Probabilistic Fundamentals in Robotics

Size: px
Start display at page:

Download "Probabilistic Fundamentals in Robotics"

Transcription

1 Probabilisic Fundamenals in Roboics Probabilisic Models of Mobile Robos Robo localizaion Basilio Bona DAUIN Poliecnico di Torino Course Ouline Basic mahemaical framework Probabilisic models of mobile robos Mobile robo localizaion problem Roboic mapping Probabilisic planning and conrol Reference exbook Thrun, Burgard, Fox, Probabilisic Roboics, MIT Press, 2006 hp:// roboics.org/ Basilio Bona 2 1

2 Mobile robo localizaion problem Mobile robo localizaion: Markov and Gaussian Inroducion Markov localizaion EKF localizaion Muli hypohesis racking UKF localizaion Mobile robo localizaion: Grid and Mone Carlo Grid localizaion Mone Carlo localizaion Localizaion in dynamic environmens Basilio Bona 3 Inroducion Localizaion or posiion esimaion is he problem of deermining he pose of a rover relaive o a given map of he environmen Localizaion is he process of esablishing he correspondence beween eenhe map reference frame and he local rover reference frame Hence coordinae ransformaion is necessary: from robo cenered local coordinae sysem o global one The pose x =(x y q) of he robo is no known and canno be measured direcly wih a sensor The pose mus be inferred from environmen daa (and odomery) A single measuremen is no sufficien o recover he pose Localizaion echniques depend on he map ype The algorihms presened are varians of he basic Bayes filer Basilio Bona 4 2

3 Example: localizaion wih known landmarks Basilio Bona 5 Localizaion as a sae sysem A graphical sae descripion of he localizaion problem The map m, he measuremens z and he conrols u are known Basilio Bona 6 3

4 Maps A 2D meric map A opological map (graph like) An occupancy grid map A mosaic map of a ceiling Basilio Bona 7 A axonomy of localizaion problems Posiion racking The iniial robo pose is known; he iniial error is assumed small; he pose uncerainy is ofen assumed o be unimodal; he problem is essenially a local one Global localizaion The iniial robo pose in unknown; he iniial error canno be assumed o be small; global localizaion includes posiion racking Kidnapped robo problem During operaion he robo ge kidnapped and suddenly reappears in anoher locaion; he robo believes o be able o esimae is posiion, bu his is no rue Basilio Bona 8 4

5 Saic vs dynamic environmens Saic environmens Environmens where he only moving objec is he robo. All oher objecs (feaures, landmarks, ec.) are fixed wr he map Dynamic environmens Oher objecs move, i.e., change heir locaion over ime. Changes usually are no episodic, i.e., hey persis over ime and are recognized by sensor readings. Changes ha affec only a single or few measuremens are reaed as noise. Examples are: people moving around he robo, doors ha open and close, movable furniure, dayligh illuminaion (for robo wih vision sensors) Two approaches: moving objec are included ino he sae Sensor daa are filered o eliminae he effec of unmodeled dynamics Basilio Bona 9 Passive vs acive localizaion Passive localizaion The localizaion module only observes he robo operaing; he localizaion resuls do no conrol he robo moion. Moion is random or deermined by oher ask accomplishmen Acive localizaion The localizaion module conrols he robo moion o maximize a performance crierion on localizaion error Basilio Bona 10 5

6 Single robo vs muli robo Single robo The localizaion is performed by a single robo ha moves (acively or passively) in he environmen. No communicaion issues are presen (excep ha wih he supervisor) Muli robo The localizaion is joinly performed by a eam or robos. In principle each one can perform single robo localizaion, bu if hey can deec each oher, his informaion can be shared and each robo belief can influence oher robos belief. Deecion and communicaion issues are imporan, as well as relaive pose measuremen. Basilio Bona 11 Markov localizaion Probabilisic localizaion algorihms are varians of he Bayes filer The simples localizaion filer (pure Bayesian) is called Markov localizaion Predicion bel( x ) = ò p( x x, u ) bel( x )dx Correcion bel( x ) = h p( z x ) bel( x ) Basilio Bona 12 6

7 Iniial belief Posiion racking bel( x ) de(2 p ) - ì ï ü T - ï = S exp í- ( x -x ) S ( x -x ) ý ï î 2 ï þ Nxx (;, S) 0 Normal disribuion Nxm (;, S) bel( x ) = Global localizaion 0 1 X Volume of he poses space Basilio Bona 13 KF summary x = Ax + Bu + e -1 ( ) px ( x, u) = N x; Ax + Bu, R -1-1 bel( x ) = ò p( x x, u ) bel( x )dx ß ß ( ; +, ) ( ;, S ) ~ N x Ax Bu R ~ N x m Basilio Bona 14 7

8 ò KF summary bel( x ) = p( x x, u ) bel( x ) dx ß ( ; + B u, R ) N ( x ; m, S ) ~ N x Ax B ~ N x, ß ì 1 ü T -1 bel( x ) = h exp ï ( x Ax B u ) R ( x Ax B u ) ï ò í ý ïî 2 ïþ ì 1 T -1 exp ï ü í- ( x -m ) S ( x -m ) ï dx ý - 1 ïî 2 ï ïþ ìï m = Am - + Bu 1 bel( x ) = ï í ï T S = AS A + R ïî -1 ß Basilio Bona 15 KF summary z = C x + d ( ) pz ( x ) = N z ; Cx, Q bel( x ) = h p( z x ) bel( x ) ß ß ( ;, ) N ( x ; m, S ) ~ N z ; C x Q ~ Basilio Bona 16 8

