Improved Exploration of Re tilinear Polygons
∗
MikaelHammar†
Bengt J. Nilsson‡
Sven S huierer§
Abstra tWeprovea5/3- ompetitivestrategyforexploringasimplere tilinear polygon inthe
L
1
metri . Thisimprovesthepreviousfa tortwoboundofDeng,KamedaandPapadimitriou.1 Introdu tion
Sear hinginanenvironmentisanimportantandwellstudiedprobleminroboti s. Inmanyrealisti
situations therobot doesnotpossess omplete knowledge aboutitsenvironment,for instan e, it
maynothaveamapofitssurroundings[1,2,5, 6,7,9,8,10,11,12,13,14℄.
Thesear hoftherobot anbeviewedasanon-line problemsin etherobot'sde isionsabout
the sear h are based only on the part of its environment that it has seen so far. We use the
frameworkof ompetitiveanalysis tomeasuretheperforman eofanon-linesear hstrategy
S
[15℄.The ompetitive ratio of
S
is dened as themaximum of theratio of the distan e traveledby arobotusing
S
totheoptimaldistan eofthesear h.Infa t,these sear h problems ome inbasi ally twoavors. Insome aseswewantto sear h
fora targetin the environment[5, 6, 9, 10, 11, 12℄. Thetarget is usuallymodeledasa pointin
theenvironmentthat hasto berea hed by therobot. Anotherproblem typeisthat ofexploring
anenvironment, i.e., withoutaprioriknowledge ofthe environmentletarobot onstru tamap
ofit[7,8,11℄.
Weareinterestedinobtainingimprovedupperboundsforthe ompetitiveratioofexploringa
re tilinearpolygon. Thesear hismodeledbyapathor losedtourfollowedbyapointsizedrobot
insidethepolygon,givenastartingpointforthesear h. Theonlyinformationthattherobothas
aboutits surrounding polygon is thepartof the polygon that it hasseenso far. Deng et al. [7℄
showadeterministi strategyhaving ompetitiveratiotwoforthisproblemifdistan eismeasured
a ordingto the
L
1
-metri . Kleinberg[11℄provesalowerbound of5/4
forthe ompetitiveratio of any deterministi strategy. We will showa strategy obtaining a ompetitive ratio of5/3
forsear hingare tilinearpolygoninthe
L
1
-metri .∗
Thisresear hissupportedbytheDFG-Proje tDiskreteProbleme,No.Ot64/8-1.
†
DepartmentofComputerS ien e,LundUniversity,Box118,S-22100Lund,Sweden.
email:Mikael.Hammar s.lth.s e
‡
Te hnologyandSo iety,MalmöUniversityCollege,S-20506Malmö,Sweden.
email:Bengt.Nilssonte.mah.s e
§
InstitutfürInformatik,AmFlughafen17,Geb.051,D-79110Freiburg,Germany.
naryresults. InSe tion3wegiveanoverviewofthestrategybyDengetal.[7℄. Se tion4 ontains
animprovedversiongivinga ompetitiveratioof
5/3
.2 Preliminaries
Wewillhen eforthalwaysmeasuredistan ea ordingtothe
L
1
metri ,i.e.,thedistan ebetween twopointsp
andq
isdenedby||p, q|| = |p
x
− q
x
| + |p
y
− q
y
|,
where
p
x
andq
x
arethex
- oordinatesofp
andq
andp
y
andq
y
arethey
- oordinates. Wedene thex
-distan ebetweenp
andq
tobe||p, q||
x
= |p
x
−q
x
|
andthey
-distan etobe||p, q||
y
= |p
y
−q
y
|
. IfC
isapolygonal urve,thenthelengthofC
,denotedlength(C)
,isdenedasthesumofthedistan esbetween onse utivepairsofsegmentendpointsin
C
.Let
P
beasimplere tilinearpolygon. TwopointsinP
aresaidtosee ea hother,orbevisibleto ea h other, if theline segment onne ting thepointslies in
P
. Forapointp
inP
wedenoteby
VP(p)
thesetofpointsofP
that arevisiblebytherobotifitstandsonpositionp
.VP(p)
isalledthevisibilitypolygon of
p
. IfC
isa urveinP
,thevisibilitypolygonVP
(C)
ofC
isdenedby
VP(C) =
S
p∈C
VP(p)
.Let
p
be apointsomewhere insideP
. A wat hman route throughp
is dened tobea losedurve
C
that passes throughp
withVP
(C) = P
. The shortest wat hman route throughp
isdenoted by
SWR
p
. It an be shown that theshortest wat hmanroute in asimple polygon is a losedpolygonal urve[3,4℄.Sin e we are only interested in the
L
1
length of a polygonal urve we an assume that the urveisre tilinear,that is,the segmentsof the urveareallaxisparallel. Notethat theshortestre tilinearwat hmanroutethroughapoint
p
isnotne essarilyunique.Forapoint
p
inP
wedenefourquadrants withrespe ttop
. Thoseare theregionsobtainedby utting
P
along thetwo maximal axis parallel line segmentsthat pass throughp
. The fourquadrantsaredenoted
Q1
(p)
,Q2
(p)
,Q3
(p)
,andQ4
(p)
intheorder fromthetoprightquadranttothebottomrightquadrant;seeFigure1(a).
Considerareex vertexof
P
. ThetwoedgesofP
onne tingatthereex vertex anea hbeextendedinside
P
untiltheextensionsrea haboundarypoint. Thesegmentsthus onstru tedarealledextensionsandtoea hextensionadire tionisasso iated. Thedire tionisthesameasthat
ofthe ollinearpolygonedgeaswefollowtheboundaryof
P
in lo kwiseorder;see Figure1(b).We use the four ompass dire tions
north
,west
,south
, andeast
to denote the dire tion of anextension. Thereasonforthissomewhatobs ure denition ishopefully lariedbythefollowing
lemma.
Lemma2.1 (Chin,Ntafos [3℄)A losed urveisawat hmanroutefor
P
ifandonlyifthe urvehasat leastonepointtotherightofeveryextensionof
P
.Ourobje tive isthusto presenta ompetitiveon linestrategy that enablesarobot tofollow
a losed urvefrom thestartpoint
s
inP
andba ktos
withthe urvebeingawat hmanrouteQ1(p)
Q2(p)
Q3(p)
Q4(p)
(a) (b) extensionsp
e
e
′
p
e
dominatese
′
e
isaleft extensionw.r.t.p
( )Figure1: Illustratingdenitions.
Anextension
e
splitsP
intotwosetsP
l
andP
r
withP
l
totheleftofe
andP
r
totheright. We sayapointp
istotheleft ofe
ifp
belongstoP
l
. Tothe right isdenedanalogously.As afurther denition we saythat an extension
e
is aleft extension with respe t to a pointp
, ifp
lies to theleft ofe
, and anextensione
dominates anotherextensione
′
, if allpointsof
P
totherightof
e
arealsoto therightofe
′
;see Figure1( ). ByLemma2.1 weareonlyinterested
intheextensionsthat areleftextensionswithrespe tto thestartingpoint
s
sin etheotheronesalreadyhaveapoint(thepoint
s
)totherightofthem. Sowithoutlossof laritywhenwementionextensionswewillalwaysmeanextensionsthat areleftextensionswithrespe tto
s
.Asapreludetothepresentationofourstrategywegiveashortoverviewofpreviousstrategies.
Inthiswaywe anintrodu e on eptsin theiroriginalsettingthat wewilluse extensivelyin the
sequel.
