Fast computational processing for mobile robots' self-localization

Fas compu a ional p ocessing o mobile obo s’ sel -localiza ion Hélde Ribei o, Ped o Sil a, Rica do Ro iz, Tiago Maia, Rui Sa ai a, Gil Lopes and A.Fe nando Ribei o Labo a ó io de Au omação e Robó ica, G upo de Con olo Au omação e Robó ica Depa men o Indus ial Elec onics, Uni e si y o Minho Azu ém, Guima ães, Po ugal { a58795, a68541, a68536, a57126, a58783}@alunos.uminho.p , {gil, e nando}@dei.uminho.p Abs ac —This pape in ends o p esen a di e en app oach o sol e he Sel -Localiza ion p oblem ega ding a RoboCup’s Middle Size League game, de eloped by MINHO eam esea che s. The me hod uses whi e ield ma kings as key poin s, o compu e he posi ion wi h leas e o , c ea ing an e o -based g aphic whe e he minimum co esponds o he eal posi ion, ha a e compu ed by compa ing he key (line) poin s wi h a p ecompu ed se o alues o each posi ion. This app oach allows a e y as local and global localiza ion calcula ion, allowing he global localiza ion o be used mo e o en, while d i ing he es ima e o i s eal alue. Di e en ly om he majo i y o o he eams in his league, i was impo an o come up wi h a new and imp o ed me hod o sol e he adi ional slow Sel -Localiza ion p oblem. Keywo ds—RoboCup; MSL; Middle Size League; MINHO eam; Sel -Localiza ion; Localiza ion; I. INTRODUCTION MINHO eam s a ed a obo ic oo ball eam in 1997 and has been pa icipa ing on RoboCup scien i ic challenge since 1999 making imp o emen s on hei pla o ms and so wa e. In 2011 ha de elopmen paused and e u ned in 2014. The es a consis ed in ebuilding bo h ha dwa e and so wa e, and he e was an u gen need o imp o e he obo s sel - localiza ion echnique, and o push he de elopmen o a new me hod. Rega ding he RoboCup MSL speci ic applica ion au onomous obo s need o know hei posi ion on he ield ( he wo ld), o be able o mo e o a ce ain posi ion, and o kick owa ds a ce ain di ec ion o o pe o m high le el agen coo dina ion, bu ha implies he need o he exis ence o a me hod ha allow he obo o sel -localize, only using local on-boa d senso s. The obus ness, p ocessing ime and alse posi i es in he line poin de ec ion a e a big conce n, and all o hem model he s uc u e and p ocedu e o he me hod. A he momen he majo i y o eams use he sel -localiza ion algo i hm c ea ed by B ains o me s T ibo s [1]. The me hod he e p esen ed ies o imp o e he compu a ional ime using a a he di e en app oach in he e o calcula ion p ocedu e. The me hod uses a p ecompu ed se o meaning ul dis ances o e e y possible posi ion in he 20x14m ield a ea (s anda d 18x12 plus 1 me e all a ound he ield, esul ing in a 20x14m wo ld), wi h a 10x10cm esolu ion. Then, gi en he obo ’s ue o ien a ion, he posi ion wi h he leas e o is compu ed using an e o modelling unc ion, coming up wi h he ue posi ion o he obo on he ield. In Sec ion II, a acking o he league’s de elopmen is gi en, while in Sec ion III he imaging solu ion (also a s anda d in he league) and he me hod used in he sea ch o in e es poin s in he image, is p esen ed. Sec ion IV p esen s he me hod i sel , explaining he wo ld iew om a ce ain poin in he ield, and he e o calcula ion p ocedu e. Sec ion V explains he wo di e en me hods o acqui ing he obo ’s ue heading, in ela ion o a known e e ence, one by ha dwa e (wi h i s complemen a y so wa e) and he o he by so wa e. Sec ion VI add esses he esul s achie ed du ing he esea ch and applica ion o he me hod, concluding his wo k. II. LEAGUE’S DEVELOPMENT The league e ol ed apidly h oughou he yea s, accomplishing new challenges and asks, complying wi h he cu en ad ances o mode n day compu ing echnologies. Wi h new came a echnologies, new compu e s and be e communica ions, he imaging quali y, he p ocessing powe and he in o ma ion sha ing eloci y has been g ea ly imp o ed. Due o hese ac s, he league s epped up new challenges o he