9 KF summary bel( x ) = h p( z x ) bel( x ) ß ß ( ;, Q ) N ( x ; m, S ) ~ N z C x ~ ; ß ì 1 belx = h ï- z -Cx Q z -Cx ïî 2 ì 1 ü T -1 exp ï í- ( x -m ) S ( x -m ) ï ý ïî 2 ïþ T -1 ( ) exp ( ) ( ) í ìï m = m + K ( z -C m ) bel( x ) = ï í wih K =SC ïs = ( I -KC ) S ïî T ü ï ý ïþ T ( C S C + Q ) - 1 Basilio Bona 17 EKF localizaion We assume ha he map is represened by a collecion of feaures (landmark based localizaion) We assume he velociy model and he measuremen model presened before All feaures are uniquely idenifiable (correspondence variables are known) = known landmarks Range and heading measuremens are available pz ( x, mc, ) Door 1 Door 2 Door 3 One dimensional environmen wih hree landmarks (doors) Correspondence is known Basilio Bona 18 9

10 Iniial belief (uniform) Measuremen model Correcion/updae sep Moion model (predicion) Measuremen model Correcion/updae sep Moion model (predicion) Basilio Bona 19 EKF_known_correspondence_localizaion (,, u, z, c, m) m - S 1-1 Predicion æ x x x ö ç ç m m m -1, x -1, y -1, q gu (, m ) -1 y y y 1: G Jacobian of g wr locaion = = x 1 m m m - -1, x -1, y -1, q q q q ç ç m m m çè -1, x -1, y -1, q ø æ x x ö v w (, ) gu m -1 y y 2: V = = Jacobian of g wr conrol u v w q q ç ç v w è ø æ 2 ö ( a v a w 1 2 ) 0 3: M + Moion noise = 2 ç 0 ( a v a w çè ) ø 4: m = gu (, m ) Prediced mean -1 5: S = GS G T + VMV T Prediced covariance -1 Basilio Bona 20 10

11 EKF_known_correspondence_localizaion (,, u, z, c, m) Correcion æ 2 2 ö ( m m ) ( m m x, x y, y) (, ) m - S 1-1 1: zˆ = ç m m m m m èçèaan y, y x, x, q ø Prediced measuremen mean 2: æ r r r ö h( m, m) m m m x, y,, q H = = x f f f ç m m m çè x, y,, q ø Jacobian of h wr locaion 3: æ 2 s 0 ö r Q = 2 ç 0 s çè r ø T 4: S = H S H + Q Prediced measuremen covariance T -1 5: K = S H S Kalman gain 6: m = m + K ( z -zˆ ) Updaed mean 7: S = I -K H S Updaed covariance ( ) Basilio Bona 21 EKF predicion sep differen moion noise parameers Small roaion and ranslaion noise High roaion noise High ranslaion noise High roaion and ranslaion noise Basilio Bona 22 11

12 EKF measuremen predicion sep innovaion z - zˆ innovaion Basilio Bona 23 EKF correcion sep measuremen uncerainy uncerainy in robo locaion Predicion uncerainy Basilio Bona 24 12

13 Esimaion Sequence (1) Accurae deecion sensor rajecories esimaed from he moion conrol uncerainy ground ruh rajecories before correced rajecories afer Basilio Bona 25 Esimaion Sequence (2) Less accurae deecion sensor Basilio Bona 26 13

14 Esimaion wih unknown correspondences When landmark correspondence is no cerain (as i is ofen he case) one should adop a sraegy o deermine he ideniy of he landmarks during localizaion The mos simple and popular echnique e is he maximum m likelihood correspondence One deermines he mos likely value of he correspondence variable and use i as he rue one This sraegy is fragile when differen landmarks have equally likely values How o reduce he risk Choose landmarks ha are far apar, so no o be confused Make sure ha he robo pose uncerainy remains small Basilio Bona 27 ML daa associaion The maximum likelihood esimaor (MLE) deermines he correspondence c ha maximizes he likelihood cˆ = arg max p( z c, z, u, m) 1: 1: -1 1: c This expression is condiioned on all prior correspondences ha are reaed as always correc c1: - 1 When he number of landmarks is high, he number of possible correspondence grows oo much and he problem becomes inracable A soluion is o maximize each erm singularly and hen proceed as in cˆ = arg max p( z c, z, u, m) i i 1: 1: -1 1: i c i i» Nz hm c m HS H T + Q i c arg max ( ; (,, ), ) Basilio Bona 28 14

15 Muli hypohesis racking filer Each color represens a differen hypohesis Basilio Bona 29 Localizaion wih MHTF Curren belief is represened by muliple hypoheses Each hypohesis is racked by a Kalman filer Addiional problems: Daa associaion: Which observaion corresponds o which hypohesis? Hypohesis managemen: When o add/delee hypoheses? Large body of lieraure on arge racking, moion correspondence ec. Basilio Bona 30 15