3 An Overview of GO
Considera re tilinear polygon
P
that is not a priori known to the robot. Lets
be the robot'sinitialpositioninside
P
.Thenext on eptweintrodu eis ru ialtotheworkingsofpreviouslypresentedstrategiesand
isgoingtobeveryimportantforourstrategyaswell. Forthestartingposition
s
oftherobot weasso iateapoint
f
0
ontheboundaryof
P
thatisvisiblefroms
and allf
0
theprin ipal proje tion
point of
s
. For instan e,we an hoosef
0
tobetherstpointontheboundarythatis hitbyan
upwardraystartingat
s
. Assumenowthattherobotmovesalonga urveC
froms
tosomepointp
. Letf
betheendpointthat therobot hasseensofaraswemove lo kwisefromf
0
alongthe
boundaryof
P
;seeFigure2. Thepointf
is alledthe urrentfrontier.Let
C
be a polygonal urve starting ats
. Formallya frontierf
ofC
is a vertexofVP(C)
adja entto anedge
e
ofVP(C)
that isnot anedge ofP
. Extende
until ithits apointq
onC
andlet
v
bethevertexofP
that isrsten ounteredaswemovealongthelinesegment[q, f ]
fromq
tof
. We denotethe left extension with respe t tos
asso iatedto the vertexv
byext(f )
; see Figures2(b)and( ).s
f
0
prin ipalproje tionf
frontierVP(s)
f = v
ext
(f )
ext
(f )
C
( ) (b) (a)q
v
C
p
f
0
f
0
f
Figure2: Illustratingdenitions.
theirstrategy,we all itBGO,goesasfollows.
Strategy BGO
1 Using
s
itselfasprin ipalproje tionpointestablishtherstfrontierf
2p := s
3 while
P
isnot ompletelyseen do3.1 Gotothepointon
ext
(f )
losesttop
,letthispointbethenewpointp
3.2 Using the urrent point
f
onthe boundaryasprin ipal proje tion point establish the newfrontierf
endwhile
4 Taketheshortestpathba kto
s
End BGO
Dengetal.showthatarobotusingstrategyBGOtoexploreare tilinearpolygonfollowsatour
with shortestlength, i.e., BGO has ompetitiveratio one. Theyalso presenta similar strategy,
alled IGO, forthe asewhen the starting point
s
lies in the interior ofP
. ForIGO theyshowa ompetitiveratiooftwo,i.e., IGOspe iesatourthat isat mosttwi e aslongastheshortest
wat hmanroute. IGOdoesthefollowing.
Strategy IGO
1 From
s
shoot aray upwards until aboundarypointf
0
is hit, use
f
0
as prin ipal proje tion
point andestablishtherstfrontier
f
2p := s
3 while
P
isnot ompletelyseen do3.1 Gotothepointon
ext
(f )
losesttop
,letthispointbethenewpointp
3.2 Using the urrent point
f
onthe boundaryasprin ipal proje tion point establish the newfrontierf
endwhile
4 Taketheshortestpathba kto
s
End IGO
ForanexamplerunofstrategyIGOseeFigure3. Thepoints
f
i
,for
1 ≤ i ≤ 19
, arethefrontiersintheorder thattheyareen ounteredalongtheboundary. Thedire ted linesare theextensions
s
f
1
f
4
f
14
f
16
f
19
f
6
f
7
f
3
f
0
f
2
f
5
f
8
f
9
f
12
f
13
f
15
f
11
f
17
f
18
f
10
Figure3: Examplerunofthegreedy-onlinestrategy.
Asis easilyseen,thetwostrategiesdieronlyin therststepwherein one aseweestablish
therstfrontierstartingdire tlyfrom
s
,whereasintheother aseweshootarayinordertondtheprin ipalproje tionpointandfromthereweestablishtherstfrontier.
It is lear that BGO ould just as wells an the boundaryanti- lo kwise insteadof lo kwise
when establishing the frontiers and still have the same ompetitive ratio. Hen e, BGO an be
seen as two strategies, one s anning lo kwise and the other anti- lo kwise. We an therefore
parameterize the twostrategies so that BGO
(p, orient )
is the strategy beginning at some pointp
on the boundary and s anning with orientationorient
whereorient
is either lo kwisecw
or anti- lo kwiseaw
.SimilarlyforIGO,we annotonly hoosetos an lo kwiseoranti- lo kwiseforthefrontierbut
also hoose toshoot theraygivingtherstprin ipalproje tionpointin anyofthefour ompass
dire tionsnorth,west,south,oreast. ThusIGOinfa tbe omeseightdierentstrategiesthat we
anparameterizeasIGO
(p, dir , orient )
andthe parameterdir
anbe oneofnorth
,south
,west
,or
east
.Wefurther dene partial versions of GOstarting atboundaryand interiorpoints. Strategies
PBGO
(p, orient , region)
and PIGO(p, dir , orient , region)
apply GO untileither therobot hasex-ploredallof
region
ortherobotleavestheregionregion
. Thestrategiesreturnasresultthepositionoftherobotwhenitleaves
region
orwhenregion
hasbeenexplored. NotethatPBGO(p, orient , P)
andPIGO
(p, dir , orient , P)
are thesamestrategiesasBGO(p, orient )
andIGO(p, dir , orient )
re-spe tivelyex eptthattheydonotmoveba kto
p
whenallofP
hasbeenseen.4 The Strategy DGO
Wepresentanewstrategydoublegreedyonline(DGO)thatstartsobyexploringtherstquadrant
of
s
,Q1
(s)
,withoutusinguptoomu h distan e. Weassumethats
liesin theinteriorofP
sin eotherwisewe anuse BGO and a hievean optimalroute. Thestrategy uses two frontier points
s
f
0
u
f
0
r
f
r
Figure4: Upperandrightfrontierstogetherwiththeirextensions.
sameastheoneusedinGO,i.e., weshootaraystraightuptogettheupperprin ipalproje tion
point
f
0
u
and s an in lo kwise dire tion forf
u
. Theright frontierf
r
isestablished byshooting araytowardstherightfortherightprin ipalproje tionpointf
0
r
andthens an theboundaryin anti- lo kwise dire tion forf
r
; seeFigure 4. Toea h frontierpointweasso iate aleft extensionext(f
u
)
andext(f
r
)
withrespe ttos
inthesamewayasbefore.ThestrategyDGO,presentedinpseudo odeonPage7,makesuseofvedierentsubstrategies:
DGO-0,DGO-1,DGO-2,DGO-3,andDGO-4,thatea htake areofaspe i asethat ano ur.
Subsequentlywewillprovethe orre tnessand ompetitiveratioforea hofthesubstrategies.
We distinguish four lassesof extensions.
A
is the lass of extensionse
whose dening edgeisabove
e
,B
is the lass ofextensionse
whosedening edge isbelowe
. Similarly,L
isthe lassof extensions
e
whose dening edge is to the left ofe
, andR
is the lass of extensionse
whosedening edge is to the right of
e
. For on iseness, we useC
1
C
2
asashorthand for the artesian produ tC
1
× C
2
ofthetwo lassesC
1
andC
2
.We will ensure that whenever the strategy performs one of the substrategies this is the last
timethattheoutermostwhile-loopisexe uted. Hen e,theloopisrepeatedonlywhenthestrategy
doesnotenter anyofthespe iedsubstrategies. Theloopwill leadthestrategyto followan
x
-y
-monotonepathinside
Q1
(s)
sin ewewill ensurethatext(f
u
)
andext(f
r
)
alwayslieeitherabove or to the right of the pointp
, the urrent position of the robot, in these ases. The strategyadditionallymaintainsthe invariant during the while-loop that the boundary parts of
Q1(s)
totheleftofthe
x
- oordinateofp
andbelowthey
- oordinateofp
havebeenseen.Wewill presentea h ofthe hosensubstrategies in sequen eand forea h of themprovethat
ifDGOexe utesthesubstrategythenthe ompetitiveratioofDGOisbounded by
5/3
. LetFR
s
bethe losedroutefollowedbystrategyDGOstartingataninteriorpoints
. LetFR
s
(p, q, orient )
denotethesubpath ofFR
s
followedin dire tionorient
frompointp
to pointq
,whereorient
an eitherbecw
( lo kwise)oraw
(anti- lo kwise). Similarly,wedenethesubpathSWR
s
(p, q, orient )
ofSWR
s
. Wedenote bySP(p, q)
ashortestre tilinearpathfromp
toq
insideP
.Webeginbyestablishingtwosimplebut usefullemmas.
Lemma4.1 If
t
isapointonsometourSWR
s
,thenlength(SWR
t
) ≤ length(SWR
s
).
