MikaelHammar Bengt J. Nilsson y Mia Persson y Abstra t
Exploringapolygonwithrobots, whentherobotsdonothaveknowledgeofthesurroundings
anbeviewedasanonlineproblem. Typi alforonlineproblemsisthatde isionsmustbemade
basedonpasteventswithout ompleteinformationaboutthefuture. Inour asetherobotsdonot
have ompleteinformationabouttheenvironment. Competitiveanalysis anbeusedtomeasure
the performan eofmethodssolvingonlineproblems. The ompetitiveratio ofsu hamethodis
theratiobetweenthemethod'sperforman eandtheperforman e ofthebestmethodhavingfull
knowledgeof the future. We are interested inobtaining good bounds onthe ompetitiveratio
ofexploringpolygonsandprove onstant ompetitivestrategiesandlowerboundsforexploringa
simplere tilinear polygonintheL
1 metri .
1 Introdu tion
Exploring an environmentis an important and well studied problem in roboti s. In many realisti
situations the robots donotpossess ompleteknowledge aboutthe environment, e.g.,they may not
haveamapofthesurroundings[1,2,4,6,7,8,9℄.
Thesear hoftherobots anbeviewedasanonlineproblem sin etherobots'de isionsaboutthe
sear harebasedonlyonthepartoftheenvironmentthattheyhaveseensofar. Weusetheframework
of ompetitive analysis to measure the performan eof an online sear h strategy S. The ompetitive
ratioofSisdenedasthemaximumoftheratioofthedistan etraveledbytherobotthatmovesthe
farthestusingS to theoptimaldistan eofthesear h.
We are interested in obtaining good bounds for the ompetitive ratio of exploring a re tilinear
polygon. Thesear h ismodeledbyapathor losed tourfollowedbyoneormorepointsizedrobots
inside thepolygon,givenastartingpointfor thesear h. Theonlyinformation that therobots have
aboutthesurrounding polygonisthepartofthepolygonthattheytogetherhaveseensofar.
Forthe ase of exploration with onerobot, Deng et al. [4℄ show adeterministi strategy having
ompetitiveratiotwoforthisproblemifdistan eismeasureda ordingtotheL
1
-metri . Hammaret
al. [5℄ proveastrategy with ompetitiveratio5=3andKleinberg[7℄provesalowerbound of 5=4for
the ompetitiveratioofanydeterministi strategy. Wewillshowadeterministi strategyobtaininga
ompetitiveratioof3=2forsear hingare tilinearpolygonintheL
1
-metri withonerobot.
DepartmentofComputerS ien e,SalernoUniversity,Baronissi(SA)-84081,Italy.
email: hammardia.unisa.it
y
Te hnologyandSo iety,MalmöUniversityCollege,S-20506Malmö,Sweden.
inre tilinearpolygonsand ompetitiveresultsonexplorationwithtwoandthree robots.
Thepaperisorganizedasfollows. Inthenextse tionwepresentsomedenitionsandpreliminary
results. In Se tion 3 we give an overview of the strategy by Deng et al. [4℄. Se tion 4 ontainsan
improvedstrategyfor singlerobot explorationgivinga ompetitiveratioof 3=2. Inse tions5and 6
we onsiderpathexplorationandexplorationwithmultiple robots.
2 Preliminaries
Wewillhen eforthalwaysmeasuredistan ea ordingtotheL
1
metri ,i.e.,thedistan ebetweentwo
pointspandq isdenedby
jjp;qjj=jp x q x j+jp y q y j; wherep x andq x
arethex- oordinatesofpandqandp
y andq
y
arethey- oordinates. Wedene the
x-distan ebetweenpandqtobejjp;qjj
x =jp
x q
x
jandthey-distan etobejjp;qjj
y =jp y q y j.
If C is a polygonal urve, then the length of C, denoted length(C), is dened the sum of the
distan esbetween onse utivepairsofsegmentendpointsin C.
LetPbeasimplere tilinearpolygon. TwopointsinParesaidto see ea hother, orbevisible to
ea hother,iftheline segment onne tingthepointsliesinP. LetpbeapointsomewhereinsideP.
Awat hman routethroughpisdened tobea losed urveC that passesthroughpsu hthat every
pointinPisseenbysomepointonC. Theshortestwat hmanroutethroughpisdenotedbySWR
p .
It anbeshownthattheshortestwat hmanrouteinasimplepolygonisa losedpolygonal urve[3℄.
Sin eweareonly interestedin theL
1
lengthof apolygonal urvewe anassumethat the urve
isre tilinear,thatis,thesegmentsofthe urveareallaxisparallel. Notethattheshortestre tilinear
wat hmanroute throughapointpisnotne essarilyunique.
Forapointpin Pwedene fourquadrants withrespe tto p. Thoseare theregionsobtainedby
uttingPalongthetwomaximalaxisparallellinesegmentsthat passthroughp. Thefour quadrants
are denoted Q 1 (p), Q 2 (p), Q 3 (p), andQ 4
(p) in anti- lo kwiseorder from the toprightquadrantto
thebottomrightquadrant. WeletQ
i;j
(p)denotetheunionofQ
i
(p)andQ
j (p).