16 MHT example The blue robo sees he door, bu canno resolve is posiion since also he oher hypohesis (in red) are likely Basilio Bona 31 MHT (1) Hypoheses are exraced from laser range finder (LRF) scans Each hypohesis is given by pose esimae, covariance marix and probabiliy of being he correc one: H = { xˆ, S, PH ( )} i i i i Hypohesis probabiliy is compued using Bayes rule sensor repor Pr ( H) PH ( ) i i PH ( r) = i Pr () Hypoheses wih low probabiliy bbl are dl deleedd New candidaes are exraced from LRF scans C = {, z R} j j j Basilio Bona 32 16

17 MHT (2) Basilio Bona 33 MHT (3) Basilio Bona 34 17

18 MHT: Implemened Sysem (3) Example run # hypoheses P(H bes ) Map and rajecory #hypoheses vs. ime Basilio Bona 35 UKF localizaion The UKF localizaion algorihm uses he unscened ransform o linearize he moion and he measuremen models Insead of compuing Jacobians of hese models i compues Gaussians using sigma poins and ransforms hem hrough he model The landmark associaion (correspondence) is cerain Basilio Bona 36 18

19 UKF Algorihm par a) Sigma poins Basilio Bona 37 UKF Algorihm par b) Cross covariance Basilio Bona 38 19

20 UKF_localizaion (,, u, z, m) æ 2 ö ( a v a w 1 2 ) 0 ç + 2 m - S 1-1 ( ) c = m m + g S m -g S a a a a a a Predicion M = 0 ( a v a w Moion noise ç çè ) ø æ 2 s 0 ö r Q = Measuremen noise 2 ç 0 s çè r ø T T a æ T ö m = m 1 1 ( 00) ( ) ç çè Augmened sae mean ø æs 0 0ö -1 a S = 0 M 0-1 Augmened covariance ç çè 0 0 Q ø c x = g( c) Predicion of sigma poins Sigma poins 2L i x m = åw c m i, Prediced mean i= 0 2L T i x x S = w å c ( c -m i, ) ( c -m i, ) Prediced covariance i= 0 Basilio Bona 39 UKF_localizaion (,, u, z, m) m - S 1-1 Correcion 1 x ( ) Z = h c + c z Measuremen sigma poins 2L i zˆ = åw Z m i, Prediced measuremen mean i= 0 ( Z ˆ)( Z ˆ) 2L T i = å - - c i, i, i= 0 S w z z ( )( zˆ,, ) 2L T xz, i x = åw - - c i i i= 0 S c m Z Pred. measuremen covariance Cross covariance K = S S xz, -1 Kalman gain m = m + K ( z -zˆ ) S = S -KSK T Updaed mean Updaed covariance Basilio Bona 40 20

21 UKF_localizaion (,, u, z, m) m - S 1-1 Correcion 2 zˆ æ 2 2 ö ( m m ) ( m m x, x y, y) ( ) = Prediced measuremen mean ç aan 2 m m, m m m çè y, y x, x, q ø æ r r r ö h( m, m) m m m x, y,, q H = = x f f f Jacobian of h w.r.. locaion ç m m m çè x, y,, q ø æ 2 s 0 ö r Q = 2 ç 0 s çè r ø S = H S H + Q T K m = m + K ( z -zˆ ) S = S H S T -1 ( I K H ) S = - Pred. measuremen covariance Kalman gain Updaed mean Updaed covariance Basilio Bona 41 UKF Predicion Sep 3) High noise in roaion small in ranslaion 1) Small noise in ranslaion and roaion Iniial esimae 2) High noise in ranslaion small in roaion 4) High noise in ranslaion and in roaion Basilio Bona 42 21

22 UKF Observaion Predicion Sep The lef plos show he sigma poins prediced from wo moion updaes along wih he resuling uncerainy ellipses. The rue robo and he observaion are indicaed by he whie circle and he bold line, respecively The righ plos show he resuling li measuremen predicion sigma poins. The whie arrows indicae he innovaions, he differences beween observed and prediced measuremens Basilio Bona 43 UKF Correcion Sep The panels on he lef show he measuremen predicion The panels on he righ he resuling correcions, which updae he mean esimae and reduce he posiion uncerainy ellipses Basilio Bona 44 22

23 EKF Correcion Sep Basilio Bona 45 Robo rajecory according o he moion conrol (dashed lines) and he resuling rue rajecory (solid lines) Landmark deecions are indicaed by hin lines Esimaion Sequence EKF PF UKF Basilio Bona 46 23

24 Predicion Qualiy EKF predicion UKF predicion Approximaion error due o linearizaion The robo moves on a circle The reference covariances are exraced from an accurae, sample based predicion Basilio Bona 47 Mobile robo localizaion Grid and Mone Carlo mehods These algorihms can process raw sensor measuremen, i.e., here is no need o exrac feaures from measuremens These mehods are non parameric, e.g., hey are no limied o unimodal disribuions ions as wih he EKF localizaion aion mehod They can solve global localizaion problem and kidnapped robo problems (in some cases); he EKF algorihm is no able o solve such problems The firs mehod is called grid localizaion The second mehod is called Mone Carlo localizaion Basilio Bona 48 24