eams, making changes in he ules and in he composi ion o he ield. As he global sys em is as e , he necessi y o ha ing obus , e ec i e and as algo i hms ( o accomplish di e en asks) o i he “p ocessing ime window” is u ging. Rega ding he ield layou , i is now la ge , wi h 18x12 me e s playable a ea, only wi h he s anda d whi e line ma kings. Taking in o accoun ha he obo ’s ca adiop ic came a sys em only co e s abou 4 me e s adius o ield a ea, he e is a lo o in o ma ion missing, gi ing a lo mo e impo ance o eam communica ion and agen coo dina ion, o accomplish eam and ac ical objec i es o he game. S a ed ha he league, and he whole game, is e ol ing as , he de elopmen o he p esen me hod, ep esen s he e olu ion o a well-se led algo i hm, o imp o e, a leas , he compu a ional ime in ol ed in he sel -localiza ion p ocess. III. IMAGING SOLUTION AND POINTS OF INTEREST A. Ca adiop ic Senso Se up When i comes o imaging, he e is a s anda d in he league, o use a ca adiop ic senso , which is ob ained by pai ing a ca adiop ic mi o and a came a, wi h a ious echnologies, ou pu ypes and p ice anges. The ca adiop ic mi o [2] [3] was de eloped speci ically o his applica ion, using simula ion ools o achie e bes pe o mances. The image p o ided by his imaging se up a i es a a equency o 30Hz, wi h a c opped esolu ion o 480x480, in YUV 411 o ma . Using came a’s se up ool, pa ame e s can change se up o o e come some ligh ing p oblems, achie ing a good and s able image quali y ha is u he ahead used in he calib a ion p ocess, o wo ld pa ame e s (like ball and line size e sus dis ance) and colou labelling h ough colou segmen a ion. Figu e 1 - Ca adiop ic Mi o and image cap u ed by he ca adiop ic senso B. Image Labelling and line poin ex ac ion The colou s in he image a e segmen ed using a 16MB YUV Look up Table [4] ha is s o ed in memo y. Du ing he calib a ion p ocedu e, he RGB space is labelled using he YUV colou space, allowing o co ec ly iden i y he na u e o each pixel. To ex ac ea u es used in he sel -localiza ion algo i hm, ins ead o analysing all o he 230400 pixels, pe o ming Hough-T ans o ms o o he line de ec ing algo i hms ha in ol e Canny-Edge de ec o , scan lines a e used. Fo iden i ying he lines, using key line poin s, 72 adial scan lines (spaced by 5º) and 72 spi al scan lines [5] (36 in one di ec ion, 36 in he opposi e di ec ion, spaced by 10º) co e a la ge and meaning ul a ea o he image, using only 63378 pixels, which a e 27.5% o he image. Inspi ed on he ASML Falcons his eam decided o use spi al scan lines because hey gua an ee ha poin a e ound when a obo is on op o a line, while he adial lines migh no see hem. This sea ching p ocess usually akes 5 milliseconds o un, unde an In el Pen ium 3805U, lea ing 25 milliseconds in he “p ocessing ime window”, o be occupied by o he p ocesses. A e de ec ing ield-line ansi ions, i is possible o analyse he de ec ed line poin s in he image. Figu e 2 - De ec ion o in e es poin s using adial and spi al scan lines. The ield shown in his example is hal he size o an o icial MSL ield. Each pixel is hen mapped using a non- linea ans o ma ion, speci ic o each ca adiop ic senso se up ha is calib a ed be o ehand. The p oposed me hod al eady b ings he ad an age ha no line poin s need o be disca ded, because, when using he me hod ha is mos widely used in he league, he e is si ua ions when poin s ha e o be disca ded due o he excessi e inc ease in p ocessing ime. Despi e he ac ha , in he p oposed me hod, he p ocessing ime also inc eases wi h he numbe o de ec ed line poin s, he inc ease is no signi ican , as he p ocessing ime pe line poin is e y small. In he nex sec ion, i s i is explained he ield model buil o compu e he e o , hen, i is p esen ed he me hod o compu e he e o , gi en a ce ain se o line poin s, in a ce ain ime s ep. IV. CALCULATING THE ROBOT’S POSITION As any o he sel -localiza ion me hod, i is equi ed o p ecompu e a “map”, o o say, a poin o iew om a localiza ion poin . Gi en he me hod p oposed by T ibo s, hei poin o iew is he nea es dis ance o he closes whi e ma king. Ins ead, he p oposed me hod de ines a se o dis ances, acqui ed by he same adial and spi al scan lines. Each adial scan line can p o ide up o 4 dis ances, gi en he maximum dis ance ha he ca adiop ic senso can de ec , and each spi al scan line can p o ide only one dis ance. This dis ances a e he dis ances whe e he scan lines encoun e whi e line ma kings. Th ee subsec ions a e p esen ed he e, as hey a e he co e p ocedu es o ob ain he obo localiza ion on he ield. A. Building he Field Map As b ie ly explained in he p e ious chap e , in o de o be possible he use his algo i hm, i was necessa y o c ea e a i ual ield model capable o ca ying ou de ec ion poin ope a ions. The model has been designed o ma ch as much as possible he so wa e ha will un in he obo in eal ime. So i was c ea ed a model o he ield, ully con igu able using he P ocessing de elopmen IDE. This model is capable o de ec ing he line in e sec ion poin s equally o ha desc ibed in sec ion III, i.e. ough 72 adial scan lines (spaced by 5º) and 72 spi al scan lines (36 in one di ec ion, 36 in he opposi e di ec ion, spaced by 10º). I is also in oduced in he model an o se om he cen al poin o sea ch, wi h he adius co esponding o he space occupied by he obo in o de o c ea e a close app oxima ion o eali y. Fi s he model pe o ms a sea ch o in e sec ion poin s using he adial lines. I consis s o a ansi ion om he RGB colo (0,255,0) co esponding o he ield o a RGB colo (255,255,255) co esponding o he lines. Fo ha adial line c ea ion and sea ch is used he ollowing equa ions: x = xi + Incpoin *cos(Angle * π/180) (1) x = xi + Incpoin *cos(Angle * π/180) (2) Whe e:  xi and yi make he cen al sea ch poin .  x and y make he nex sea ch poin .  Incpoin de ine he dis ance o nex sea ch poin  Angle de ine he angle o he adial (spaced by 5º) A e his, he sea ch is done by spi al lines using he ollowing equa ions: x = xi + MASK*cos(Spi alAngle * π/180) (3) y = yi + MASK*sin(Spi alAngle * π/180) (4) Whe e:  xi and yi make he cen al sea ch poin .  x and y make he nex sea ch poin .  MASK simula es he obo adius.  Spi alAngle de ines he angle o he nex spi al poin In bo h sea ch me hods i is de ined a sea ch dis ance limi o each one o he lines, which is he maximum obo isualiza ion dis ance as shown in Figu e 3. Impo an o no ice ha he dimensions o his ield model co esponds o hal o he o icial ield size. Figu e 3 - Vi ual poin s’ ex ac ion In o de o ge all he in e sec ion poin s a ull scan o he ield is ca ied ou , wi h inc emen s o 10cm. A he end, a ile is c ea ed wi h all hese alues om he adial and spi al lines p ope ly di ided, in o de o acili a e i s use in he e o compu ing me hod desc ibed in he ollowing sec ion. B. Compu ing and Modelling he E o A he beginning o he e o compu a ion algo i hm he p e iously c ea ed ile is ead in o memo y and s o ed in he o m o s uc u es o acili a e and inc ease calcula ion speed, be ween he eal poin s acqui ed by he obo and he ones con ained in he ile. Fo his compa ison, wo undamen al aspec s a e aken in o accoun . One is ha he i ual pa is mo e accu a e in de ec ing he in e cep ions poin s in compa ison o he segmen a ion algo i hm in eal ime, i.e. o each o he lines used in he i ual sea ch he e is a high p obabili y o de ec ing a leas one ansi ion. Bu , in eal ime on he obo ision algo i hm, he de ec ed ansi ion poin s will be lowe . This occu s because he colo segmen a ion algo i hm is no pe ec due o b igh ness changes in he ield o he impe ec ions in he mi o , o e en due o dynamic obs acles p esen on he playing ield (o he obo s o humans), no all he sea ch lines will e u n ansi ion poin s. So he e o compu a ion algo i hm igno es ows ha