Proof: Sin e
SWR
s
passesthrought
,therouteisawat hmanroutethrought
. Butsin eSWR
t
istheshortestwat hmanroutethrought
, thelemmafollows.2
If
f
isafrontierpoint,thenletreg(f )
denote theregiontotherightofext
(f )
.StrategyDGO
Establishtheupperand rightprin ipalproje tionpoints
f
u
[s]
andf
r
[s]
whileQ1(s)
isnot ompletelyseendoObtaintheupperandrightfrontiers,
f
u
andf
r
Case 1:
reg(f
u
) ⊆ reg(f
r
)
Case 1.1:
ext
(f
r
) ∈ A ∪ R
Goto the losestpointon
ext(f
r
)
Case 1.2:
ext
(f
r
) ∈ B ∪ L
ApplysubstrategyDGO-1
Case 2:
reg(f
r
) ⊆ reg(f
u
)
Case 2.1:
ext
(f
u
) ∈ A ∪ R
Goto the losestpointon
ext(f
u
)
Case 2.2:
ext
(f
u
) ∈ B ∪ L
ApplysubstrategyDGO-1
Case 3:
ext
(f
u
)
interse tsext
(f
r
)
in onepointCase 3.1:
(ext(f
u
), ext(f
r
)) ∈ AR
Goto theinterse tionpointbetween
ext
(f
u
)
andext(f
r
)
Case 3.2:
(ext(f
u
), ext(f
r
)) ∈ RB ∪ LA
ApplysubstrategyDGO-1
Case 3.3:
(ext(f
u
), ext(f
r
)) ∈ LB
ApplysubstrategyDGO-4
Case 4:
reg(f
u
)
andreg(f
r
)
aredisjointCase 4.1:
(ext(f
u
), ext(f
r
)) ∈ AA ∪ AR ∪ RR ∪ RA
ApplysubstrategyDGO-2
Case 4.2:
(ext(f
u
), ext(f
r
)) ∈ LR ∪ LA ∪ AB ∪ RB
ApplysubstrategyDGO-3
Case 4.3:
(ext(f
u
), ext(f
r
)) ∈ LB
ApplysubstrategyDGO-4
end while
if
P
isnot ompletelyvisiblethenApplysubstrategyDGO-0
Lemma4.2 If
q
andq
′
arepointsonsometour
SWR
s
lyinginQ1
(s)
,thenlength(SWR
s
) ≥ 2 max(||s, q||
x
, ||s, q
′
||
x
) + 2 max(||s, q||
y
, ||s, q
′
||
y
).
Proof: Sin e
SWR
s
passesthroughq
andq
′
andwe al ulatelengtha ordingto the
L
1
metri , thesmallestpossibletourthatpassesthroughthetwopointsisthere tanglewithatleasttwoofthepoints
s
,q
,andq
′
ontheperimeter. Sin e
q
andq
′
liein
Q1(s)
,thepoints
hastobethelowerleft ornerofthere tangleandthelengthoftheperimeterofthere tangleisthenasstated.
2
The stru ture of the following proofs are very similar to ea h other. In ea h ase we will
establishapoint
t
that we anensureispassedbySWR
s
andthateither liesontheboundaryofP
or anbeviewed asto lieon theboundaryofP
. We then onsider the tourSWR
t
(spe ied bystrategyBGO) and omparehowmu h longer thetourFR
s
is. ByLemma 4.1weknowthatlength(SWR
t
) ≤ length(SWR
s
)
,hen etheratiolength(FR
s
)
length(SWR
t
)
isan upperbound on the ompetitiveratioof strategyDGO.Onequirk of theproofsis that we
requirethat there existsa tour
SWR
t
that ontainsthe startingpoints
in itsinterior oron the boundaryofSWR
t
. ThisrequirementwillberemovedinLemma 4.8.4.1 Strategy DGO-0
WestartbypresentingDGO-0.
Strategy DGO-0
1 Let
p
bethe urrentrobotposition 2 if||s, p||
y
≤ ||s, p||
x
then2.1
r := PIGO(p, east , cw , P)
else
/∗
if||s, p||
y
> ||s, p||
x
then∗/
2.2r := PIGO(p, east , aw , P)
endif
3 From
r
goba ktothestartingpoints
andhaltEnd DGO-0
Lemma4.3 IfthestrategyappliessubstrategyDGO-0,then
length(FR
s
) ≤
3
2
length(SWR
s
).
Proof: Sin ethepath
FR
s
(s, p, orient )
thatthestrategyhasfollowedwhenitrea hespointp
isx
-y
-monotone, the pointp
isthe urrently topmost andrightmost point of thepath. Hen e, we anaddahorizontalspikeissuing fromthe boundarypointimmediatelyto therightofp
,givinganewpolygon
P
′
having
p
on theboundaryand furthermorewith the sameshortestwat hmanroutethrough
p
asP
;seeFigure5. ThismeansthatperformingstrategyIGO(p, east , orient )
inP
yieldsthesameresultasperformingBGO
(p, orient)
inP
′
, p
beingaboundarypointinP
′
, orient
beingeither
cw
oraw
. Thetourfollowedisthereforeashortestwat hmanroutethroughthepointp
in bothP
′
(a) (b)
||s, p||
y
≤ ||s, p||
x
s
p
r
||s, p||
y
> ||s, p||
x
p
s
SWR
p
r
SWR
p
Figure5:Illustratingthe asesintheproofofLemma4.3.
Also the point
p
lies on a left extension with respe t tos
, by the wayp
is dened, and itis the losest point to
s
su h that all ofQ1(s)
has been seen by the pathFR
s
(s, p, orient ) =
SP(s, p)
. Hen e,thereisarouteSWR
s
thatpassesthroughp
andbyLemma4.1length(SWR
p
) ≤
length(SWR
s
)
.Assume nowfortherst asethat DGO-0performsStep2.1,i.e., thatwhen
FR
s
rea hesthe pointp
, then||s, p||
y
≤ ||s, p||
x
; see Figure 5(a). Theresultfor these ond asein whi hDGO-0 performsStep2.2followsbysymmetry. ThetourfollowedequalsFR
s
=
SP(s, p) ∪ SWR
p
(p, r, cw ) ∪ SP (r, s).
(1)Sin ebyourassumption,that
s
doesnotlieoutsideSWR
p
,we anassumewithoutlossofgenerality thatthepointr
isthelast interse tionpointofSWR
p
withtheverti alboundaryaxisofQ1(s)
; seeFigure5(a).We provethis asfollows. Letthe point
r
be asdened in thestrategy DGO-0 and letr
′
be
the last interse tion point of
SWR
p
with the verti al boundary axis ofQ1(s)
. Bythe way the strategyPIGOisdenedthesubpathSP(s, p) ∪ SWR
p
(p, r, cw )
seesallofP
. Hen e,thesubpathSWR
p
(r, p, cw )
equalsSP(r, p)
withr
eitherlyinginQ2
(s)
orQ3
(s)
. Ifr ∈ Q3(s)
thens ∈ SWR
p
,and
r
′
= s
. If
r ∈ Q2(s)
thenwe an hooser
′
sothat||r, s||
y
= ||r
′
, s||
y
,andSP(r, s) = SP (r, r
′
)∪
SP(r
′
, s)
. In both ases
length(SWR
p
(p, r, cw ) ∪ SP(r, s)) = length(SWR
p
(p, r
′
, cw ) ∪ SP(r
′
, s))
.
Thisprovesthatwe an usethepoint
r
′
insteadofthepoint
r
in (1).Wehavethat
length(FR
s
) = ||s, p|| + length(SWR
p
) − ||r, p|| + ||r, s||
≤ length(SWR
s
) + ||s, p|| + ||r, s|| − ||r, p||
= length(SWR
s
) + ||s, p||
x
+ ||s, p||
y
+ ||r, s||
x
+ ||r, s||
y
− ||r, p||
x
− ||r, p||
y
= length(SWR
s
) + ||s, p||
y
+ ||r, s||
y
− ||r, p||
y
≤ length(SWR
s
) + 2||s, p||
y
≤ length(SWR
s
) + ||s, p|| ≤
3
length(SWR
s
),
sin e
length(SWR
s
) ≥ 2||s, p||
byLemma4.2and||s, p||
y
≤ ||s, p||
x
.2
4.2 Strategy DGO-1
NextwepresentDGO-1.