Considera reex vertex of P. The two edges of P onne ting at the reex vertex an ea h be
extended inside P until the extensions rea h aboundary point. The segments thus onstru tedare
alledextensions and toea hextensionadire tionisasso iated. Thedire tionisthesameasthatof
the ollinearpolygonedgeaswefollowtheboundaryofPin lo kwiseorder. Weusethefour ompass
dire tionsnorth,west,south, andeast todenotethedire tionofanextension.
Lemma 2.1 (Chin, Ntafos [3℄) A losed urveis awat hmanroute forP ifand only ifthe urve
hasatleastonepointto therightofeveryextensionofP.
Ourrstobje tiveistopresenta ompetitiveonlinestrategythatenablesarobottofollowa losed
urvefromthestartpointsinPandba ktoswiththe urvebeingawat hmanrouteforP.
AnextensionesplitsPintotwosetsP
l andP
r
withP
l
totheleftofeandP
r
totheright. Wesay
apointpistothe leftof eifpbelongstoP
l
. Totheright isdened analogously.
Asafurtherdenitionwesaythatanextensioneisaleft extensionwithrespe ttoapointp,ifp
liestotheleft ofe, andanextensionedominates anotherextensione 0
f=v ext(f) C f 0 (b) ext(f) ( ) q v C p f 0 f 0 prin ipalproje tion s (a) f frontier s f 0 r f 0 l f r fl (d)
Figure1: Illustratingdenitions.
of eare also to therightof e 0
. By Lemma 2.1weare only interestedin the extensionsthat are left
extensions with respe t to the starting point s sin ethe other ones already havea point (the point
s)to the rightof them. Sowithout lossof laritywhen we mentionextensionswewill alwaysmean
extensionsthatareleftextensionswithrespe ttos.
3 An Overview of GO
Considerare tilinearpolygonPthat isnotaprioriknown totherobot. Let sbetherobot'sinitial
positioninsideP. Forthestartingposition softherobotweasso iateapointf 0
ontheboundaryof
Pthatisvisiblefrom sand all f 0
theprin ipal proje tion point ofs. Forinstan e, we an hoosef 0
tobetherstpointontheboundarythatishitbyanupwardraystartingats. Letf betheendpoint
oftheboundarythat therobotseesaswes antheboundaryofPin lo kwise order;seeFigure1(a).
Thepointf is alledthe urrentfrontier.
Let C beapolygonal urvestartingat s. Formallyafrontier f of C is avertexof thevisibility
polygon,VP (C)ofCadja enttoanedgeeofVP (C)that isnotanedgeofP. Extendeuntilithits
apointqonCandletvbethevertexofPthatisrsten ounteredaswemovealongthelinesegment
[q;f℄fromqtof. Wedenotetheleftextensionwithrespe ttosasso iatedtothevertexv byext(f);
seeFigures1(b) and( ).
Dengetal. [4℄introdu eanonlinestrategy alledgreedy-online,GO forshort,toexploreasimple
re tilinearpolygonPintheL
1
metri . IfthestartingpointsliesontheboundaryofP,theirstrategy,
we allitBGO,goesasfollows: fromthestartingpoints antheboundary lo kwiseandestablishthe
rstfrontierf. Moveto the losestpointonext(f)and establishthenextfrontier. Continuein this
fashionuntilallofPhasbeenseenandmoveba ktothestartingpoint.
Deng etal. showthat arobotusing strategyBGOto explore are tilinear polygonfollowsatour
withshortestlength,i.e.,BGOhas ompetitiveratioone. Theyalsopresentasimilarstrategy, alled
IGO,forthe asewhenthestartingpointsliesintheinteriorofP. ForIGOtheyshowa ompetitive
ratio oftwo, i.e., IGO spe ies atour that is at mosttwi e aslongastheshortest wat hmanroute
throughs.
IGOshootsarayupwardstoestablishaprin ipalproje tionpointf 0
andthens anstheboundary
lo kwisetoobtainthefrontier. Next,itpro eedsexa tlyasBGO,movingtothe losestpointonthe
extensionofthefrontier,updatingthefrontier,andrepeatingthepro essuntilallofthepolygonhas
establishingthefrontiers andstill havethesame ompetitiveratio. Hen e,BGO anbeseenastwo
strategies, one s anning lo kwise and the other anti- lo kwise. We antherefore parameterize the
twostrategiessothat BGO(p;orient)isthestrategybeginningatsomepointpontheboundaryand
s anningwithorientationorient whereorient iseither lo kwise w oranti- lo kwiseaw.
