial solution – Solomon algorithm

Start with an empty route and add waypoints as long as possible

The waypoints we chose are those that maximize the time difference

Then we start with a fresh route until we finish with all the way points

Page 10: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Minimizing vehicle number

Discarding all empty routes

Trying to insert all the way-points of a route to the others

Upon success in discarding a route we start from the beginning of the stage

Eddie S
ha?
Page 11: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Single route changes

Page 12: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Double route changes

Page 13: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Triple route changes

Page 14: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

The search algorithm

Taboo search

A greedy search

Stop upon reaching local minima

Page 15: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

The break-out

Reversing the optimizing direction

Restarting the search upon reaching a local maximum

A fast break out but does not guarantee finding a new local minimum

Page 16: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

No fly zones

As many points as needed in each polygon

Any kind of polygons, convex or not

Any kind of combination of polygons, overlapping or not

Page 17: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

VRPTW Results – Solomon Instances

Difficulties with assessing results

Instance

# Waypoint

s

Capacity Run- Time

# UAVs

Benchmark

r101 100 200 55 sec 19 19

c101 100 200 3 sec 10 10

rc101 100 200 100 sec 15 14

r206 100 1000 50sec 3 3

c201 100 700 7 sec 3 3

c108 100 200 11 sec 10 10

Page 18: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

-10 0 10 20 30 40 50 60 70 80 90 1000

10

20

30

40

50

60

70

80

90Final Plan: 10 Vehicles

X [m]

Y [

m]

-10 0 10 20 30 40 50 60 70 80 90 1000

10

20

30

40

50

60

70

80

90Final Plan: 3 Vehicles

X [m]

Y [

m]

Clustered Formation – Low Vs. High Capacity

Page 19: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

100 Nodes - Random Formation

0 10 20 30 40 50 60 700

10

20

30

40

50

60

70

80Final Plan: 18 Vehicles

X [m]

Y [

m]

0 10 20 30 40 50 60 700

10

20

30

40

50

60

70

80Final Plan: 19 Vehicles

X [m]

Y [

m]

Page 20: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

0 10 20 30 40 50 60 700

10

20

30

40

50

60

70Final Plan: 8 Vehicles

X [m]

Y [

m]

0 10 20 30 40 50 60 700

10

20

30

40

50

60

70Final Plan: 8 Vehicles

X [m]

Y [

m]

No Fly Zones

0 10 20 30 40 50 60 700

10

20

30

40

50

60

70Final Plan: 8 Vehicles

X [m]

Y [

m]

Page 21: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Results Discussion

Versatile algorithm

Very fast

Quality results

Surpassed expectations

Page 22: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Summary

The problem

Finding shortest paths

Building initial solution

Moving in solution space

Taboo search

Results & capabilities

Page 23: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Future Work & Improvements

Graphical User Interface

Soft time windows

Improved coding (object oriented)

Page 24: Project Presentation by Eddie Smolyansky & Shilo Abramovitch Supervisor: David Erdos

Thank You For Listening!

Questions?

-10 0 10 20 30 40 50 60 70 80 90 1000

10

20

30

40

50

60

70

80

90Final Plan: 10 Vehicles

X [m]

Y [

m]