Strategy DGO-1
1 if
ext
(f
u
) ∈ L
then MirrorP
atthediagonalofQ1(s)
endif 2 Gotov
,theverteximmediatelytotherightofthe urrentpoint 3q := PBGO(v, aw , Q1(v))
4 if||s, q||
y
≤ ||s, q||
x
then 4.1 Goba ktov
4.2r := PBGO(v, cw , P)
else/∗
if||s, q||
y
> ||s, q||
x
then∗/
4.3r := PIGO(q, north, aw , P)
endif5 Goba ktothestartingpoint
s
andhaltEnd DGO-1
Lemma4.4 IfthestrategyappliessubstrategyDGO-1,then
length(FR
s
) ≤
3
2
length(SWR
s
).
Proof: Sin ethe point
v
rea hedby pathFR
s
(s, v, orient ) = SP(s, v)
isaboundarypoint, the result of performing strategy BGO(v, orient )
, withorient
being eithercw
oraw
, is a shortestwat hmanroutethrough
v
; seeFigure6.Also the point
v
lies on a shortest path between the horizontal boundary ofQ1(s)
and theextension
ext(f
u
)
. Hen e, there is a routeSWR
s
that passes throughv
and by Lemma 4.1length(SWR
v
) ≤ length(SWR
s
)
.Assume now for the rst ase that DGO-1 performs Steps 4.1 and 4.2, i.e., that when
FR
s
rea hesthepointq
,then||s, q||
y
≤ ||s, q||
x
; seeFigure6(a). ThetourfollowedequalsFR
s
= SP (s, v) ∪ SWR
v
(v, q, aw ) ∪ SP (q, v) ∪ SWR
v
(v, r, cw ) ∪ SP(r, s)
andsin ebyourassumption,that
SWR
v
ontainss
initsinterior,we anassumewithoutlossof generalitythatthepointr
isthelastinterse tionpointofSWR
v
with theverti alboundaryaxisof
Q1(s)
;seeFigure6(a). This anbeprovedin thesamewayasin Lemma4.3. Wehavelength(FR
s
) =
||s, q|| + length(SWR
v
) − ||r, q|| + ||r, s||
≤
length(SWR
s
) + ||s, q|| + ||r, s|| − ||r, q||
=
length(SWR
s
) + ||s, q||
x
+ ||s, q||
y
+ ||r, s||
x
+ ||r, s||
y
− ||r, q||
x
− ||r, q||
y
=
length(SWR
s
) + ||s, q||
y
+ ||r, s||
y
− ||r, q||
y
=
length(SWR
s
) + 2||s, q||
y
≤
3
2
length(SWR
s
),
f
r
s
v
||s, q||
y
> ||s, q||
x
(b)SWR
v
q
FR
s
f
r
FR
s
s
v
r
SWR
v
q
||s, q||
y
≤ ||s, q||
x
(a)p
p
Figure6:Illustratingthe asesintheproofofLemma4.4.
sin e
length(SWR
s
) ≥ 2||s, q||
byLemma4.2and||s, q||
y
≤ ||s, q||
x
.Assume forthese ond asethatDGO-1performsStep4.3,i.e.,on e
FR
s
rea hesthepointq
,then
||s, q||
y
> ||s, q||
x
;see Figure6(b). ThetourfollowedequalsFR
s
= SP(s, v) ∪ SWR
v
(v, r, aw ) ∪ SP (r, s),
with
r
beingthelast interse tionpointofSWR
v
withthehorizontalboundaryaxisofQ1(s)
;see Figure6(b). Inthesamewayasbeforewe annowobtainlength(FR
s
) ≤
length(SWR
s
) + 2||s, q||
x
≤
3
2
length(SWR
s
).
This on ludes theproof.
2
4.3 Strategy DGO-2
We ontinuetheanalysisbyrstshowingthesubstrategyDGO-2andthenprovingits ompetitive
1 Let
e
betheleftmostverti aledgeinQ1(s)
ontheboundarypartseparatingext
(f
u
)
andext
(f
r
)
. Similarly lete
′
bethe bottommosthorizontal edgein
Q1(s)
onthe boundarypart separatingext
(f
u
)
andext
(f
r
)
2 Let
v
bethetopmostvertexone
andletv
′
betherightmostvertexon
e
′
3 if
||s, v||
y
< ||s, v
′
||
x
then3.1 Mirror
P
atthediagonal ofQ1(s)
/∗ v
andv
′
arenowswapped
∗/
endif 4 Gotov
5q := PBGO(v, aw , Q1(v))
6 if||s, v|| ≤ ||s, q||
y
then 6.1r := PIGO(q, north, aw , P)
else/∗
if||s, v|| > ||s, q||
y
then∗/
6.2 Goba ktov
6.3r := PBGO(v, cw , P)
endif7 Goba kto
s
andhaltEnd DGO-2
Lemma4.5 IfthestrategyappliessubstrategyDGO-2,then
length(FR
s
) ≤
5
3
length(SWR
s
).
Proof: Before we begin thea tual proofwehaveto ensure that thepoints
v
andv
′
dened in
substrategyDGO-2 are su h that in all ases it holds that
length(SWR
v
) ≤ length(SWR
s
)
andlength(SWR
v
′
) ≤ length(SWR
s
)
. In asesAA
andAR
,thepointv
liesonsometourSWR
s
sin ev
liesonashortestpathfroms
toext
(f
u
)
andonashortestpathfromext(f
r
)
toext
(f
u
)
. Hen e, forthese two asesitfollowsbyLemma4.1thatlength(SWR
v
) ≤ length(SWR
s
)
;seeFigures7(a) and (b). Furthermore, sin ev
′
lies on some tour
SWR
v
sin ev
′
lies on a shortest path from
ext(f
r
)
toext
(f
u
)
,wehavebythesametokenthatlength(SWR
v
′
) ≤ length(SWR
v
)
,thus provingthese ondinequality.
The ase
RR
followsbya ompletelysymmetri argumentifweex hangethepointsv
andv
′
forea hother;see Figure7( ).
The nal ase
RA
is a little more involved sin e neitherv
norv
′
an be ensured to lie on
a tour
SWR
s
. To prove the inequalities onsider an interse tion pointu
between some tourSWR
s
andext(f
r
)
. The pointu
must exist otherwiseSWR
s
violates the visibility property ofawat hmanroute. Now, byLemma 4.1, wehave that
length(SWR
u
) ≤ length(SWR
s
)
. Sin ev
andv
′
lie onsome shortestpath from
u
toext(f
u
)
we additionallyhavethat there is sometourSWR
u
passingthroughboththesepoints. ApplyingLemma4.1againgivesusthetwoinequalitieslength(SWR
v
) ≤ length(SWR
u
)
andlength(SWR
v
′
) ≤ length(SWR
u
)
,thusprovingour laimalsointhis ase;seeFigure7(d).
Assume fortherst asethat DGO-2performsStep6.1,i.e., that
||s, v||
y
≤ ||s, v
′
||
x
andthatwhen
FR
s
rea hesthepointq
,then||s, v|| ≤ ||s, q||
y
;seeFigure 8(a). Thetourfollowedequalss
ext
(f
u
)
v
v
′
ext
(f
r
)
s
ext
(f
r
)
ext
(f
u
)
v
v
′
(b) ( ) (a) (d)s
ext
(f
r
)
v
v
′
ext
(f
u
)
s
v
ext
(f
u
)
v
′
ext
(f
r
)
p
p
p
p
Figure7: Illustratingthestru tural aseshandledbysubstrategyDGO-2.
