Upload
others
View
12
Download
0
Embed Size (px)
Citation preview
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Ant Colony Optimization
Ant Colony OptimizationAlgorithm and Approaches in Robot Path Planning
Katinka Bohm
Universitat HamburgFakultat fur Mathematik, Informatik und NaturwissenschaftenFachbereich Informatik
Technische Aspekte Multimodaler Systeme
January 4th, 2016
Katinka Bohm 1
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Ant Colony Optimization
Structure
1. Introduction
2. Theoretical Approach
3. Robot Path Planning with ACO
4. Analysis
5. Interesting Applications
6. Recap
7. References
Katinka Bohm 2
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Introduction - Motivation Ant Colony Optimization
Motivation
Natural Inspiration! based on the the behavior of ants seeking a path between theircolony and a source of food
Stigmergy
Unorganized actions of individuals serve as a stimuli for other individuals bymodifying their environment and result in a single outcome .
In short: A group of individuals that behave as a sole entity.
Katinka Bohm 3
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Introduction - Motivation Ant Colony Optimization
Motivation (contd.)
I Swarm Intelligence method
I probabilistic technique ! non-deterministic
I solve hard combinatorial optimization problems
Definition
Combinatorial Optimization Problem P = (S ,⌦, f )S . . . finite set of decision variables,⌦ . . . constraints,f . . . objective function to be minimized
Prominent example: Traveling Salesman
Katinka Bohm 4
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Introduction - Metaheuristic Ant Colony Optimization
Metaheuristic
Ant Colony Optimization (ACO)
Set parameters
Initialize pheromone trails
while termination condition not met do
ConstructAntSolutions
DaemonActions (optional)
UpdatePheromones
endwhile
Katinka Bohm 5
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Theoretical Approach - Ant System Ant Colony Optimization
Ant System (AS)
I oldest most basic algorithm
I by Marco Dorigo in the 90s
Ant Movement
Probability for ant k to move from i to j in the next step:
pkij =
⌧↵ij · ⌘�ijP8cil
cil feasible
⌧↵il · ⌘�il
where ↵ and � control importance of pheromone ⌧ vs. heuristic value ⌘
Standard heuristic: ⌘ij = 1dij
where dij is the distance between i and j
Katinka Bohm 6
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Theoretical Approach - Ant System Ant Colony Optimization
Ant System (AS) (cont.)
Pheromone Update
Pheromone update for all ants that have built a solution in that iteration:
⌧ij (1� ⇢) · ⌧ij +mX
k=1
�⌧ kij
where ⇢ is the evaporation rate and �⌧ kij is the quantity of pheromone laid onedge (ij) with
�⌧ kij =Q
L k
where Q is a constant and Lk is the total length of the tour of ant k
Katinka Bohm 7
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Theoretical Approach - Max-Min Ant System Ant Colony Optimization
Max-Min Ant System (MMAS)
I pheromone values are bound
I only the best ant updates its pheromone trails after solutionshave been found
Pheromone Update
⌧ij h(1� ⇢) · ⌧ij +�⌧ bestij
i⌧max
⌧min
where �⌧ bestij = 1Lbest
Lbest can be the iteration best or global best tour
Katinka Bohm 8
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Theoretical Approach - Ant Colony System Ant Colony Optimization
Ant Colony System (ACS)
I diversify the search through a local pheromone update
I pseudorandom proportional rule for ant movement
Local Pheromone Update
Performed by all ants after each construction step to the last traversed edge
⌧ij = (1� ) · ⌧ij + · ⌧0
where 2 (0, 1] is the pheromone decay coe�cient
and 0 is the initial pheromone value
Pheromone Update
⌧ij ((1� ⇢) · ⌧ij + ⇢ ·�⌧ij if (i , j) belongs to the best tour
⌧ij otherwise
Katinka Bohm 9
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Theoretical Approach - Overview Ant Colony Optimization
Theoretical ApproachOverview
Algorithm Ant Movement Pheromones UpdateEvaporation
Ant System (AS)1991
random proportional ⌧ij (1� ⇢) · ⌧ij +Pm
k=1 �⌧kij all paths
Max-Min Ant Sys-tem (MMAX)2000
random proportional
⌧ij h(1� ⇢) · ⌧ij + �⌧bestij
i⌧max
⌧min
best-so-far tourmin/max bound