SimilarlyforIGO, we annotonly hoose tos an lo kwise oranti- lo kwiseforthefrontierbut
also hoose to shoot the ray giving the rst prin ipal proje tion point in any of the four ompass
dire tionsnorth,west,south,oreast. ThusIGOinfa tbe omeseightdierentstrategiesthatwe an
parameterizeasIGO(p;dir;orient)andtheparameterdir anbeoneofnorth,south,west, oreast.
We further dene partial versions of GO starting at boundary and interior points. Strategies
PBGO(p;orient;region)andPIGO(p;dir;orient;region)applyGOuntileithertherobothasexplored
allof region ortherobot leavestheregionregion. Thestrategiesreturnasresulttheposition of the
robot when it leaves region or when region has been explored. Note that PBGO(p;orient;P) and
PIGO(p;dir;orient;P) arethesamestrategiesasBGO(p;orient)andIGO(p;dir;orient)respe tively
ex eptthattheydonotmoveba ktopwhenallofPhasbeenseen.
4 The Strategy CGO
Wepresentanewstrategy ompetitive-greedy-online(CGO)thatexplorestwoquadrantssimultaneosly
without using uptoomu h distan e. We assume that s lies in the interiorof P sin e otherwisewe
anuseBGO anda hievean optimalroute. The strategyusestwofrontierpointssimultaneouslyto
improvethe ompetitiveratio. However,toinitiatetheexploration,thestrategybeginsbyperforming
a s an of the polygon boundary to de ide in whi h dire tion to start the exploration. This is to
minimizethelossini teduponusbyour hoi eofinitialdire tion.
The initial s an works as follows: onstru t the visibility polygon VP(s) of the initial point s.
Considertheset of edgesin VP (s) not oin idingwith theboundaryof P. Theend pointsof these
edges dene aset of frontier points ea h having an asso iated left extension. Let e denote the left
extension that is furthest from s(distan e beingmeasured orthogonallyto theextension). Letl be
theinnitelinethroughe. Werotatetheviewpointofssothat Q
3 (s)andQ 4 (s)interse tlwhereas Q 1 (s)and Q 2
(s) do not. Hen e, e is a horizontal extension lying below s. Theinitial dire tion of
exploration is upwards throughQ
1
(s)and Q
2
(s). Thetwo frontierpoints used by thestrategy are
obtainedas follows: the left frontier f
l
is established byshooting aray towardsthe left for theleft
prin ipalproje tionpointf 0
l
andthens antheboundaryin lo kwisedire tionforf
l
;seeFigure1(d).
Therightfrontier f
r
isestablishedbyshootingaraytowardstherightfortherightprin ipalproje tion
pointf 0
r
andthens antheboundaryinanti- lo kwisedire tionforf
r
;seeFigure1(d). Toea hfrontier
pointweasso iatealeftextensionext(f
l
)andarightextensionext(f
r
)withrespe ttos.
The strategy CGO, presented in pseudo ode below makes use of three dierent substrategies:
CGO-0, CGO-1,and CGO-2,that ea htakes are ofspe i asesthat ano ur. Subsequentlywe
willprovethe orre tnessand ompetitiveratioforea hofthesubstrategies.
Ourstrategy ensuresthatwheneveritperformsoneofthe substrategiesthisis thelast timethat
the outermost while-loop is exe uted. Hen e, the loop is repeated only when the strategy does not
enteranyof thespe iedsubstrategies. Theloopwill leadthestrategy to followa straightline and
wewillmaintaintheinvariantduringthewhile-loopthat alloftheregionQ
3;4
(p)\Q
1;2
s f l =u (a) s (b) u s ( )
Figure2: Illustratingthekeypointu.
seen.
Wedistinguishfour lassesofextensions. Aisthe lassofextensionsewhosedeningedgeisabove
e,Bisthe lassofextensionsewhosedeningedgeisbelowe. Similarly,Listhe lassofextensionse
whosedeningedgeisto theleftofe,andRisthe lassofextensionsewhosedeningedgeisto the
rightofe. For on iseness,weuseC
1 C
2
asashorthandforthe Cartesianprodu tC
1 C 2 of thetwo lassesC 1 andC 2 .
Wedenetwokeyverti esuandvtogetherwiththeirextensionsext(u)andext(v)thatareusefulto
establishthe orre tsubstrategytoenter. ThevertexuliesinQ
2 (s)andvinQ 1 (s). Ifext(f l )2A[B,
thenuisthevertexissuingext(f
l
)andext(u)=ext(f
l
). Ifext(f
l
)2Landext(f
l
) rossestheverti al
linethroughs,thenuisthevertexissuing ext(f
l
)andagainext(u)=ext(f
l
). Ifext(f
l
)2Ldoesnot
rosstheverti allinethroughs,thenuistheleftmostvertexofthebottommostedgevisiblefromthe
robot,ontheboundarygoingfromf
l
lo kwise untilweleaveQ
2
(s). Theextensionext(u)istheleft
extensionissued byu,and hen e,ext(u)2A; seeFigures 2(a),(b), and( ). The vertexv is dened
symmetri allyin Q
1
(s)withrespe ttof
r .