Hen e,wehavethatthelengthof
FR
s
isboundedbylength(FR
s
) =
||s, v|| + length(SWR
v
(v, r, aw )) + ||r, s||
≤
length(SWR
v
) + 2||s, v|| ≤ length(SWR
s
) +
4
3
||s, v|| +
2
3
||s, v||
≤
length(SWR
s
) +
4
3
||s, q||
y
+
2
3
||s, v||
x
+
2
3
||s, v||
y
≤
length(SWR
s
) +
4
3
||s, q||
y
+
2
3
||s, v
′
||
x
+
2
3
||s, v
′
||
x
=
length(SWR
s
) +
4
3
||s, q||
y
+
4
3
||s, v
′
||
x
≤
length(SWR
s
) +
2
3
length(SWR
s
) =
5
3
length(SWR
s
),
sin elength(SWR
s
) ≥ 2||s, q||
y
+ 2||s, v
′
||
x
byLemma4.2.Forthese ond aseassumethatDGO-2performsSteps6.2and6.3,i.e.,that
||s, v||
y
≤ ||s, v
′
||
x
andthatwhen
FR
s
rea hesthepointq
,then||s, v|| > ||s, q||
y
;seeFigure8(b). Thetourfollowed equalsFR
s
= SP(s, v) ∪ SWR
v
(v, q, aw ) ∪ SP(q, v) ∪ SWR
v
(v, r, cw ) ∪ SP (r, s).
Withoutlossofgeneralitywe anassumethat
r
isthelastinterse tionpointbetweenSWR
v
and theverti alboundaryaxisofQ1
(s)
. ThelengthofFR
s
isboundedbylength(FR
s
) =
||s, q|| + length(SWR
v
(v, q, aw )) + length(SWR
v
(v, r, cw )) + ||r, s||
=
||s, q|| + length(SWR
v
) − ||r, q|| + ||r, s||
s
v
′
SWR
v
v
||s, v||
y
≤ ||s, v
′
||
x
||s, v|| > ||s, q||
y
(b)q
r
s
q
v
′
SWR
v
v
r
||s, v||
y
≤ ||s, v
′
||
x
||s, v|| ≤ ||s, q||
y
p
p
(a)Figure8:Illustratingthe asesintheproofofLemma4.5.
=
length(SWR
s
) +
4
3
||s, q||
y
+
2
3
||s, q||
y
< length(SWR
s
) +
4
3
||s, q||
y
+
2
3
||s, v||
=
length(SWR
s
) +
4
3
||s, q||
y
+
2
3
||s, v||
x
+
2
3
||s, v||
y
≤
length(SWR
s
) +
4
3
||s, q||
y
+
2
3
||s, v
′
||
x
+
2
3
||s, v
′
||
x
=
length(SWR
s
) +
4
3
||s, q||
y
+
4
3
||s, v
′
||
x
≤
length(SWR
s
) +
2
3
length(SWR
s
) =
5
3
length(SWR
s
),
sin elength(SWR
s
) ≥ 2||s, q||
y
+ 2||s, v
′
||
x
byLemma4.2.This on ludes theproof.
2
4.4 Strategy DGO-3
1 if
(ext (f
u
), ext (f
r
)) ∈ LR ∪ LA
then MirrorP
atthediagonalofQ1(s)
endif2 Let
e
betheleftmostverti aledgeinQ1(s)
ontheboundarypartseparatingext
(f
u
)
andext
(f
r
)
andletv
bethetopmostvertexone
.Letv
′
betheverteximmediatelytotherightofthe urrent
point 3 if
||s, v
′
||
x
≤ ||s, v||
x
then 3.1 Gotov
′
3.2q := PBGO(v
′
, aw , Q1(v
′
))
3.3 if||s, q||
y
≤ ||s, q||
x
then 3.3.1 Goba ktov
′
3.3.2r := PBGO(v
′
, cw , P)
else/∗
if||s, q||
y
> ||s, q||
x
then∗/
3.3.3r := PIGO(q, north , aw , P)
endif else/∗
if||s, v
′
||
x
> ||s, v||
x
then∗/
3.4 if||s, v||
y
≤ ||s, v
′
||
x
then 3.4.1 Gotov
3.4.2q := PBGO(v, aw , Q1(v))
3.4.3 if||s, v|| ≤ ||s, q||
y
then 3.4.3.1r := PIGO(q, north, aw , P)
else/∗
if||s, v|| > ||s, q||
y
then∗/
3.4.3.2 Goba ktov
3.4.3.3r := PBGO(v, cw , P)
endif else/∗
if||s, v||
y
> ||s, v
′
||
x
then∗/
3.5 Gotov
′
3.6r := PBGO(v
′
, aw , P)
endif endif4 Goba kto
s
andhaltEnd DGO-3
Lemma4.6 IfthestrategyappliessubstrategyDGO-3,then
length(FR
s
) ≤
5
3
length(SWR
s
).
Proof: Sin ethepoint
v
′
rea hedbypath
FR
s
(s, v
′
, orient ) = SP(s, v
′
)
isaboundarypointthe result of performing strategy BGO(v
′
, orient )
, withorient
being eithercw
oraw
, is a shortestwat hmanroutethrough
v
′
;seeFigure9.
Also, thepoint
v
′
lieson ashortest pathbetween thehorizontal boundaryof
Q1(s)
and theextension
ext
(f
r
)
. Hen e, there is a routeSWR
s
that passes throughv
′
and by Lemma 4.1
length(SWR
v
′
) ≤ length(SWR
s
)
. The pointv
lies on a shortest path fromext(f
r
)
toext(f
u
)
and therefore we havethat there is aroute
SWR
s
that passes throughv
andlength(SWR
v
) ≤
s
q
r
v
v
′
SWR
v
′
s
||s, v
′
||
x
≤ ||s, v||
x
v
v
′
q
r
||s, q||
y
> ||s, q||
x
p
p
SWR
v
′
||s, v
′
||
x
≤ ||s, v||
x
||s, q||
y
≤ ||s, q||
x
(a) (b)s
v
′
v
q
r
SWR
v
||s, v||
y
≤ ||s, v
′
||
x
||s, v|| ≤ ||s, q||
y
s
||s, v
′
||
x
> ||s, v||
x
v
′
SWR
v
v
r
q
||s, v||
y
≤ ||s, v
′
||
x
||s, v|| > ||s, q||
y
p
p
( ) (d)||s, v
′
||
x
> ||s, v||
x
||s, v
′
||
x
> ||s, v||
x
v
′
q
v
s
r
p
SWR
v
′
||s, v||
y
> ||s, v
′
||
x
(e)Assumefortherst asethatDGO-3performsSteps3.3.1and3.3.2,i.e.,that
||s, v
′
||
x
≤ ||s, v||
x
and
||s, q||
y
≤ ||s, q||
x
; seeFigure9(a). ThetourfollowedequalsFR
s
= SP(s, v
′
) ∪ SWR
v
′
(v
′
, q, aw ) ∪ SP(q, v
′
) ∪ SWR
v
′
(v
′
, r, cw ) ∪ SP (r, s),
with
r
beingthelastinterse tionpointofSWR
v
′
withtheverti alboundaryaxisofQ1(s)
. Inthe asethat thestrategyperformsStep 3.3.3i.e., that||s, v
′
||
x
≤ ||s, v||
x
and||s, q||
y
> ||s, q||
x
; seeFigure9(b). Thetourfollowedequals
FR
s
= SP (s, v
′
) ∪ SWR
v
′
(v
′
, r, aw ) ∪ SP(r, s),
with
r
beingthelast interse tionpointofSWR
v
′
withthehorizontalboundaryaxisofQ1
(s)
. In thesamewayasintheproofofLemma4.4we anobtaina ompetitiveratioof3/2
inthis ase.Assume for the se ond ase that DGO-3 performs Step 3.4.3.1, i.e., that
||s, v
′
||
x
> ||s, v||
x
,||s, v||
y
≤ ||s, v
′
||
x
and||s, v|| ≤ ||s, q||
y
;see Figure9( ). ThetourfollowedequalsFR
s
= SP(s, v) ∪ SWR
v
(v, r, aw ) ∪ SP (r, s),
with
r
being the last interse tion point ofSWR
v
with the horizontal boundary axis ofQ1(s)
. In the ase that the strategy performs Steps 3.4.3.2 and 3.4.3.3, i.e., that||s, v
′
||
x
> ||s, v||
x
,||s, v||
y
≤ ||s, v
′
||
x
and||s, v|| > ||s, q||
y
;see Figure9(d). ThetourfollowedequalsFR
s
= SP (s, v) ∪ SWR
v
(v, q, aw ) ∪ SP (q, v) ∪ SWR
v
(v, r, cw ) ∪ SP(r, s)
with
r
beingthelastinterse tionpointofSWR
v
withtheverti alboundaryaxisofQ1(s)
. Inthe samewayasintheproofofLemma4.5we anobtaina ompetitiveratioof5/3
inthis ase.Assumeforthethird asethatDGO-3performsSteps3.5and3.6,i.e.,that
||s, v
′
||
x
> ||s, v||
x
and
||s, v||
y
> ||s, v
′
||
x
;seeFigure9(e). ThetourfollowedequalsFR
s
= SP (s, v
′
) ∪ SWR
v
′
(v
′
, r, aw ) ∪ SP(r, s),
length(FR
s
) =
||s, v
′
|| + length(SWR
v
′
(v
′
, r, aw )) + ||r, s||
≤
||s, v
′
|| + length(SWR
v
′
) − ||r, v
′
|| + ||r, s|| ≤ length(SWR
s
) + 2||s, v
′
||
x
<
length(SWR
s
) + ||s, v
′
||
x
+ ||s, v||
y
≤
3
2
length(SWR
s
),
sin elength(SWR
s
) ≥ 2||s, v
′
||
x
+ 2||s, v||
y
by Lemma 4.2. From our assumption, thatSWR
v
′
ontains
s
in its interior, we an assume without loss of generality that the pointr
is the lastinterse tionpointof
SWR
v
′
withtheverti alboundaryaxisofQ1(s)
. This anbeprovedin the samewayasinLemma 4.3.This on ludes theproof.