do no mee ansi ion poin s. Ano he e y impo an aspec o conside is he con idence ha is gi en o he dis ance calcula ed om he ansi ion poin o he obo cen e image. The ansi ion poin s de ec ed in sho e dis ances will ha e highe weigh in he calcula ion e o han poin s de ec ed a a la ge dis ance om he cen e o he image (cen e o he obo ). The weigh assigned o each compa ison poin is calcula ed based on he ollowing equa ion and g aphic: Weigh = 1-(c2)/(c2+e2) (5) Whe e e is he e o and c is he alue o e whe e he weigh becomes cons an . Figu e 4 - E o modelling unc ion A e unning h ough he posi ions ha a e being sea ched, ca ying ou a local o global sea ch, he cu en posi ion o he obo is he one ha minimizes he e o e m, being he e o e m, he accumula ion o he e o be ween he ideal (model) dis ances, and he dis ances p o ided by he de ec ed line poin s. A e an es ima ion om he ision sys em, i is necessa y o use i wi h ha dwa e da a, p e e ably, odome y. C. Co ec ing he es ima es using Kalman Fil e ing Kalman Fil e ing [6] [7] is o su e one o he mos used p edic ion and co ec ion mechanisms, in he whole ield o enginee ing. I s capabili ies o p edic he s a e o a sys em, gi en i s pas and i s ma hema ical model, and co ec ha es ima e using senso s eedback, makes i one o he mos amous me hods in enginee ing o acking, p edic ion, e en used in ai planes and hei missiles o ack and chase hei a ge s, based on hei measu es on i and he ma hema ical model o a mo ing objec . Two indi idual Kalman Fil e s mus be compu ed, hen a me ging Kalman Fil e en e s in place, building an es ima ion o he obo posi ion ha is based bo h on so wa e ( ision) and ha dwa e (odome y). Fi s , he sepa a e Kalman Fil e s shall be p esen ed, concluding his subsec ion wi h he usion algo i hm o he indi idual Kalman Fil e s. 1) Indi idual Fil e o Co ec Vision Es ima es When he ision sys em ou pu s an es ima e o he obo ’s cu en posi ion, i has o be co ec ed be o e i is me ged wi h he odome y. Gi en he ou pu sen o he omnidi ec ional mo o con olle o he obo , applying he omnidi ec ional ma hema ic model, an es ima e o he posi ion o he obo ela i e o i s p e ious posi ion is compu ed. Then, using s anda d Kalman Fil e ing equa ions, he ision es ima e ( ha ac s as he senso componen ) co ec s he es ima e, yielding a be e es ima i e o s a . 2) Indi idual Fil e o Co ec Odome y Es ima es In he same manne ha he ision es ima e is he sensed co ec ing componen , he odome y also is he sensing componen . Using he communica ion lines o he ha dwa e modules, one can e ie e he alues gi en by he encode s, pe o ming ma ix calcula ions, con e ing mo o eloci y (gi en by encode icks e sus ime) in o obo ’s angula and linea eloci ies, ha ing also i s di ec ion o mo emen . Again, using he omnidi ec ional ma hema ical mo emen s and he inpu s gi en o he mo o con olle , i is possible o p edic he s a e o he sys em, ela i e o he p e ious posi ion and hen co ec ing i wi h he odome y measu es. This indi idual Kalman Fil e ing p ocedu e, assu es ha he alues inpu o he usion algo i hm a e he bes ones possible, as applying Kalman Fil e ing yields be e esul s han any o he sou ces alone, meaning ha , he ou pu o he il e is mo e eliable and us wo hy han he ou pu o he ma hema ical model o he senso s ( ision and odome y) alone. To be e comp ehend he p ocess, i is possible o summa ize he p e iously desc ibed me hods using he ollowing diag am. Figu e 5 - Indi idual Kalman Fil e s p ocedu e 3) Fusion Algo i hm o p o ide Final Es ima e A e imp o ing indi idual es ima es’ pe o mance wi h s anda d Kalman Fil e s, i is ime o use hei es ima es, in o de o yield a mo e obus , s able and accu a e es ima e. The e a e se e al me hods o pe o m he usion o wo s able measu es, like a s anda d weigh ed a e age, an In o ma ion Fil e , e c. The usion is ca ied ou using a Decen alized Kalman Fil e [8] (DKF). A