Ea h of the substrategies is presented in sequen e and for ea h of them we prove that if CGO
exe utes the substrategy, then the ompetitive ratio of CGO is bounded by 3=2. Let FR
s
be the
losedroutefollowedbystrategyCGOstartingataninteriorpoints. LetFR
s
(p;q;orient)denotethe
subpath of FR
s
followedin dire tion orient from point pto pointq, where orient aneither be w
( lo kwise)oraw (anti- lo kwise). Similarly,wedenethesubpathSWR
s
(p;q;orient)ofSWR
s . We
denotebySP(p;q)ashortestre tilinearpathfrom pto qinsideP.
Webeginbyestablishingtwosimplebut usefullemmas.
Lemma 4.1 Ift isapointonsometourSWR
s ,thenlength(SWR t )length(SWR s ): Proof: Sin eSWR s
passesthrought,therouteis awat hmanroute throught. Butsin eSWR
t is
theshortestwat hmanroutethrought,thelemmafollows. 2
Lemma 4.2 LetS beasetofpointsthatareen losedbysometourSWR
s ,andletS 1 =S\Q 1;2 (s), S 2 =S\Q 2;3 (s),S 3 =S\Q 3;4 (s),andS 4 =S\Q 1;4 (s). Then length(SWR s )2max p2S 1 fjjs;pjj y g+2max p2S 2 fjjs;pjj x g+2max p2S 3 fjjs;pjj y g+2max p2S 4 fjjs;pjj x g: Proof: SWR s
en losesallthepointsin S andsin ewe al ulate lengtha ordingto theL
1
metri ,
the smallesttour en losing the pointsis thesmallest re tangle ontaining them. Thelength of the
apointt thatwe anensureispassedbySWR
s
and thateither liesontheboundaryofPor anbe
viewedasto lieontheboundaryofP. Wethen onsider thetourSWR
t
and ompareits lengththe
lengthof FR
s
. ByLemma 4.1 weknowthat length(SWR
t
)length(SWR
s
),hen ethe dieren ein
lengthbetweenFR
s
andSWR
t
isanupperbound onthelossprodu edbyCGO.
Strategy CGO
1 Establishtheexplorationdire tionbyperformingtheinitials anofthepolygonboundary
2 Establishtheleftandrightprin ipalproje tionpointsf 0
l andf
0
r
forQ2(s)andQ1(s)respe tively
3 while Q1;2(s)isnot ompletelyseen do
3.1 Obtaintheleftandrightfrontiers,flandfr
3.2 if f l liesinQ 2 (s)andf r liesinQ 1 (s) then
3.2.1 Updateverti esuandvasdes ribedinthetext
3.2.2 if (ext(u);ext(v))2LRor (ext(u);ext(v))2AR[LAand ext(u) rossesext(v)
then
3.2.2.1 Gotothe losesthorizontalextension
elseif (ext(u);ext(v))2BR[LB or (ext(u);ext(v))2AR[LAand ext(u) doesnot
rossext(v)
then
3.2.2.2 ApplysubstrategyCGO-1
elseif (ext(u);ext(v))2AA[AB[BA[BB then
3.2.2.3 ApplysubstrategyCGO-2
endif
else
3.2.3 ApplysubstrategyCGO-0
endif
endwhile
4 if Pisnot ompletelyseen then
4.1 ApplysubstrategyCGO-0
endif
End CGO
Westart by presenting CGO-0, that does the following: Letp bethe urrentrobot position. If
Q
1
(p) is ompletely seenfrom p then we run PIGO(p;north;aw;P) and move ba k to the starting
points,otherwiseQ
2
(p)is ompletely seenfrom pandwerunPIGO(p;north; w;P) andmoveba k
tothestartingpoints.
Lemma 4.3 IfthestrategyappliessubstrategyCGO-0,thenlength(FR
s
)=length(SWR
s ):
Proof: Assume that CGO-0realizes that when FR
s
rea hesthepointp, then Q
1
(p) is ompletely
seenfrom p. Theother ase,thatQ
2
(p)is ompletelyseenfrom pissymmetri .
Sin ethepathFR
s
(s;p;orient)thatthestrategyhasfollowedwhenitrea hespointpisastraight
line,thepointpisthe urrentlytopmostpointofthepath. Hen e,we anaddaverti alspikeissued
bytheboundarypointimmediatelyabovep,givinganewpolygonP 0
havingpontheboundaryand
furthermore with the same shortest wat hmanroute through p as P. This means that performing
strategyIGO(p;north;orient)inPyieldsthesameresultasperformingBGO(p;orient)inP 0
aboundarypointin P, and orient beingeither w oraw. The tourfollowedistherefore ashortest
wat hmanroute throughthepointpin bothP 0
andP.