2
4.5 Strategy DGO-4
1 Let
v
betheverteximmediatelyabovep
andletv
′
betheverteximmediatelytotherightof
p
. 2 if||s, v||
y
≤ ||s, v
′
||
x
then 2.1 Gotov
2.2r := PBGO(v, cw , P)
else/∗
if||s, v||
y
> ||s, v
′
||
x
then∗/
2.3 Gotov
′
2.4r := PBGO(v
′
, aw , P)
endif3 Goba kto
s
andhaltEnd DGO-4
Lemma4.7 IfthestrategyappliessubstrategyDGO-4,then
length(FR
s
) ≤
3
2
length(SWR
s
).
Proof: Consider the point
v
. It lies on the shortest path between the horizontal boundaryofQ1(s)
and the extensionext(f
u
)
. Hen e, there is a routeSWR
s
that passes throughv
and byLemma4.1
length(SWR
v
) ≤ length(SWR
s
)
. Bya ompletely symmetri argumentitalsofollows thatthereisarouteSWR
s
thatpassesthroughv
′
,andhen e,
length(SWR
v
′
) ≤ length(SWR
s
)
. Assume rstthat DGO-4performsSteps2.1and2.2,i.e.,||s, v||
y
≤ ||s, v
′
||
x
;seeFigure10(a). ThetourfollowedequalsFR
s
= SP(s, v) ∪ SWR
v
(v, r, cw ) ∪ SP (r, s).
Withoutlossofgeneralitywe anassumethat
r
isthelastinterse tionpointbetweenSWR
v
and theverti alboundaryaxisofQ1(s)
. This anbeprovedinthesamewayasinLemma 4.3. Thelengthof
FR
s
isboundedbylength(FR
s
) =
||s, v|| + length(SWR
v
(v, r, cw )) + ||r, s||
≤
||s, v|| + length(SWR
v
) − ||r, v|| + ||r, s||
≤
length(SWR
s
) − ||r, v|| + ||s, v|| + ||r, s||
≤
length(SWR
s
) + 2||s, v||
y
≤
length(SWR
s
) + ||s, v||
y
+ ||s, v
′
||
x
≤
3
2
length(SWR
s
),
sin elength(SWR
s
) ≤ 2||s, v||
y
+ 2||s, v
′
||
x
byLemma4.2. IfDGO-4performsSteps2.3and2.4,i.e.,||s, v||
y
> ||s, v
′
||
x
;seeFigure10(b);thetourfollowed equalsFR
s
= SP(s, v
′
) ∪ SWR
v
′
(v
′
, r, aw ) ∪ SP (r, s)
andsimilarlywe anassumethat
r
isthelastinterse tionpointbetweenSWR
v
′
andthehorizontal boundaryaxisofQ1(s)
asprovedin Lemma4.3. Inthesamewayasbefore weobtainthatlength(FR
s
) ≤ length(SWR
s
) + 2||s, v
′
||
x
≤
3
s
||s, v||
y
> ||s, v
′
||
x
(b)v
v
′
r
s
||s, v||
y
≤ ||s, v
′
||
x
v
′
v
SWR
v
r
(a)p
p
Figure10: Illustratingthe asesintheproofofLemma4.7.
This on ludes theproof.
2
Inea hoftheproofsofLemmas4.34.7weestablishedakeypoint,fromnowondenoted
t
,inQ1(s)
that ispassed by sometourSWR
s
. IfSWR
t
ontainsthestarting points
in the interiororontheboundarythentheresultsofLemmas4.34.7holdaswehaveshown. Now onsiderthe
asewhen
s
liesoutsideSWR
t
. Weneedto showthat theresultsofthelemmasstillhold inthis ase.Lemma4.8 Usingthenotationabove,theresultsofLemmas4.34.7stillholdif
s
isnot ontainedin
SWR
t
.Proof: Ofallthepossibletours
SWR
t
onsideronethathasapoints
′
visiblefrom
s
andas loseto
s
aspossible. Wewillshowthatlength(SWR
s
) ≥ 2||s, s
′
|| + length(SWR
s
′
)
andthat
length(FR
s
) ≤ 2||s, s
′
|| + length(FR
s
′
).
Thus,wehavethat
length(FR
s
) ≤ 2||s, s
′
|| + length(FR
s
′
) ≤ 2||s, s
′
|| + c · length(SWR
s
′
)
≤ 2c||s, s
′
|| + c · length(SWR
s
′
) ≤ c · length(SWR
s
)
forany
c ≥ 1
su hthatlength(FR
s
′
) ≤ c · length(SWR
s
′
)
. Hen e,we anapplyc = 3/2
orc = 5/3
onea h ofthe asesDGO-0 toDGO-4 appropriatelysin ethepoint
s
′
isbydenition ontained
ontheboundaryof
SWR
t
.Toprovethe twoinequalities,werst assumethat
SWR
t
is ontainedinQ1
(s)
and possiblys
q
s
′
t
SWR
t
(a)s
q
t
(b)SWR
t
s
′
Figure11: IllustratingtheproofofLemma4.8.
symmetri albyamirroringoperationofthepolygon
P
atthediagonalofQ1(s)
andtherefore anbehandledin thesameway.