25 Grid localizaion Grid localizaion uses a hisogram filer o represen poserior belief The grid dimension is a key facor for he performances of he mehod When he cell grid dimension is small, he algorihm can be exremely slow When he cell grid dimension is large, here can be an informaion loss ha makes he algorihm no working in some cases Basilio Bona 49 Example Measuremen model Correcion/updae sep Moion model (predicion) Measuremen model Correcion/updae sep Moion model (predicion) Basilio Bona 50 25

26 Topological grid map A opological map is a graph annoaion of he environmen Topological maps assign nodes o paricular places and edges as pahs if direc passage beween pairs of places (end nodes) exis Humans manage spaial knowledge primarily by opological informaion This informaion is used o consruc a hierarchical opological map ha describes he environmen Basilio Bona 51 Topological grid map Topological grid map represenaion: Coarse gridding Cells of varying size Resoluion influenced by he srucure of he environmen (significan places and landmarks as doors, corridors windows, T juncions, dead ends, ac as grid elemens) Grid elemens Basilio Bona 52 26

27 Meric grid map Meric grid represenaion: Finer gridding Cells of uniform size Resoluion no influenced by he srucure of he environmen Usually cell sizes of 15 cm x 15 cm up o 1 m x 1m Moion model affeced by robo velociy and cell size q Basilio Bona 53 Cell size influence Cell size influence he performance of he grid localizaion algorihm Average localizaion error as a funcion of grid cell size, for ulrasound sensors and laser range finders Average CPU ime needed for global localizaion as a funcion of grid resoluion, shown for boh ulrasound sensors and laser range finders Basilio Bona 54 27

28 Example wih meric grids 1 Basilio Bona 55 Example wih meric grids 2 Basilio Bona 56 28

29 Example wih meric grids 3 Basilio Bona 57 Example wih sonar daa 1 Basilio Bona 58 29

30 Example wih sonar daa 2 Basilio Bona 59 Example wih sonar daa 3 Basilio Bona 60 30

31 Example wih sonar daa 4 Basilio Bona 61 Example wih sonar daa 5 Basilio Bona 62 31

32 Mone Carlo localizaion Mone Carlo localizaion (MCL) is a relaively new, ye very popular algorihm I is a versaile mehod, where he belief is represened by a se of paricles (i.e., a paricle filer) I can be used boh for local and for global localizaion problems In he conex of localizaion, he paricles are propagaed according o he moion model They are hen weighed according o he likelihood of he observaions In a re sampling sep, new paricles are drawn wih a probabiliy proporional o he likelihood of he observaion The mehod firs appeared in 70 s, and was re discovered by Kiagawa and Isard & Blake in compuer vision Basilio Bona 63 Paricle filer localizaion (MCL) Paricle filers based localizaion (Mone Carlo Localizaion MCL) uses a se of weighed random samples o approximae he robo pose belief Paricle se size n ( ) [] i ( ˆ [] i» å w d - ) bel p p p i= 1 n wih å i= 1 w [] i = 1 Pose of he paricle Weigh of he paricle Basilio Bona 64 32

33 Paricle filer localizaion (MCL) Paricle based Represenaion of posiion belief Paricle based approximaion Gaussian approximaion (ellipse: 95% accepance region) Basilio Bona 65 Paricle filer localizaion (MCL) Using paricle filers approximaion, he Bayes Filer can be reformulaed as follows (saring from he robo iniial belief a ime zero) 1. PREDICTION: Generae a new se of paricles given he moion model and he applied conrols 2. UPDATE: Assign o each paricle an imporance weigh according o he sensor measuremens 3. RESAMPLING: Randomly resample paricles in funcion of heir imporance weigh Basilio Bona 66 33

34 Predicion Predicion Generae a new se of paricles given he moion model and he applied conrols For each paricle: Given he paricle pose a ime sep -1 and he commands, he paricle pose a ime is prediced using he moion model i ( -1 ) [] i [] pˆ = f pˆ, u, w The variable is randomly exraced according o he noise disribuion We obain a se of prediced paricles Basilio Bona 67 Updae Updae Assign o each paricle an imporance weigh according o he sensor measuremens For each paricle: Compare paricle s predicion of measuremens wih acual [ i ] measuremens z vs h( pˆ ) Paricles whose predicions mach he measuremens are given a high weigh In red he paricles wih high weighs Basilio Bona 68 34

35 Resampling Resampling Randomly resample paricles in funcion of heir imporance weigh For each paricle: For n imes draw (wih replacemen) a paricle from G - wih probabiliy bili given by he imporance weighs ih and pu i in he se G Paricles whose predicions mach he measuremens are given a high weigh G The new se provides he paricle based approximaion of he robo pose a ime Basilio Bona 69 Example Measuremen model Correcion/updae sep Moion model (predicion) Measuremen model Correcion/updae sep Moion model (predicion) Basilio Bona 70 35

36 Example Basilio Bona 71 MC_localizaion (, u, z, m) c - 1 1: c = c = Æ 2: for m = 1o M do = [ m] [ m] 3: x sample_moion_model moion model ( u, x ) -1 [ m] [ m] w = measuremen_model z x m [ m] [ m] 5: c = c + x, w 4: (,, ) 6: endfor 7: for m = 1o M do 8: 9 : 10: endfor draw i wih probabiliy add x 11 : reurn c [] i o c µ w [] i see previous slides Basilio Bona 72 36