Also the point p lies on an extension with respe t to s, by the way p is dened, and it is the
losestpointtossu hthatallofQ
1
(s)hasbeenseenbythepathFR
s
(s;p;orient)=SP(s;p). Hen e,
there isarouteSWR
s
that ontainspand byLemma 4.1length(SWR
p )length(SWR s ). Thetour followedequals FR s =SP(s;p)[SWR p
(p;s;aw);and wehavethat length(FR
s )=length(SWR p ) length(SWR s );andsin eFR s
annotbestri tlyshorterthanSWR
s
theequalityholdswhi h on ludes
theproof. 2
Next wepresentCGO-1. Let uand v be the key verti esas dened earlier. The strategydoes
thefollowing: if(ext(u);ext(v))2LA[LB, we mirrorthe polygon Pat theverti alline throughs
and swapthe namesof uand v. Hen e,(ext(u);ext(v))2AR[BR. We ontinuemovingupwards
updatingf
r
and vuntileither allofQ
1
(s)hasbeenseenorext(v) nolonger rossestheverti alline
throughs.
Ifallof Q
1
(s)hasbeenseenthenweexploretheremainingpartofPusingPIGO(p;east;aw;P),
wherepisthe urrentrobotposition.
Ifext(v)nolonger rossestheverti allinethroughsthenweeitherneedto ontinuetheexploration
bymovingtotherightorreturntouandexploretheremainingpartofthepolygonfromthere.
Ifjjs;pjj y +jjs;ujj x jjs;vjj x
we hooseto returntou. Ifext(u)2AwerunPBGO(u;aw;P)and
ifext(u)2BweusePBGO(u; w;P);seeFigure3. Otherwise,jjs;pjj
y +jjs;ujj x >jjs;vjj x andinthis
asewemovetothe losestpointv 0
onext(v). Bydenition,theextensionofv iseitherinAorBin
this ase.
Ifext(v)2Bthenv=v 0
andwe hoosetorun PBGO(v;aw;P). Otherwise,ext(v)2A. IfQ
1 (v 0 ) isseenfromv 0
thentheentirequadranthasbeenexploredandwerunPIGO(v 0
;east;aw;P)toexplore
the remainder of thepolygon. If Q
1 (v
0
) is notseen from v 0
then there are still things hidden from
therobotinQ
1
(v). WeexploretherestofthequadrantusingPBGO(v 0
;north;aw;Q
1
(v))rea hinga
pointqwherease ondde isionneedsto bemade.
Ifvisseenfromthestartingpointandjjs;qjj
x
jjs;vjj,wegoba ktov andrunPBGO(v;aw;P),
otherwisewerunPIGO(q;east; w;P)fromtheinteriorpointq;seeFigure 5.
Ifv isnotseenfromthestartingpointsthenwegoba ktov andrunPBGO(v;aw;P).
TonishthesubstrategyCGO-1ourlaststepisto returntothestartingpoints.
Lemma 4.4 IfthestrategyappliessubstrategyCGO-1,thenlength(FR
s ) 3 2 length(SWR s ):
Proof: Wehandle ea h aseseparately. Assume fortherst asethat when FR
s
rea hesthepoint
p,thenQ
1
(p)is ompletelyvisible. Hen e,wehavethesamesituationasin theproofof Lemma4.3
andusingthesameproofte hniqueitfollowsthatlength(FR
s
)=length(SWR
s ).
Assumeforthese ond asethatCGO-1de idestogoba ktou,i.e.,thatjjs;pjj
y +jjs;ujj x jjs;vjj x ;
seeFigures3(a)and(b). Thetourfollowedequals oneof
FR s = SP(s;p)[SP(p;u)[SWR u [SP(u;s) SP(s;p)[SP(p;u)[SWR u (u;r; w)[SP(r;s)
wherer isthelastinterse tion pointofFR
s
withthehorizontalline throughs. Usingthatjjs;pjj
y + jjs;ujj x jjs;vjj x
itfollowsthatthelengthofFR
s
in both asesisboundedby
length(FR s ) = jjs;pjj+jjp;ujj+length(SWR u )+jju;sjj = length(SWR u )+2jjs;pjj y +2jjs;ujj x
s FR s SWR u v u r (b) s FR s u SWR u v (a)
Figure3:Illustratingthe asesinLemma4.4whenjjs;pjj
y +jjs;ujj x jjs;vjj x . s v FRs r SWRv u p (a) s FRs r SWRv p v v 0 u (b)
Figure4:IllustratingtheproofofLemma4.4whenjjs;pjjy+jjs;ujjx>jjs;vjjx.
length(SWR s )+jjs;pjj y +jjs;ujj x +jjs;vjj x 3 2 length(SWR s ):
Theinequalitiesfollowfrom theassumptiontogetherwithLemmas 4.1and4.2.