Next,notethatthepoint
s
′
liesonan
x
-y
-monotonepathfroms
tot
. Thisisbe auses
′
annot
lieabovethe
y
- oordinateoft
andit annotlietotherightofthex
- oordinateoft
. Ifs
′
did,then
itwouldbe possibleto moveit loserto
s
ontradi tingourinitialassumption. (Indeed ifSWR
t
liesinboth
Q1(s)
andQ4(s)
,thens
′
liesonthehorizontalboundaryof
Q1(s)
.)Considera tour
SWR
s
. Sin eSWR
t
has no pointsinQ2(s
′
)
or
Q3(s
′
)
it followsthat these
quadrants annot ontain any verti alleft extensions with respe t to
s
′
. Hen e, itis possibleto
moveany verti aledge of
SWR
s
(partially) visiblefroms
and tothe leftofs
′
towardstheright
so that the new tour onstru ted is a wat hman route, passes through
s
′
, has the same length
as
SWR
s
, and has no verti al edge inQ2
(s
′
)
orQ3
(s
′
)
; see Figures 11(a) and (b). The tour onstru ted in this way onsists of a wat hman route passing throughs
′
and an
x
-y
-monotonepath onne ting
s
ands
′
. Ex hangingthewat hmanroute passingthrough
s
′
foratour
SWR
s
′
possiblyshortensthetotallengthofthetourandthisprovestherstinequality.Toprovethese ondinequalitywenotethat
FR
s
onsistsofanx
-y
-monotonepathfroms
tot
followedbyapartofSWR
t
andapathba ktos
. Hen ewe an(forthisproofonly)assumethat theinitialpathfroms
tot
passesthroughs
′
. Furthermorewenotethat whi heversubstrategyis
appliedfor
FR
s
thesamesubstrategywillbeappliedforFR
s
′
. Howeverwithinasubstrategynot ne essarilythesamestepswillbeperformedforFR
s
andFR
s
′
. Wedenethisdieren eformally asfollows.Considerthetour
FR
p
foranypointp
. IfDGOonp
appliesasubstrategythatdoesoneofthe following:1. itsrst appli ationof PIGOorPBGO iswith anti- lo kwise orientationand the(possible)
subsequentappli ationofPIGOorPBGOisalsowithanti- lo kwiseorientation,
2. its rst appli ation of PIGO or PBGO is with lo kwise orientation and the subsequent
appli ationofPIGOorPBGO iswithanti- lo kwiseorientation,
urrentpointto
t
,applyoneofPIGOorPBGOuntilwerea hapointq
whereade isionismadedepending on whether
||s, q||
y
≤ ||s, q||
x
or||s, q||
y
> ||s, q||
x
. This de isionis pre isely the one thatdetermineswhetherthetourwillbeforwardorientedorba kwardoriented.If both
FR
s
andFR
s
′
are forwardorientedorba kwardoriented,then these ond inequality follows dire tly. Hen e, we have a problem if eitherFR
s
is forward oriented whereasFR
s
′
is ba kwardorientedorthereverse aseo urs. Wedoa aseanalysistoprovethese ondinequality.Assume rstthat
FR
s
is ba kward orientedandFR
s
′
is forwardoriented. Leto
denote the orientationthat therstappli ation ofPIGO orPBGO in DGOons
performs inasubstrategy.Let
¯
o
betheotherorientation,i.e.,ifo = aw
,theno = cw
¯
and onversely.Wehavethat
length(FR
s
) =
||s, s
′
|| + length(FR
s
(s
′
, q, o)) + ||q, t|| + ||t, r|| +
+ length(FR
s
(r, s
′
, ¯
o)) + ||s
′
, s||
=
length(FR
s
′
(s
′
, q, o)) + ||q, s
′
|| + length(FR
s
′
(s
′
, r, o)) + ||r, s
′
|| +
+ 2||s, s
′
|| + ||t, r|| + ||q, t|| − ||q, s
′
|| − ||r, s
′
||
=
length(FR
s
′
) + 2||s, s
′
|| + ||t, r|| + ||q, t|| − ||q, s
′
|| − ||r, s
′
||
≤
length(FR
s
′
) + 2||s, s
′
||,
sin elength(FR
s
(r, s
′
, ¯
o)) = length(FR
s
′
(s
′
, r, o))
,length(FR
s
(s
′
, q, o)) = length(FR
s
′
(s
′
, q, o))
, and||t, r|| + ||q, t|| − ||q, s
′
|| − ||r, s
′
|| =
||t, r||
x
+ ||t, r||
y
+ ||q, t||
x
+ ||q, t||
y
− ||q, s
′
||
x
− ||q, s
′
||
y
− ||r, s
′
||
x
− ||r, s
′
||
y
=
||t, r||
y
+ ||q, t||
y
− ||q, s
′
||
x
− ||q, s
′
||
y
− ||r, s
′
||
x
− ||r, s
′
||
y
=
||r, q||
y
− ||q, s
′
||
x
− ||q, s
′
||
y
− ||r, s
′
||
x
− ||r, s
′
||
y
≤
−||q, s
′
||
x
− ||r, s
′
||
x
− ||r, s
′
||
y
≤
0,
if
o = aw
;seeFigure12(a). Thepenultimateinequalityfollowssin e||s
′
, q||
y
≥ ||r, q||
y
. Ifo = cw
,thenasimilar al ulation aneasilybemade.
Assume nowthat
FR
s
isforwardorientedandFR
s
′
isba kwardoriented. IfSWR
t
interse ts bothQ1
(s)
andQ4
(s)
(thismeansthats
′
liesonthehorizontalboundaryof
Q1
(s)
),weargueasfollows. Sin e
||s, q||
y
= ||s
′
, q||
y
inthis ase,thismeansthat if||s, q||
x
< ||s, q||
y
,then||s
′
, q||
x
<
||s
′
, q||
y
, and hen e, it is notpossible forFR
s
to be forward oriented andFR
s
′
to be ba kward oriented. ThereforeSWR
t
mustbe ompletely ontainedinQ1(s)
.We an furthermoreassume that
s
′
is sele ted sothat
SWR
t
lies ompletely inQ1
(s
′
)
. Wehave,ifthesubstrategyusedisnotDGO-4,that
length(FR
s
)
= ||s, s
′
|| + length(FR
s
(s
′
, q, o)) + ||q, t|| + ||t, r
′
|| +
+ length(FR
s
(r
′
, r, o)) + ||r, s
′
|| + ||s
′
, s||
= length(FR
s
′
(s
′
, q, o)) + ||q, t|| + ||t, r|| + length(FR
s
′
(r, r
′
, ¯
o)) + ||r
′
, s
′
|| +
+ 2||s
′
, s|| + ||r, s
′
|| + ||t, r
′
|| − ||r
′
, s
′
|| − ||t, r||
s
q
FR
s
′
s
q
s
′
s
′
t
r
t
r
r
′
FR
s
′
s
FR
s
′
t
t
′
s
′
( ) (b) (a)Figure12: IllustratingtheproofofLemma4.8.
≤ length(FR
s
′
) + 2||s, s
′
||,
sin elength(FR
s
′
(r, r
′
, ¯
o)) = length(FR
s
(r
′
, r, o))
,length(FR
s
(s
′
, q, o)) = length(FR
s
′
(s
′
, q, o))
, and||r, s
′
|| + ||t, r
′
|| − ||r
′
, s
′
|| − ||t, r|| ≤ 0
by a similar al ulation as above; see Figure 12(b)
for an example of this ase when
o = aw
. The keyinsight is that on e the pointq
is rea hed,thenallof
Q1(s)
that alsoliesinQ1(t)
,Q2(t)
,andQ3(t)
hasalreadybeenseen. Hen eit onlyremainstoexplorethepartin
Q4(t)
.Finally, ifthe substrategyused isDGO-4, then wehave two dierentkeypoints, thepoint
t
rea hedby
FR
s
andthepointt
′
rea hedbyFR
s
′
,givinglength(FR
s
) =
||s, s
′
|| + ||s
′
, t|| + length(FR
s
(t, t
′
, aw )) + ||t
′
, s
′
|| + ||s
′
, s||
=
||s
′
, t
′
|| + length(FR
s
′
(t
′
, t, cw )) + ||t, s
′
|| + 2||s
′
, s|| +
+ ||s
′
, t|| + ||t
′
, s
′
|| − ||s
′
, t
′
|| − ||t, s
′
||
=
length(FR
s
′
) + 2||s
′
, s|| + ||s
′
, t|| + ||t
′
, s
′
|| − ||s
′
, t
′
|| − ||t, s
′
||
=
length(FR
s
′
) + 2||s, s
′
||,
sin elength(FR
s
(t, t
′
, aw )) = length(FR
s
′
(t
′
, t, cw ))
and||s
′
, t|| + ||t
′
, s
′
|| − ||s
′
, t
′
|| − ||t, s
′
|| = 0
;see Figure12( ).This on ludes theproof.
2
Next,weprovethe orre tnessof ourstrategy.
Lemma4.9 IfthestrategyDGOdivertsfroman
x
-y
-monotonepaththenoneofthesubstrategiesDGO-0toDGO-4isperformed.