37 Mone Carlo Localizaion wihin a sensor infrasrucure Fixed sensors deployed in known posiions in he environmen STEP 1: Acquire odomery Basilio Bona 73 STEP 1: Acquire odomery Filer Predicion Basilio Bona 74 37

38 STEP 1: Acquire odomery Filer Predicion Acquire meas. Basilio Bona 75 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Basilio Bona 76 38

39 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling Basilio Bona 77 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 1: Acquire odomery Basilio Bona 78 39

40 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 1: Acquire odomery Filer Predicion Basilio Bona 79 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Basilio Bona 80 40

41 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Basilio Bona 81 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling Basilio Bona 82 41

42 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 3: Acquire odomery Basilio Bona 83 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 3: Acquire odomery Filer Predicion Basilio Bona 84 42

43 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 3: Acquire odomery Filer Predicion Acquire meas. Basilio Bona 85 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 3: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Basilio Bona 86 43

44 STEP 1: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 2: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling STEP 3: Acquire odomery Filer Predicion Acquire meas. Weighs Updae Resampling Basilio Bona 87 Example for landmark based localizaion Basilio Bona 88 44

45 Properies of MCL MCL can approximae any disribuion, as i can represen complex muli modal disribuions and blend hem wih Gaussian syle disribuions Increasing he oal number of paricles increases he accuracy of he approximaion The number of paricles M is a rade off parameer beween accuracy and necessary compuaional resources Paricles number shall remain large enough o avoid filer divergence The paricles number may remain fixed or change adapively Basilio Bona 89 Adaping he paricle size In he example below, he number of paricles is very high (~ ) o allow an accurae represenaion of he belief during early sages of he algorihm, bu is unnecessarily high in laer sages, when he belief concenraes in smaller regions an adapive sraegy is required Basilio Bona 90 45

46 KLD sampling Kullback Leibler divergence (KLD) sampling is a varian of MCL ha adaps he number of paricles over ime KLD (also known as informaion divergence, informaion gain, relaive enropy)is a measure of he difference beween wo probabiliy disribuions For probabiliy disribuions P and Q of adiscree random variable heir KLD is defined as For disribuions P and Q of a coninuous random variable, he KLD is defined as an inegral Basilio Bona 91 KLD sampling The idea is o se adapively he number of paricles based on a saisical bound on he sample based approximaion qualiy A each ieraion of he PF, KLD sampling deermines he number of samples such ha, wih probabiliy 1- d, he error beween he rue poserior and he sample based approximaion is less han e To derive his bound, we assume ha he rue poserior is given by a discree, piecewise consan disribuion such as a discree densiy ree or a muli dimensional hisogram D. Fox, Adaping he sample size in paricle filers hrough KLDsampling, Inernaional Journal of Roboics Research, 2003 Basilio Bona 92 46

47 KLD sampling Given a discree disribuion wih k bins, and given he number n of samples o be chosen, we deermine n as follows 1 2 n = ck -1,1-d 2e 2 where c is he chi square disribuion wih k -1 d.o.f. and k-1,1-d probabiliy 1 - d we can guaranee ha wih probabiliy 1- dhe KL disance beween he MLE and he rue disribuion is less han e Basilio Bona 93 Approximaion formula n k -1 ì 2 2 ü ï 1 z ï í d ý 2e ï 9( k -1) 9( k -1) î ïþ 3 where z is he upper 1 - d quanile of he nomal N(0,1) disribuion 1-d Basilio Bona 94 47

48 Dynamic environmen Markov assumpions are good for a saic environmen Ofenhe he environmen where robos operae is full of people moving here and here People dynamics is no modeled by he sae x Probabilisic approaches are robus since hey incorporae sensor noise, bu... Sensor noise mus be independen a each ime sep... while people dynamics is highly dependen in ime Basilio Bona 95 Dynamic environmen: which soluion Sae Augmenaion include he hidden sae ino he sae esimaed by he filer i is more general, bu suffer from high compuaional complexiy: he pose of each subjec moving around he robo mus be esimaed, and he number of saes varies in ime Oulier rejecion pre process sensor measuremens o eliminae measuremens affeced by hidden sae i may work well when he people presence affecs he sensors (laser range finder, and, o lesser exen, vision sensors) reading Basilio Bona 96 48

49 Oulier rejecion The Expecaion maximizaion (EM) learning algorihm is used for oulier deecion and rejecion I comes from he beam model of range finders, where z shor and p shor parameers relaes o unexpeced objecs k we have o inroduce and compue a correspondence variable c ha can ake one of four values {hi, shor, max, rand} he desired probabiliy is compued as his inegral does no have a closed form soluion ò k p ( z x, m) z bel( x )dx k k shor shor pc ( = shor z, z, u, m) = 1: -1 1: k ò å p ( z x, m) z bel( x )dx c c c approximaion wih a represenaive sample of he poserior he measuremen is rejeced if he probabiliy exceeds a given hreshold Basilio Bona 97 Comparison of differen ML implemenaions EKF MHT Topological grid Meric grid MCL Measuremens landmarks landmarks landmarks Measuremen noise raw measuremens raw measuremens Gaussian Gaussian any any any Poserior Gaussian mixure of Gaussians hisogram hisogram paricles Efficiency (mem) Efficiency (ime) Ease of implemenaion i Resoluion Robusness Global localizaion no no yes yes yes Basilio Bona 98 49

