Robot Recognition of Complex Swarm Behaviors
description
Transcript of Robot Recognition of Complex Swarm Behaviors
Robot Recognition of Robot Recognition of Complex Swarm Complex Swarm
BehaviorsBehaviorsAisha Walcott-MAS622J-Dec. 11, 2006Aisha Walcott-MAS622J-Dec. 11, 2006
IntroductionIntroduction
A Swarm is a large collection of autonomous mobile A Swarm is a large collection of autonomous mobile robotsrobots
No centralized controlNo centralized control
Group behaviors are produced from local interactions of Group behaviors are produced from local interactions of many individual robotsmany individual robots
Goal is to develop a suite of primitive global behaviors Goal is to develop a suite of primitive global behaviors that combine to form more complex group programsthat combine to form more complex group programs
Dispersion
Courtesy James McLurkinCourtesy James McLurkin
Orbit
Project GoalProject Goal
0 0.5 1 1.5 2 2.5 30
0.5
1
1.5
2
2.5
3
0 0.5 1 1.5 2 2.5 30
0.5
1
1.5
2
2.5
3
0 0.5 1 1.5 2 2.5 30
0.5
1
1.5
2
2.5
3
0 0.5 1 1.5 2 2.5 30
0.5
1
1.5
2
2.5
3
DisperseDisperse OrbitOrbit ClusterCluster Bubble SortBubble Sort
source
*
Example FeaturesExample Features hi source
low source
source
Build multi-classifiers to classify Complex Swarm BehaviorsBuild multi-classifiers to classify Complex Swarm Behaviors
ApproachApproach
Collect raw behavior data sets Collect raw behavior data sets
Determine Features (8D)Determine Features (8D)
Pattern Recognition AlgorithmsPattern Recognition Algorithms
KNNKNN
Neural NetsNeural Nets
Bayes NetsBayes Nets
Analyze results of each algorithmAnalyze results of each algorithm
KNNKNN
Tested a range of values for nearest Tested a range of values for nearest neighbors random tie breakneighbors random tie break
1 2 3 4 5 6 7 8 9 10 110.52
0.54
0.56
0.58
0.6
0.62
0.64
0.66
k neighbors
perc
ent
corr
ect
KNN for varying k on 4 Swarm robot behaviors
Overall Correct ClassificationOverall Correct Classification Average Class ClassificationAverage Class Classification
Cluster= 100%Cluster= 100%
Disperse = Disperse = 12.5%12.5%
Clump = 50%Clump = 50%
Orbit = 44%Orbit = 44%
Bubble Sort = Bubble Sort = 82%82%
Neural NetsNeural Nets
Single Hidden LayerSingle Hidden LayerLayer 1: nodes [50,70]Layer 1: nodes [50,70]Max percent = 65%Max percent = 65%LogsigLogsig
Two Hidden LayerTwo Hidden LayerLayer 1: nodes [50,70]Layer 1: nodes [50,70]Layer 2: nodes [25,25]Layer 2: nodes [25,25]Max percent = 65%Max percent = 65%
0 5 10 15 20 250.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
number of nodes neighbors
perc
ent
corr
ect
2 Layer NN for with number of units = [50, 70] on 4 Swarm robot behaviors
0 5 10 15 20 250.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
number of nodes
perc
ent
corr
ect
1 Layer NN for with number of units = [50, 70] on 4 Swarm robot behaviors
Bayes NetsBayes Nets
Mapping to discrete domain by Mapping to discrete domain by applying k-means clustering to applying k-means clustering to each featureeach feature
11
Cluster, Disperse,Clump,Orbit, Bubble SortCluster, Disperse,Clump,Orbit, Bubble Sort
22 88
Preliminary ResultsPreliminary ResultsClassification of ClusterClassification of Cluster
Possible bug in codePossible bug in code
Modify the discrete mappingModify the discrete mapping
DiscussionDiscussion
KNN and Neural Net performed wellKNN and Neural Net performed wellDetermining the mapping from real Determining the mapping from real
numbers to a discrete domain may numbers to a discrete domain may affect Bayes Nets classifiersaffect Bayes Nets classifiers
Overall high classification of clustering Overall high classification of clustering behaviorbehavior-Features tuned to behavior-Features tuned to behavior-Not enough variety of samples-Not enough variety of samples
Need more samples of varying Need more samples of varying behaviorbehavior
Next StepsNext Steps
Feature selection-which group of Feature selection-which group of features work best for each classifierfeatures work best for each classifier
Additional experiments to determine Additional experiments to determine why certain classifications are much why certain classifications are much betterbetter
FutureFuture
Use the temporal information to learn Use the temporal information to learn hidden emergent sub-behaviorshidden emergent sub-behaviors
Thank YouThank You