求高手迅速英语翻译成中文
whentherobotsenses,andwhenitmoves,respectively.
Supposetherobotjustsenseds.Markovlocalizationthen
P(ljs)=ffP(sjl)P(l)
whereffisanormalizerthatensuresthattheresultingprob-
abilitiessumuptoone.Whentherobotmoves,Markov
localizationupdatesP(l)
ability:
P(0l)=
usingtheTheoremoftotalprob-
Z
P(0lja;l)P(l)dl
Hereadenotesanactioncommand.Thesetwoupdate
equationsformthebasisofMarkovlocalization.Strictly
speaking,theyareonlyapplicableiftheenvironmentmeets
aconditionalindependenceassumption(Markovassump-
tion),whichspecifiesthattherobot'sposeistheonlystate
therein.Putdifferently,Markovlocalizationappliesonlyto
staticenvironments.
Unfortunately,thestandardMarkovlocalizationap-
proachispronetofailindenselypopulatedenvironments,
sincethoseviolatetheunderlyingMarkovassumption.In
themuseum,peoplefrequentlyblockedtherobot'ssensors,
asillustratedinFigure1.Figurativelyspeaking,ifpeople
lineupasa"wall"infrontoftherobot—whichtheyoften
did—,thebasicMarkovlocalizationparadigmmakesthe
roboteventuallybelievethatitisindeedinfrontofawall.
Toremedythisproblem,RHINOemploysan"entropy
filter"(Foxetal.1998b).Thisfilter,whichisappliedtoall
proximitymeasurementsindividually,sortsmeasurements
intotwobuckets:onethatisassumedtocontainallcor-
ruptedsensorreadings,andonethatisassumedtocontain
onlyauthentic(non-corrupted)ones.Todeterminewhich
sensorreadingiscorruptedandwhichoneisnot,theen-
tropyfiltermeasurestherelativeentropyofthebeliefstate
beforeandafterincorporatingaproximitymeasurement:
P(l)logP(l)dl+P(ljs)logP(ljs)dl
lSensorreadingsthatincreasetherobot'scertainty
(_H(l;s)>0)areassumedtobeauthentic.Allothersen-
sorreadingsareassumedtobecorr