Ant Colony System(ACS)1997
pseudorandomproportional
local: ⌧ij = (1� ) · ⌧ij + · ⌧0global:
⌧ij ((1� ⇢) · ⌧ij + ⇢ · �⌧ij⌧ij
last stepbest-so-far tour
Katinka Bohm 10
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Problem Types Ant Colony Optimization
Problem Types
I Routing Problems! Traveling Salesman, Vehicle Routing, Network Routing
I Assignment Problems! Graph Coloring
I Subset Problems! Set Covering, Knapsack Problem
I Scheduling! Project Scheduling, Timetable Scheduling
I Constraint Satisfaction Problems
I Protein Folding
Katinka Bohm 11
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Robot Path Planning with ACO Ant Colony Optimization
Robot Path Planning
I NP-complete problemI static vs. dynamic environmentI known vs. unknown environmentI rerouting on collisionI shortest path
Katinka Bohm 12
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Robot Path Planning with ACO Ant Colony Optimization
Robot Path Planning Alg1Mohamad Z. et al. [8]
Shortest Path in a static environmentMap ConstructionGenerate a global free space map where the robot can traversebetween the yellow nodesFree space nodes (white) can be traversed by the robot
Katinka Bohm 13
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Robot Path Planning with ACO Ant Colony Optimization
Robot Path Planning Alg1Mohamad Z. et al. [8]
Ant Movement
Probability
pij = ⌘�ij · ⌧↵ij with ↵ = 5, � = 5
Heuristic ⌘ =1
distance between next point withintersect point at reference line
Katinka Bohm 14
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Robot Path Planning with ACO Ant Colony Optimization
Robot Path Planning Alg1Mohamad Z. et al. [8]
Pheromone Updatelocal ! after each step from one node to anotherglobal ! after path calculation is finished
Local Evaporation
prevents accumulation of pheromone⌧ij = (1� ⇢) · ⌧ij with ⇢ = 0.5
Global Reinforcement (AS)
⌧ij = ⌧ij +Pm
k=1 �⌧kij , �⌧ij = Q
LkwhereQ . . . number of nodes
Lk . . . length of path chosen by ant k
Katinka Bohm 15
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Robot Path Planning with ACO Ant Colony Optimization
Robot Path Planning Alg1Mohamad Z. et al. [8]
Results
I comparison to a standard GA algorithmI ACO faster with smaller number of iterations
(due to good state transition rule - distance to baseline)
Katinka Bohm 16
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Robot Path Planning with ACO Ant Colony Optimization
Robot Path Planning Alg2Michael Brand et al. [2]
Shortest path in a dynamic environment
I grid world of 20x20, 30x30 and 40x40four possible movement directions: left, right, up, down
I basic AS approachI re-routing after obstacles are addedI focus on re-initialization of pheromones
Global Initialization
⌧ij = 0.1 for every transition between blocks
Local Initialization
Gradient of pheromones around every objectPheromone levels are decreased in a cyclic fashion by a certain fraction (50%)
Katinka Bohm 17
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Robot Path Planning with ACO Ant Colony Optimization
Robot Path Planning Alg2Michael Brand et al. [2]
Results
Global Initialization
Map Size 20x20 30x30 40x40
Iterations 151 277 148Path Length 39 66 138
Local Initialization
Map Size 20x20 30x30 40x40
Iterations 122 84 69Path Length 39 64 128
Local Initialization: 1st iteration
Local Initialization: 1000th iteration
Katinka Bohm 18
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Analysis - Comparison Ant Colony Optimization
Comparison to other meta-heuristic techniques
I other techniques:Genetic Algorithms (GA), Simulated Annealing (SA),Particle Swarm Optimization (PSO), Tabu Search (TS)
I hard to compare in general ! dependent on specific probleminstance, algorithm implementation and parameter settings(No free lunch theorem)
I slow convergence compared to other approaches! long runtime for small easy instances and fast, pretty goodresults for complex instances
I ACO often performs really bad or really good
Katinka Bohm 19
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Analysis - TSP Ant Colony Optimization
Traveling Salesman Problem
results for a small TSP instance with 20 nodes over multiple runs
MeasuresACO GA SA PSO TS