Decen alized Kalman Fil e consis s o a ne wo k o nodes in which he me ging/ usion p ocess occu s in each node, sha ing he in o ma ion wi h he neighbo nodes. This allows he compu a ional s ess o a cen alized usion p ocesso o be elimina ed. The sensing node pe o ms he local (indi idual) Kalman Fil e and sha es he in o ma ion wi h he o he nodes, all o hem also pe o ming usion algo i hms. A e ha , i assimila es he in o ma ion ecei ed by he neighbo s, p oducing and imp o ed local es ima e, using bo h local and global in o ma ion. This a chi ec u e b ings he modula i y concep o he able, ega ding he Kalman Fil e ing compu a ional “scene”, gi en he ac ha no a p io i knowledge is needed and he local es ima es a e co ec ed using global in o ma ion. Ano he imp o emen is in he obus ness o he sys em, whe e he pe o mance o he sys em does no depend on any o he p ocesso s, bu on he communica ion link ins ead, being he sys em lexible o loss o addi ion o nodes, since he e is no need o know he con igu a ion o he ne wo k, only i s s a e. The DKF is ep esen ed by he ollowing equa ions, indexed by node i. Fi s , one should compu e bo h he s a e e o in o ma ion and he co a iance e o in o ma ion. ei(k) = Pi(k)-1Xi(k)-Pi(k-1)-1Xi(k-1) (6) Ei(k) = Pi(k)-1-Pi(k-1)-1 (7) Then, using he p e ious equa ions one can compu e he global ( used/me ged) co a iance (8) and s a e (9). As he senso s a e independen and do no in e e e wi h each o he , he p ocess co a iance a iable Q, should be nea ze o, bu ne e ze o. P(k) = (P(k-1)-1+Σ(Ei))-1 (8) X(k) = P(k)( P(k-1)-1X(k-1) + Σ(ei) ) (9) Wi h he applica ion o he Kalman Fil e ing, he es ima es become a lo mo e s able, using only simple ma hema ic ope a ions, no consuming much p ocessing ime, yielding a a he good es ima e abou he localiza ion o he obo . V. FINDING ROBOT’S ORIENTATION The obo ’s o ien a ion is a majo componen o he p oposed me hod, as he me hod capabili ies ely on he quali y o es ima ion o he obo ’s heading di ec ion, o es ima e he obo ’s posi ion. In his Sec ion, wo di e en me hods a e p esen ed, complemen a y i needed, o es ima e he obo ’s heading in he ield. As desc ibed in he eam desc ip ion pape [9], e e y MINHO eam obo is equipped wi h a 9 Deg ees o F eedom IMU – Ine ial Measu emen Uni – ha p o ides o ien a ion, Yaw, Pi ch and Roll. This ou alues a e used o achie e he bes o ien a ion possible, compu ing he alues using a DCM – Di ec ion Cosine Ma ix- algo i hm [10], while using i s o he ea u es o help imp o ing he sel -localiza ion algo i hm. The Yaw componen is e y impo an o co ec and imp o e he alue gi en by he compass, being also used o de ec collisions wi h he obo , oge he wi h he Pi ch and Roll componen s. A. Robo o ien a ion using an Ine ial Measu emen Uni An Ine ial Measu emen Uni (IMU) is equi ed o he p ope ope a ion o he localiza ion algo i hm, as desc ibed in he p e ious Sec ion. Al hough you can ge he obo ’s heading using only a compass, i was decided o use an IMU o wo main easons. Fi s , o i s abili y o adap o su ounding magne ic ields, secondly, due o he possibili y o de ec ing collisions be ween obo s. The IMU uses an implemen a ion o Di ec ion Cosine Ma ix (DCM). The mo i a ion o using DCM, was he need o g ea e s abili y when ob aining XYZ alues. Wi hou he need o go in o u he explana ions and de ails, as i is no he main ocus o his wo k, he basics o his algo i hm will be explained. Essen ially wha DCM does, is o ep esen he o ien a ion o he obo in ela ion o he o ien a ion o he Ea h, using he ollowing o a ion ma ix: Figu e 6 - DCM Ro a ion Ma ix I is possible o ans o m a sys em ec o in o ano he sys em, by mul iplying i by he o a ion ma ix. The e e se could also be achie ed simply by mul iplying wi h he in e se ma ix, aking ad an age o he o a ion ma ix key p ope ies, i.e., i s o hogonali y. A e he calcula ions and con e sions o ma ices, a di