Assume forthethird asethat CGO-1goestotheright,i.e.,that jjs;pjj
y +jjs;ujj x >jjs;vjj x . We
beginby handlingthe dierentsub ases that are independent of whether s sees v; see Figures 4(a)
and(b). Thetourfollowedequalsoneof
FR s = SP(s;v)[SWR v (v;r;aw)[SP(r;s) SP(s;v 0 )[SWR v 0 (v 0 ;r;aw)[SP(r;s) Sin ejjs;vjj x =jjs;v 0 jj x thelengthofFR s
isin bothsub asesboundedby
length(FR s ) length(SWR s )+2jjs;vjj x < length(SWR s )+jjs;pjj y +jjs;ujj x +jjs;vjj x 3 2 length(SWR s );
y x x
indeedseenfroms;seeFigures5(a)and(b). Thetourfollowedin this aseisoneof
FR s = SP(s;v)[SWR v (v;q; w)[SP(q;v)[SWR v (v;r;aw)[SP(r;s) () SP(s;v)[SWR v [SP(v;s)
whereq istheresultinglo ationafterexploring Q
1
(v). Hereweusethatv isseenfroms,and hen e,
thattheinitials anguaranteesthatthereisapointtofSWR
s inQ 3;4 (s)su hthatjjs;tjj y jjs;vjj x , thusFR s isboundedby length(FR s ) = length(SWR v )+2minfjjs;vjj;jjs;qjj x g length(SWR s )+jjs;vjj y +jjs;vjj x +jjs;qjj x < length(SWR s )+jjs;vjj y +jjs;tjj y +jjs;qjj x +jjs;ujj x 3 2 length(SWR s ):
Ontheotherhand,whenvisnotseenfroms,thetourfollowsthepathmarkedwith()above;see
Figure5( ). Thus,thepolygonboundaryobs urestheviewfromstov,andhen e,thereisapointq 0
ontheboundarysu hthat theshortestpath froms to v 0
ontainsq 0
. The pathourstrategyfollows
betweens andv 0
is ashortestpathand we anthereforeassume that italso passedthroughq 0 . We usethatjjs;q 0 jj x jjs;vjj x jjs;qjj x
togetthebound.
length(FR s ) = length(SWR q 0 )+2jjs;q 0 jj x length(SWR s )+jjs;vjj x +jjs;qjj x < length(SWR s )+jjs;vjj y +jjs;ujj x +jjs;qjj x 3 2 length(SWR s ):
TheinequalitiesabovefollowfromLemmas4.1and4.2andthis on ludestheproof. 2
We ontinuetheanalysisbyrstshowingthesubstrategyCGO-2andthenprovingits ompetitive
ratio. Thestrategydoesthefollowing:ifjjs;ujj
x
jjs;vjj
x
thenwemirrorPattheverti allinethrough
salsoswappingthenamesofuandv. Thismeansthatv is losertothe urrentpointpwithrespe t
to x-distan ethanu. Next,go tov 0
, the losestpointonext(v). Ifext(v)2B, run PBGO(v;aw;P)
sin ev=v 0 . Ifext(v)2AandQ 1 (v)isseenfromv 0
thenwerunPIGO(v 0
;east;aw;P). Ifext(v)2A
but Q
1
(v) is not ompletely seenfrom v 0 then we exploreQ 1 (v) using PBGO(v 0 ;north; w;Q 1 (v 0 )). On eQ 1
(v)isexplored wehaverea hedapointqandwemakease ondde ision. Ifjjs;qjj
x
jjs;vjj,
goba ktov andrunPBGO(v;aw;P),otherwiserunPIGO(q;east; w;P). Finallygoba ktos.
Lemma 4.5 IfthestrategyappliessubstrategyCGO-2,thenlength(FR
s ) 3 2 length(SWR s ):
Proof: Assume withoutlossof generalitythatjjs;ujj
x
>jjs;vjj
x
. Theother aseisproved
symmet-ri ally.
Next,assumethat FR
s
passesthroughv; seeFigures6(a),(b), and( ). Thetourfollowedequals
oneof FR s = 8 < : SP(s;v)[SWR v (v;r;aw)[SP(r;s) SP(s;v)[SWR v (v;q; w)[SP(q;v)[SWR v (v;r;aw)[SP(r;s) SP(s;v)[SWR v [SP(v;s)
whereris thelast interse tionpointofFR
s
withthehorizontallinethroughs. ThelengthofFR
s is
inea h asebounded by
length(FR s )=length(SWR v )+2minfjjs;qjj x ;jjs;vjjg:
s FR s r SWRv p v 0 u (a) p v 0 u q FRs s SWRv (b) v s p v v 0 u q SWRv FRs q 0 r ( )
Figure5: IllustratingtheproofofLemma4.4.
Wehavethat minfjjs;qjj x ;jjs;vjjg(jjs;qjj x +jjs;vjj)=2(jjs;qjj x +jjs;vjj y +jjs;ujj x )=2length(SWR s )=4
provingthebound inthis ase. Thelast inequalityfollowsfromLemma 4.2.