Parameters pheromone eva-poration
population,crossover,mutation
temperatureannealing rate
population size,velocity
tabu list length
Convergence slow due to phe-romone evapo-ration
rapid avoids trappingby deteriorationmoves
less rapid tabulist avoidstrappingin local optima
IntensificationDiversification
ant movement,pheromone up-date
crossover,mutation
cooling,solution accep-tance strategy
local search,fitness
tabulist,neighbor selecti-on
CPUTime(s)
250 200 101 220 140
Path Length 300 200 99 250 97
Katinka Bohm 20
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Analysis - Advantages and Drawbacks Ant Colony Optimization
Advantages and Drawbacks
AdvantagesI inherent parallelism
I easy to implement on a basiclevel ! few parameters
I possible to solve NP-hardproblems
I fast in finding near optimalsolutions in comparison toclassical approaches
I robust ! suitable fordynamic applications
Drawbacks
I randomness ! notguaranteed to find theoptimal solution
I slow convergence
I theoretical research is hard! mostly rely onexperimental results
Katinka Bohm 21
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Interesting Applications Ant Colony Optimization
Interesting Applications
Dexterous Manipulation:Gripper Configuration
Determine forces extracted by robotgrippers to guarantee stability of thegrip without causing defect or damageto the object.Non-linear problem containing fiveobjective functions, nine constraintsand seven variables.
Image Processing: Edge Detection
Ants move from one pixel to anotherand are directed by the local variationof the images intensity values stored ina heuristics matrix. The highest densityof the pheromone is deposited at theedges.
Katinka Bohm 22
Universitat Hamburg
MIN-FakultatFachbereich Informatik
Recap Ant Colony Optimization
Recap
I Swarm Intelligence
I Inspired by ant colony movementI Three basic approaches
I Ant SystemI Min-Max Ant SystemI Ant Colony System
I Application: Robot Path PlanningI Shortest path in static environment with free space mapI Shortest path in dynamic environment
I slow convergence but fast good solutions for complex problems
Katinka Bohm 23
Universitat Hamburg
MIN-FakultatFachbereich Informatik
References Ant Colony Optimization
References
[1] Toolika Arora and Yogita Gigras.A Survey of Comarison Between Various Meta-Heuristic Techniques for Path Planning Problem.International Journal of Computer Engineering & Science, 3(2):62–66, November 2013.
[2] Michael Brand, Michael Masuda, Michael Masuda, Nicole Wehner, and Xiao-Hua Yu.Ant Colony Optimization Algorithm for Robot Path Planning.International Conference On Computer Design And Appliations (ICCDA 2010), 3:436–440, 2010.
[3] Marco Dorigo and Thomas Stutzle.Ant colony optimization.MIT Press, 2004.
[4] Marco Dorigo and Thomas Stutzle.International Series in Operations Research & Management Science 146, Ant Colony Optimization: Overviewand Recent Advances, chapter 8, pages 227–263.Springer Science+Business Media, 2010.
[5] Yogita Gigras and Kusum Gupta.Ant Colony Based Path Planning Algorithm for Autonomous Robotic Vehicles.International Journal of Artificial Intelligence & Applications (IJAIA), 3(6), November 2012.
[6] Mauro Birattari Marco Dorigo and Thomas Stutzle.Ant Colony Optimization Artificial Ants as a Computational Intelligence Technique.IEEE Computational Intelligence Magazine, 1556-603X(06):28–39, 2006.
Katinka Bohm 24
Universitat Hamburg
MIN-FakultatFachbereich Informatik
References Ant Colony Optimization
References (cont.)
[7] Seedarla Moses Mullar and R. Satya Meher.Optimizing of Robot Gripper Configurations Using Ant Colony Optimization.International Journal of Engineering Research & Technology (IJERT), 2(9), September 2013.
[8] Buniyamin N., Sari↵ N., Wan Ngah W.A.J., and Mohamad Z.Robot global path planning overview and a variation of ant colony system algorithm.International Journal of Mathematics and Computers in Simulation, 5(1), 2011.
[9] Alpa Reshamwala and Deepika P Vinchurkar.Robot Path Planning using An Ant Colony Optimization Approach: A Survey.International Journal of Advanced Research in Artificial Intelligence, 2(3), 2013.
Katinka Bohm 25