Proof: Theobje tiveofthisproofistoensurethatall asesaretaken areof. Weenumerateall
thepossible asesandshowthatunlessoneofthesubstrategiesDGO-0toDGO-4isenteredthen
strategywill ontinuealongan
x
-y
-monotonepathinsideP
.Webeginbyassumingthat the urrentpointoftherobotisapoint
p
inQ1(s)
. Furthermore,p
has been seenso far. All these assumptions are true for the starting point, i.e., whenp = s
.Let
ext(f
u
)
andext(f
r
)
betheupperandrightfrontierestablishedfromthepointp
havingtheirissuingedgesinside
Q1(s)
. Thisgivesriseto threepossible ases.Therst aseisthat neither
ext(f
u
)
norext(f
r
)
exist. Thismeansthat thepathfroms
top
sees allofQ1
(s)
and substrategyDGO-0 guaranteesa ompetitive ratioof3/2
in this ase; seeLemma4.3.
The se ond aseis that the extensions
ext(f
u
)
andext(f
r
)
exist butext(f
u
) = ext(f
r
)
, i.e., in essen e we only have one extension. If the extension is inB
orL
, then substrategyDGO-1guarantees a ompetitive ratio of
3/2
in this ase; see Lemma 4.4. On the other hand, if theextensionis in
A
orR
, thenthe robot movestothe losestpointon theextension. This anbedone with an
x
-y
-monotone path moving upwardsand to the rightfrom the urrentpoint thusshowingthatthepathfrom
s
tothenew urrentpointisx
-y
-monotone. Furthermore,everythingin
Q1(s)
to theleftandbelowthenew urrentpointhasbeenseen.The third ase is that the extensions
ext(f
u
)
andext(f
r
)
exist and are dierent. This ase has a number of sub ases. Ifext(f
u
)
is inB
, thenext(f
r
)
must also lie inB
, and hen e, one extension dominatesthe other. Sin eext(f
u
)
lies inB
is symmetri with respe t to amirroring operationofP
along thediagonal ofQ1(s)
toext(f
r
)
liesinL
, wehaveby thesame argument thatoneextension dominatestheotherin this ase. So,ifweassumethat thetwoextensionsdonotdominateea hother andthat theydo notinterse t,wehavethat the pairof extensions an
bein oneofthefollowingnine ases:
(ext(f
u
), ext (f
r
)) ∈ AA ∪ AB ∪ AR ∪ LA ∪ LB ∪ LR ∪ RA ∪ RB ∪ RR.
If
(ext(f
u
), ext(f
r
)) ∈ AA ∪ AR ∪ RR ∪ RA
,thensubstrategyDGO-2guaranteesa ompetitiveratioof
5/3
inthis ase;seeLemma 4.5.If
(ext(f
u
), ext (f
r
)) ∈ LA ∪ LR ∪ RB ∪ AR
,thensubstrategyDGO-3guaranteesa ompetitiveratioof
5/3
inthis ase;seeLemma 4.6.If
(ext(f
u
), ext(f
r
)) ∈ LB
, then substrategyDGO-4 guarantees a ompetitive ratio of3/2
inthis ase;seeLemma4.7.
Next, we assume that the two extensions interse t, hen e they are orthogonal. The pair of
extensions anbein oneofthefollowingve ases:
(ext (f
u
), ext(f
r
)) ∈ AR ∪ LA ∪ LB ∪ RA ∪ RB.
The ase
(ext (f
u
), ext (f
r
)) ∈ RA
annoto urwiththetwoextensionsinterse tingleavinguswiththefourremaining ases.
If
(ext(f
u
), ext (f
r
)) ∈ LA ∪ RB
,thensubstrategyDGO-1guaranteesa ompetitiveratioof3/2
inthis ase;seeLemma 4.6.
If
(ext(f
u
), ext(f
r
)) ∈ LB
, then substrategyDGO-4 guarantees a ompetitive ratio of3/2
inthis ase;seeLemma4.7.
If
(ext(f
u
), ext(f
r
)) ∈ AR
, thenby ourindu tive assumption we anmovethe urrent pointwithan
x
-y
-monotonepathtotheinterse tionpointoftheextensions. Everythingtotheleftandbelowtheinterse tionpointisseenbyan
x
-y
-monotonepathfroms
to theinterse tionpoint.hasbeenseensofar.
This on ludes the aseanalysisandtheproof.
2
Lemmas4.34.9leadustoestablishthetotal ompetitiveratioofDGO.Wehavethetheorem.
Theorem 1
length(FR
s
) ≤
5
3
length(SWR
s
).
Proof: It only remains to prove the orre tness of our strategy, i.e., that on e the strategy
terminatesthe omplete polygonhas been explored. But this followsfrom Lemma 2.1sin e the
strategyensuresthat
FR
s
hasatleastonepointto therightofeveryextension.2
5 Con lusions
We haveproved a
5/3
- ompetitivedeterministi strategy alled DGO for exploring a re tilinearpolygonin the
L
1
metri . Weanti ipatethatwithasimilarmethodthe ompetitiveratioshould bepossibletobeimprovedto3/2
althoughthedetailsareyettobeironedout. ThisnewstrategymakesextensiveuseofthestrategyDGOanditssubstrategiesDGO-0to DGO-4.
Closing thegapto theknown lowerboundof
5/4
is stillanopenproblem. Wehopethat ourmethodwillgivenewinsightsothatthegap anbenarrowedfurtheroreven losedaltogether.
Referen es
[1℄ Margrit Betke, Ronald L. Rivest, Mona Singh. Pie emealLearningof anUnknown
Environment. In Sixth ACM Conferen e on Computational Learning Theory (COLT 93),
pages277286,July1993.
[2℄ K-F. Chan, T.W. Lam. An on-linealgorithmfor navigatinginanunknownenvironment.
International Journalof ComputationalGeometry&Appli ations,3:227244,1993.
[3℄ W. Chin, S.Ntafos. OptimumWat hmanRoutes. InformationPro essingLetters,28:39
44, 1988.
[4℄ W. Chin, S. Ntafos. Shortest Wat hmanRoutes in SimplePolygons. Dis reteand
Com-putational Geometry,6(1):931,1991.
[5℄ A.Datta,Ch.Hipke,S.S huierer.CompetitiveSear hinginPolygonsBeyond
General-izedStreets.InPro .SixthAnnualInternationalSymposiumonAlgorithmsandComputation,
pages3241.LNCS1004,1995.
[6℄ A. Datta,Ch.I king. CompetitiveSear hinginaGeneralizedStreet. InPro . 10thAnnu.
ACM Sympos. Comput.Geom.,pages175182,1994.
[7℄ X. Deng,T.Kameda, C.H.Papadimitriou. HowtoLearnanUnknownEnvironment. In
Polygon.InPro eedingsoftheEighthAnnualACM-SIAMSymposiumonDis reteAlgorithms,
pages166174,1997.
[9℄ Christian I king, Rolf Klein. Sear hing for the Kernel of a Polygon: A Competitive
Strategy. InPro . 11thAnnu.ACMSympos.Comput. Geom., pages258266,1995.
[10℄ R. Klein. Walkinganunknownstreetwithboundeddetour. Comput.Geom. Theory Appl.,
1:325351,1992.
[11℄ J. M. Kleinberg. On-line sear hin asimple polygon. InPro . of 5th ACM-SIAM Symp.
on Dis reteAlgorithms,pages815,1994.
[12℄ A. López-Ortiz, S. S huierer. GoingHomeThroughan Unknown Street. InS. G.Akl,
F.Dehne, J.-R.Sa k,editors,Pro . 4thWorkshoponAlgorithms andDataStru tures,pages
135146.LNCS955,1995.
[13℄ AohanMei, YoshihideIgarashi.AnE ientStrategyforRobotNavigationinUnknown
Environment. Inform. Pro ess.Lett.,52:5156,1994.
[14℄ C. H. Papadimitriou, M. Yannakakis. Shortest paths without a map. In Pro . 16th
Internat.Colloq.AutomataLang.Program.,volume372ofLe tureNotesinComputerS ien e,
pages610620.Springer-Verlag,1989.
[15℄ D. D. Sleator,R. E.Tarjan. Amortizede ien yoflistupdateandpagingrules.