If FR
s
doesnotpassthroughv;see Figure6(d); then thetourfollowedequalsFR
s =SP(s;v 0 )[ SWR v 0(v 0
;r;aw)[SP(r;s) where r is the last interse tion point of FR
s
with the horizontal line
throughs. ThelengthofFR
s is length(FR s )=length(SWR v 0 )+2jjs;v 0 jj x length(SWR s )+jjs;v 0 jj x +jjs;ujj x 3 2 length(SWR s ):
Theinequalitiesfollowfrom Lemmas4.1and4.2andthat jjs;v 0
jj
y
0,whi h on ludes theproof. 2
Wehaveprovedthefollowingtheorem.
Theorem 1 CGOis3=2- ompetitive.
5 The Path Problem
Considernowthesituationin whi h we,insteadofa losedtour,wishto obtainashortestpaththat
explorestheinteriorofourre tilinearpolygon,i.e.,thepathfollowedbytherobotdoesnothavetoend
atthestartingpoint. LetOPT P
s
beashortestexplorationpathbeginningats. Itfollowsimmediately
thatlength(OPT P
s
)length(SWR
s
)=2,sin efollowingthepathOPT P
s
toitsendpointandthenba k
s SWR v FRs r v p (a) s SWR v v q FRs p v 0 r jjs;qjjxjjs;vjj (b) s SWR v u FRs q v v 0 p ( ) jjs;qjj x >jjs;vjj s SWR v v FRs p v 0 u r (d)
s p x (b) p s (a) p s ( ) x
Figure7: IllustratingtheproofofTheorem2.
Fromthiswededu ethat thestrategyCGO presentedpreviouslyis 3- ompetitiveforpath
explo-ration. We ontinueto showthatanystrategyforpathexplorationmustbeatleast2- ompetitive.
Theorem 2 Thereisnodeterministi strategyforpathexplorationofare tilinearpolygonthat has
ompetitiveratio2 forany>0.
Proof: We onstru ta ounterexampleasinFigure7. Thestartingpointfortherobotisatthelower
left ornerofthepolygonanditessentiallyseesonlythetwowallsadja enttoit;see Figure7(a).
Therobotnowhastomovetooneoftheextensionsthatitsees. Thesearebothatdistan e1from
thestartingpoint. Assume withoutlossofgeneralitythat itmovesto thehorizontal extension,then
the robot realizesthat there is afurther horizontal extensionat distan e Æ aboveit. Therobot has
theoptionofeither ontinuingupwardsuntilithasvisitedallthehorizontalextensions(ofwhi h the
orresponding frontierpoint is onlyvisible from the previousextension and theextensions are only
separatedbyadistan eofÆ);seeFigure7(b). Thisoptionwillmaketherobotmoveadistan eof1=Æ
upwardsuntilit rea hesthelast extension andmovesto theverti al extensionwhere itrealizesthat
thereisanot hatpointx (whi hliesatdistan eÆbelowthestartingpoints)for ingittomoveba k
downto thispoint. Thetotaldistan emovedis2=Æ+1+Æ.
Theoptimal path is to move to the verti al extension rst, visit the not h at x and then move
upwardsuntil allhorizontal extensionshavebeenvisited, requiringonlyadistan eof1+2Æ+1=Æto
bemoved. Theratiobe omes
2=Æ+1+Æ 1=Æ+1+2Æ =2 Æ+3Æ 2 1+Æ+2Æ 2 2 ifÆ=4<1=4.
Ontheotherhand,iftherobotatsomepointaftermovingtothersthorizontalextensionde ides
tomovetotheverti alextensionitthenrealizesthatithastomoveba ktothenot hatx. Ifitde ides
to ontinueupwardsuntilallhorizontalextensionshavebeenvisitedwehavetheprevioussituation. If
terminatesatpointt;seeFigure7( ). Assume thattherobothasmovedaverti aldistan eofD1
whenitde idestomovedownandvisitthenot hatx. Thetotaldistan emovedisthen3D+1+3Æ
whereastheoptimalpathhaslengthatmostD+1+3Æand theratiobe omes
3D+1+3Æ D+1+3Æ =3 2+6Æ D+1+3Æ 2
ifÆ=2<1=2,thus on ludingtheproof. 2
6 Exploration with Multiple Robots
Wenowlookatthesituationwhenseveralrobotstogetherarerequiredtoexploreare tilinearpolygon.
Againwelookatthetourvariant,i.e.,ea hrobotmustterminatetheaxplorationatthestartingpoint.
Wegiveupperandlowerbounds forthesituationwith twoandthree robotsallstartingat thesame
point. themeasurethatweoptimizeonisthelengthofthelongesttourthatanyoftherobotsfollow.
LetOPT k
s
bethetouroftherobotthatmovesthelongestlengthofallthekrobots. Sin easingle
robot anfollow ea h of thetours that thek robots followand thus getawat hmanroute, wehave
thatlength(OPT k s )length(SWR s )=k.
We prove lower bounds on the ompetitive ratio of any exploration strategy using twoor three
robots.
Theorem 3 Therearenodeterministi strategiesforexploringare tilinearpolygonwithtwoorthree
robotshavingsmaller ompetitiveratiothan3/2.