ec ela ionship be ween he c ea ed DCM and he Eule angles is gi en, using he ollowing equa ion. R = (10) Whe e:  is he angle be ween he x axis and he N axis.  is he angle be ween he z axis and he Z axis.  is he angle be ween he N axis and he X axis Figu e 7 - P ope Eule angles ep esen ing o a ions abou axis z, N and Z. Using he Yaw, Pi ch and Roll da a ga he ed by he IMU (X, Y and Z), one can es ima e he posi ion o he pla o m in ela ion o he g ound plane, knowing i he obo is i led in a ce ain axis. As desc ibed in Sec ion IV, he odome y posi ion es ima ion is used o complemen he sel -localiza ion algo i hm, p o iding addi ional in o ma ion o i . When wo obo s collide, usually i is he case when he obs acle de ec ion and a oidance ailed, and he obo s will s ill be d i ing (wheels mo ing). A leas one o he mo o s will cease o make con ac wi h he g ound, becoming a mo o wi h ee- spinning o skidding mo ion, in oducing la ge e o s o he odome y es ima ion p ocess. Figu e 8 - Collision de ec ion using he IMU So, his ea u e p o ided by he IMU and he ans o ma ion algo i hm, allows o hal he usion algo i hm ha me ges he ision and odome y es ima es, only conside ing he ision es ima es, no aking in o accoun he e o s in oduced by he odome y es ima e, when a collision happens. This helps he ac ha , he obo will no lose he posi ion acking, and he e is no need o pe o m a global localiza ion, con inuing o co ec and o sel -localize using only local sea ching pa e ns. B. Robo ’s o ien a ion using His og ams Despi e he ac ha he IMU and he algo i hm desc ibed be o e ake in o accoun ex e nal magne ic ields, a oiding eading e o s due o he p esence o o he magne ic ields, he e is always he necessi y o ha e a complemen a y sys em, o p o ide he mos impo an in o ma ion, in his case, using eliable imaging and ea u e ex ac ion algo i hms. One o he mos impo an achie emen s o he olde e a o MINHO eam, was o de elop an algo i hm ha p o ide he obo ’s o ien a ion, wi h he de ec ed line poin s, using his og ams [11]. Fi s , a his og am is buil , coun ing he numbe o line poin s de ec ed in a ce ain di ec ion, e ically and ho izon ally. In o de o calcula e he o ien a ion o he obo , he his og am is o a ed om θ−40º o θ+40º, being θ is he las known o ien a ion, and 40º he maximum o a ion possible be ween ames. Ve i ying he his og am maximum alue (peak alue) o each o a ion, he angle in which he his og am’s alue is he maximum one, summing bo h e ical and ho izon al his og ams’ alues, is he new o ien a ion. Figu e 9 - Robo 's o ien a ion using His og ams example. A g aphic ep esen a ion o he o a ion o he his og ams and he alues ob ained, show he maximum alues and he new o ien a ion. Figu e 10 - G aphical ep esen a ion o o a ion esul s and maximum alue. As shown, he new o ien a ion alue has a e y p onounced maximum, being se a 30º. This me hod allows o compu e he new o ien a ion be ween ames, using al eady ga he ed da a, and simple nume ic ope a ions and compa isons. VI. DISCUSSION AND RESULTS The main objec i e o his wo k was o imp o e he compu a ional e iciency and speed o he Sel -Localiza ion algo i hm, used in all RoboCup’s MSL Robo s, while ying o keep he obus ness and accu acy o he exis en me hods. I was also p esen ed a usion algo i hm using a Decen alized Kalman Fil e , in o de o pe o m usion o so wa e and ha dwa e senso s, yielding smoo h and accu a e es ima es o he obo ’s posi ion, while p e en ing local e oneous es ima es by any o he sou ces o he usion algo i hm. The applica ion o Kalman Fil e ing in senso usion is widely used, p o en o be a e y obus , e icien and e sa ile me hod, pe o ming complex il e ing wi h simple ma hema ical ope a ions, allowing o compu e h ee Kalman Fil e s, wi hou dis u bing he “p ocessing ime window”. The usion algo i hm also b ings an imp o emen o he s anda d senso usion algo i hms used, while, using