50 Example Basilio Bona 99 Example Basilio Bona

51 Example Basilio Bona 101 Thank you. Any quesion? Basilio Bona

Probabilistic Robotics

Probabilistic Robotics Probabilisic Roboics Bayes Filer Implemenaions Gaussian filers Bayes Filer Reminder Predicion bel p u bel d Correcion bel η p z bel Gaussians : ~ π e p N p - Univariae / / : ~ μ μ μ e p Ν p d π Mulivariae

More information

Introduction to Mobile Robotics

Introduction to Mobile Robotics Inroducion o Mobile Roboics Bayes Filer Kalman Filer Wolfram Burgard Cyrill Sachniss Giorgio Grisei Maren Bennewiz Chrisian Plagemann Bayes Filer Reminder Predicion bel p u bel d Correcion bel η p z bel

More information

Probabilistic Fundamentals in Robotics

Probabilistic Fundamentals in Robotics Probabilistic Fundamentals in Robotics Probabilistic Models of Mobile Robots Robot localization Basilio Bona DAUIN Politecnico di Torino June 2011 Course Outline Basic mathematical framework Probabilistic

More information

Probabilistic Fundamentals in Robotics. DAUIN Politecnico di Torino July 2010

Probabilistic Fundamentals in Robotics. DAUIN Politecnico di Torino July 2010 Probabilistic Fundamentals in Robotics Probabilistic Models of Mobile Robots Robot localization Basilio Bona DAUIN Politecnico di Torino July 2010 Course Outline Basic mathematical framework Probabilistic

More information

Sequential Importance Resampling (SIR) Particle Filter

Sequential Importance Resampling (SIR) Particle Filter Paricle Filers++ Pieer Abbeel UC Berkeley EECS Many slides adaped from Thrun, Burgard and Fox, Probabilisic Roboics 1. Algorihm paricle_filer( S -1, u, z ): 2. Sequenial Imporance Resampling (SIR) Paricle

More information

Two Popular Bayesian Estimators: Particle and Kalman Filters. McGill COMP 765 Sept 14 th, 2017

Two Popular Bayesian Estimators: Particle and Kalman Filters. McGill COMP 765 Sept 14 th, 2017 Two Popular Bayesian Esimaors: Paricle and Kalman Filers McGill COMP 765 Sep 14 h, 2017 1 1 1, dx x Bel x u x P x z P Recall: Bayes Filers,,,,,,, 1 1 1 1 u z u x P u z u x z P Bayes z = observaion u =

More information

Zürich. ETH Master Course: L Autonomous Mobile Robots Localization II

Zürich. ETH Master Course: L Autonomous Mobile Robots Localization II Roland Siegwar Margaria Chli Paul Furgale Marco Huer Marin Rufli Davide Scaramuzza ETH Maser Course: 151-0854-00L Auonomous Mobile Robos Localizaion II ACT and SEE For all do, (predicion updae / ACT),

More information

CSE-571 Robotics. Sample-based Localization (sonar) Motivation. Bayes Filter Implementations. Particle filters. Density Approximation

CSE-571 Robotics. Sample-based Localization (sonar) Motivation. Bayes Filter Implementations. Particle filters. Density Approximation Moivaion CSE57 Roboics Bayes Filer Implemenaions Paricle filers So far, we discussed he Kalman filer: Gaussian, linearizaion problems Paricle filers are a way o efficienly represen nongaussian disribuions

More information

7630 Autonomous Robotics Probabilistic Localisation

7630 Autonomous Robotics Probabilistic Localisation 7630 Auonomous Roboics Probabilisic Localisaion Principles of Probabilisic Localisaion Paricle Filers for Localisaion Kalman Filer for Localisaion Based on maerial from R. Triebel, R. Käsner, R. Siegwar,

More information

Using the Kalman filter Extended Kalman filter

Using the Kalman filter Extended Kalman filter Using he Kalman filer Eended Kalman filer Doz. G. Bleser Prof. Sricker Compuer Vision: Objec and People Tracking SA- Ouline Recap: Kalman filer algorihm Using Kalman filers Eended Kalman filer algorihm

More information

Announcements. Recap: Filtering. Recap: Reasoning Over Time. Example: State Representations for Robot Localization. Particle Filtering

Announcements. Recap: Filtering. Recap: Reasoning Over Time. Example: State Representations for Robot Localization. Particle Filtering Inroducion o Arificial Inelligence V22.0472-001 Fall 2009 Lecure 18: aricle & Kalman Filering Announcemens Final exam will be a 7pm on Wednesday December 14 h Dae of las class 1.5 hrs long I won ask anyhing

More information

Robot Motion Model EKF based Localization EKF SLAM Graph SLAM

Robot Motion Model EKF based Localization EKF SLAM Graph SLAM Robo Moion Model EKF based Localizaion EKF SLAM Graph SLAM General Robo Moion Model Robo sae v r Conrol a ime Sae updae model Noise model of robo conrol Noise model of conrol Robo moion model