Proof: Werstshowthelowerbound fortworobotsandlaterextenditforthreerobots. Thelower
boundisbasedonessentiallythesame ounterexamplethat Kleinbergusesforthelowerbound fora
singlerobot[7℄.
Theinitial polygon is given in Figure 8(a)and onsists of asquare with not hes in the orners.
Thelengthofthesides ofthesquareis 2. Ea hof thetworobots hastomoveadistan eof 2before
itgets to a ornerof thepolygon andfurthermore at mosttwo omplete orners anbeseenby the
robots. Hen epla inganot hinoneofthe ornersthatisnotyet ompletelyseenrequiresoneofthe
robots to move6units whereastheoptimal motion anbedonewith only4units, thus proving the
result.
Thesameproofa tuallygoesthroughforthreerobotson eyourealizethat independentlyofhow
thethreerobotsstarttheirexplorationwe anfor eoneofthemtomove6unitspla ingatmostthree
not hesasshownin Figure8(b). 2
Wealso show astrategy fortworobots that has ompetitive ratio2. We allthe strategyTGO
(two-robot GO) sin e it is based on the GO-strategyof Deng et al; [4℄. The two robots ea h run
IGO(s;north;orient),onewithorient = w andtheotherwithorient=aw untilthetworobotshave
seenthe ompletepolygonafterwhi h theybothmoveba ktothestartingpoint.
Theorem 4 The strategy TGO is 2- ompetitive for exploration of a re tilinear polygon with two
s not h (a) s (b) not h
Figure8: IllustratingtheproofofTheorem3.
Proof: Let r bethe interse tion pointof SWR
s
andthe verti al axisissuingfrom s upwards. The
interse tion point r lies at distan e D 0from s; see Figure 9(a). We an viewthe strategyTGO
asrstmovingthe tworobotsfrom s tor and thenseparating, onemoving lo kwise and theother
ounter lo kwise,followingSWR
r
in twodire tionsuntiltherobotshaveseenallof thepolygonand
moveba kto s.
Considernow the robot that movesthe farthest. Lett bethe pointof interse tion between the
robot'swalkandthelastextensionthatitvisitsbeforeitrealizesthat thewholepolygonisexplored.
The robot thenmoves thedistan e L =D+length(SWR
r
(r;t;dir))+length(SP(t;s));where dir is
thedire tionthattherobotmoves.
SupposethatwefollowthetourSWR
s
fromsin thedire tionthat visitsthepointrbeforet. We
an assume that t is a point on SWR
s
sin e it is an interse tion point with an extension. Assume
withoutlossofgeneralitythatthisdire tion is lo kwise. Wethushavethat
L = D+length(SWR r (r;t;dir))+length(SP(t;s)) length(SWR s (s;r; w))+length(SWR s (r;t; w))+length(SWR s (t;s; w)) = length(SWR s ) 2length(OPT 2 s )
whi hprovestheresult. Thattheanalysisistightfollowsfromtheexamplein Figure9(b). 2
7 Con lusions
We havepresented onstant ompetitivestrategies and lowerbounds to explore are tilinear simple
polygonintheL
1
metri withoneormorerobots. Unfortunatelynoneofourresultsaretightsoobvious
open problems are to redu e thegaps betweenthe lowerbounds and the upper bounds. Espe ially
SWRs
t SP(t;s)
s
(a) (b)
Figure9: IllustratingtheproofofTheorem4.
Referen es
[1℄ MargritBetke, Ronald L.Rivest, MonaSingh. Pie emealLearningofanUnknown
Envi-ronment. Ma hine Learning,18(23):231254,1995.
[2℄ K-F. Chan, T. W. Lam. An on-line algorithm for navigating in an unknown environment.
International JournalofComputational Geometry &Appli ations, 3:227244,1993.
[3℄ W. Chin, S. Ntafos. Optimum Wat hman Routes. Information Pro essing Letters, 28:3944,
1988.
[4℄ X. Deng, T. Kameda, C.H. Papadimitriou. Howto LearnanUnknown Environment I:The
Re tilinearCase. Journalofthe ACM,45(2):215245,1998.
[5℄ M.Hammar,B.J.Nilsson,S.S huierer.ImprovedExplorationofRe tilinearPolygons.Nordi
Journalof Computing,9(1):3253,2002.
[6℄ F. Hoffmann, C. I king, R. Klein, K. Kriegel. ThePolygonExplorationProblem. SIAM
Journalon Computing,31(2):577600,2001.
[7℄ J. M. Kleinberg. On-line sear h in a simplepolygon. In Pro . of 5th ACM-SIAM Symp. on
Dis reteAlgorithms,pages815,1994.
[8℄ Aohan Mei, Yoshihide Igarashi. An E ient Strategy for Robot Navigation in Unknown
Environment. Inform. Pro ess.Lett.,52:5156,1994.
[9℄ C.H.Papadimitriou,M.Yannakakis. ShortestPathsWithoutaMap.Theoret.Comput.S i.,