an Ine ial Measu emen Uni b ough o he bene i s, ha we e no possible be o e and also, we e no used in any pla o m in he league. The p oposed me hod achie ed he ask o educing he algo i hm compu a ional ime, educing d ama ically he global localiza ion p ocessing ime o only 150 milliseconds, ega ding an O icial RoboCup MSL ield. When pe o ming local sea ching, in an a ea o 4m2 a ound he cen e o he obo , he p oposed me hod only akes 3 o 4 milliseconds. The use o wo di e en me hods o acqui e he obo ’s o ien a ion, one by ha dwa e and he o he by so wa e, makes he ask o es ima ing he obo ’s heading e y e icien and us wo hy, while aking ad an age o o he capabili ies o he IMU, like de ec ing collisions in o de o elimina e he noise in oduced by he odome y in hose si ua ions, when calcula ing he posi ion o he obo . Al hough he me hod assu es a e y accu a e and as localiza ion es ima e, he iden i ica ion o line poin s needs o be ca ied ou co ec ly, which ep esen s a down-side when compa ing o he mos common me hod in he league. ACKNOWLEDGMENT This wo k was de eloped a he Labo a ó io de Au omação e Robó ica by MINHO eam´s esea ching and de eloping eam, a Uni e si y o Minho, unde he supe ision o P o esso A. Fe nando Ribei o and A. Gil Lopes. The knowledge exchanging be ween he RoboCup’s MSL eams and communi y con ibu ed g ea ly o he de elopmen o his wo k. This wo k has been suppo ed by COMPETE: POCI-01- 0145-FEDER-007043 and FCT – Fundação pa a a Ciência e Tecnologia wi hin he P ojec Scope: UID/CEC/00319/2013. REFERENCES [1] Ma in Laue , Sascha Lange, and Ma in Riedmille , “Calcula ing he Pe ec Ma ch: an E icien and Accu a e App oach o Robo Sel - Localiza ion”, In A. B eden eld, A. Jaco , I. Noda and Y. Takahashi, edi o s, RoboCup 2005: Robo Socce Wo ld Cup IX, LNCS. Sp inge , 2005. [2] Gil Lopes, Fe nando Ribei o, Nino Pe ei a, “Ca adiop ic sys em op imisa ion o omnidi ec ional RoboCup MSL obo s”, T. Rö e e al. (Eds.): RoboCup 2011, Lec u e No es in Compu e Science (including subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics) 7416 LNCS, pp. 318-328. Sp inge -Ve lag Be lin Heidelbe g 2012. [3] P. Lima, A. Bona ini, C. Machado, F. M. Ma chese, C. Ma ques, F. Ribei o, D. G. So en i, “Omni-Di ec ional Ca adiop ic Vision o Socce Robo s”, Special Issue o he Robo ics and Au onomous Sys ems Jou nal, Else ie , Ames e dão, Volume 36, nº 2, 31 de Agos o de 2001. [4] Gou ab Sen Gup a and Donald Bailey, “Disc e e YUV look-up ables o as colou segmen a ion o obo ic applica ions”, IEEE, Con e enced in May 2008. [5] Jaap Vos, e al, “Team Desc ip ion Pape ”, ASML Falcons, Veldho en, The Ne he lands, 2016. [6] Kalman, R. E. 1960. “A New App oach o Linea Fil e ing and P edic ion P oblems”, T ansac ion o he ASME--Jou nal o Basic Enginee ing, pp. 35-45 (Ma ch 1960). [7] G eg Welch and Ga y Bishop,”An In oduc ion o he Kalman Fil e ”, Depa men o Compu e Science Uni e si y o No h Ca olina a Chapel Hill (2001&2006). [8] Z. Hidaya , R. Babuska, B. De Schu e , and A. Nunez : “Decen alized Kalman il e compa ison o dis ibu ed-pa ame e sys ems: A case s udy o a 1D hea conduc ion p ocess” (2011). [9] F. Ribei o, G. Lopes, H. Ribei o, P. Sil a, T. Maia, R. Ro iz, A. Gomes and N. Fe ei a, “Minho Team ‘2016: Team Desc ip ion Pape ”, Uni e si y o Minho, Guima ães, Po ugal, 2016. [10] P eme lani, W., Biza d, P.: “Di ec ion cosine ma ix IMU: heo y”, h p://gen lena .googlecode.com/ iles/ DCMD a 2.pd [11] Fe nando Ribei o, Gil Lopes, B uno Pe ei a, João Sil a, Paulo Ribei o, Joao Cos a, Sé gio Sil a, João Rod igues, and Paulo T iguei os, “Robo O ien a ion wi h His og ams on MSL”, (2012) T. Rö e e al. (Eds.): RoboCup 2011, Lec u e No es in Compu e Science (including subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics) 7416 LNCS, pp. 507-514. Sp inge -Ve lag Be lin Heidelbe g 2012. DOI: 10.1007/978-3-642-32060-6_43.