More information

2016 Possible Examination Questions. Robotics CSCE 574

2016 Possible Examination Questions. Robotics CSCE 574 206 Possible Examinaion Quesions Roboics CSCE 574 ) Wha are he differences beween Hydraulic drive and Shape Memory Alloy drive? Name one applicaion in which each one of hem is appropriae. 2) Wha are he

More information

SEIF, EnKF, EKF SLAM. Pieter Abbeel UC Berkeley EECS

SEIF, EnKF, EKF SLAM. Pieter Abbeel UC Berkeley EECS SEIF, EnKF, EKF SLAM Pieer Abbeel UC Berkeley EECS Informaion Filer From an analyical poin of view == Kalman filer Difference: keep rack of he inverse covariance raher han he covariance marix [maer of

More information

Estimation of Poses with Particle Filters

Estimation of Poses with Particle Filters Esimaion of Poses wih Paricle Filers Dr.-Ing. Bernd Ludwig Chair for Arificial Inelligence Deparmen of Compuer Science Friedrich-Alexander-Universiä Erlangen-Nürnberg 12/05/2008 Dr.-Ing. Bernd Ludwig (FAU

More information

Probabilistic Robotics SLAM

Probabilistic Robotics SLAM Probabilisic Roboics SLAM The SLAM Problem SLAM is he process by which a robo builds a map of he environmen and, a he same ime, uses his map o compue is locaion Localizaion: inferring locaion given a map

More information

Probabilistic Robotics SLAM

Probabilistic Robotics SLAM Probabilisic Roboics SLAM The SLAM Problem SLAM is he process by which a robo builds a map of he environmen and, a he same ime, uses his map o compue is locaion Localizaion: inferring locaion given a map

More information

L07. KALMAN FILTERING FOR NON-LINEAR SYSTEMS. NA568 Mobile Robotics: Methods & Algorithms

L07. KALMAN FILTERING FOR NON-LINEAR SYSTEMS. NA568 Mobile Robotics: Methods & Algorithms L07. KALMAN FILTERING FOR NON-LINEAR SYSTEMS NA568 Mobile Roboics: Mehods & Algorihms Today s Topic Quick review on (Linear) Kalman Filer Kalman Filering for Non-Linear Sysems Exended Kalman Filer (EKF)

More information

Localization. Mobile robot localization is the problem of determining the pose of a robot relative to a given map of the environment.

Localization. Mobile robot localization is the problem of determining the pose of a robot relative to a given map of the environment. Localizaion Mobile robo localizaion is he problem of deermining he pose of a robo relaive o a given map of he environmen. Taxonomy of Localizaion Problem 1 Local vs. Global Localizaion Posiion racking

More information

EKF SLAM vs. FastSLAM A Comparison

EKF SLAM vs. FastSLAM A Comparison vs. A Comparison Michael Calonder, Compuer Vision Lab Swiss Federal Insiue of Technology, Lausanne EPFL) michael.calonder@epfl.ch The wo algorihms are described wih a planar robo applicaion in mind. Generalizaion

More information

Augmented Reality II - Kalman Filters - Gudrun Klinker May 25, 2004

Augmented Reality II - Kalman Filters - Gudrun Klinker May 25, 2004 Augmened Realiy II Kalman Filers Gudrun Klinker May 25, 2004 Ouline Moivaion Discree Kalman Filer Modeled Process Compuing Model Parameers Algorihm Exended Kalman Filer Kalman Filer for Sensor Fusion Lieraure

More information

CSE-473. A Gentle Introduction to Particle Filters

CSE-473. A Gentle Introduction to Particle Filters CSE-473 A Genle Inroducion o Paricle Filers Bayes Filers for Robo Localizaion Dieer Fo 2 Bayes Filers: Framework Given: Sream of observaions z and acion daa u: d Sensor model Pz. = { u, z2, u 1, z 1 Dynamics

More information

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Inroducion o Mobile Roboics SLAM: Simulaneous Localizaion and Mapping Wolfram Burgard, Maren Bennewiz, Diego Tipaldi, Luciano Spinello Wha is SLAM? Esimae he pose of a robo and he map of he environmen

More information

Probabilistic Robotics The Sparse Extended Information Filter

Probabilistic Robotics The Sparse Extended Information Filter Probabilisic Roboics The Sparse Exended Informaion Filer MSc course Arificial Inelligence 2018 hps://saff.fnwi.uva.nl/a.visser/educaion/probabilisicroboics/ Arnoud Visser Inelligen Roboics Lab Informaics

More information

Recursive Bayes Filtering Advanced AI

Recursive Bayes Filtering Advanced AI Recursive Bayes Filering Advanced AI Wolfram Burgard Tuorial Goal To familiarie you wih probabilisic paradigm in roboics! Basic echniques Advanages ifalls and limiaions! Successful Applicaions! Open research

More information

Overview. COMP14112: Artificial Intelligence Fundamentals. Lecture 0 Very Brief Overview. Structure of this course

Overview. COMP14112: Artificial Intelligence Fundamentals. Lecture 0 Very Brief Overview. Structure of this course OMP: Arificial Inelligence Fundamenals Lecure 0 Very Brief Overview Lecurer: Email: Xiao-Jun Zeng x.zeng@mancheser.ac.uk Overview This course will focus mainly on probabilisic mehods in AI We shall presen

More information

Physics 235 Chapter 2. Chapter 2 Newtonian Mechanics Single Particle

Physics 235 Chapter 2. Chapter 2 Newtonian Mechanics Single Particle Chaper 2 Newonian Mechanics Single Paricle In his Chaper we will review wha Newon s laws of mechanics ell us abou he moion of a single paricle. Newon s laws are only valid in suiable reference frames,

More information

Tracking. Many slides adapted from Kristen Grauman, Deva Ramanan

Tracking. Many slides adapted from Kristen Grauman, Deva Ramanan Tracking Man slides adaped from Krisen Grauman Deva Ramanan Coures G. Hager Coures G. Hager J. Kosecka cs3b Adapive Human-Moion Tracking Acquisiion Decimaion b facor 5 Moion deecor Grascale convers. Image

More information

Tracking. Announcements

Tracking. Announcements Tracking Tuesday, Nov 24 Krisen Grauman UT Ausin Announcemens Pse 5 ou onigh, due 12/4 Shorer assignmen Auo exension il 12/8 I will no hold office hours omorrow 5 6 pm due o Thanksgiving 1 Las ime: Moion

More information

Notes on Kalman Filtering

Notes on Kalman Filtering Noes on Kalman Filering Brian Borchers and Rick Aser November 7, Inroducion Daa Assimilaion is he problem of merging model predicions wih acual measuremens of a sysem o produce an opimal esimae of he curren

More information

Temporal probability models

Temporal probability models Temporal probabiliy models CS194-10 Fall 2011 Lecure 25 CS194-10 Fall 2011 Lecure 25 1 Ouline Hidden variables Inerence: ilering, predicion, smoohing Hidden Markov models Kalman ilers (a brie menion) Dynamic

More information

Mapping in Dynamic Environments

Mapping in Dynamic Environments Mapping in Dynaic Environens Wolfra Burgard Universiy of Freiburg, Gerany Mapping is a Key Technology for Mobile Robos Robos can robusly navigae when hey have a ap. Robos have been shown o being able o

More information

Tracking. Many slides adapted from Kristen Grauman, Deva Ramanan

Tracking. Many slides adapted from Kristen Grauman, Deva Ramanan Tracking Man slides adaped from Krisen Grauman Deva Ramanan Coures G. Hager Coures G. Hager J. Kosecka cs3b Adapive Human-Moion Tracking Acquisiion Decimaion b facor 5 Moion deecor Grascale convers. Image

More information

An introduction to the theory of SDDP algorithm

An introduction to the theory of SDDP algorithm An inroducion o he heory of SDDP algorihm V. Leclère (ENPC) Augus 1, 2014 V. Leclère Inroducion o SDDP Augus 1, 2014 1 / 21 Inroducion Large scale sochasic problem are hard o solve. Two ways of aacking

More information

Temporal probability models. Chapter 15, Sections 1 5 1

Temporal probability models. Chapter 15, Sections 1 5 1 Temporal probabiliy models Chaper 15, Secions 1 5 Chaper 15, Secions 1 5 1 Ouline Time and uncerainy Inerence: ilering, predicion, smoohing Hidden Markov models Kalman ilers (a brie menion) Dynamic Bayesian

More information

CS 4495 Computer Vision Tracking 1- Kalman,Gaussian

CS 4495 Computer Vision Tracking 1- Kalman,Gaussian CS 4495 Compuer Vision A. Bobick CS 4495 Compuer Vision - KalmanGaussian Aaron Bobick School of Ineracive Compuing CS 4495 Compuer Vision A. Bobick Adminisrivia S5 will be ou his Thurs Due Sun Nov h :55pm

More information

Object tracking: Using HMMs to estimate the geographical location of fish

Object tracking: Using HMMs to estimate the geographical location of fish Objec racking: Using HMMs o esimae he geographical locaion of fish 02433 - Hidden Markov Models Marin Wæver Pedersen, Henrik Madsen Course week 13 MWP, compiled June 8, 2011 Objecive: Locae fish from agging

More information

Hidden Markov Models

Hidden Markov Models Hidden Markov Models Probabilisic reasoning over ime So far, we ve mosly deal wih episodic environmens Excepions: games wih muliple moves, planning In paricular, he Bayesian neworks we ve seen so far describe

More information

Fundamental Problems In Robotics

Fundamental Problems In Robotics Fundamenal Problems In Roboics Wha does he world looks like? (mapping sense from various posiions inegrae measuremens o produce map assumes perfec knowledge of posiion Where am I in he world? (localizaion

More information

Anno accademico 2006/2007. Davide Migliore

Anno accademico 2006/2007. Davide Migliore Roboica Anno accademico 2006/2007 Davide Migliore migliore@ele.polimi.i Today Eercise session: An Off-side roblem Robo Vision Task Measuring NBA layers erformance robabilisic Roboics Inroducion The Bayesian

More information

Speaker Adaptation Techniques For Continuous Speech Using Medium and Small Adaptation Data Sets. Constantinos Boulis

Speaker Adaptation Techniques For Continuous Speech Using Medium and Small Adaptation Data Sets. Constantinos Boulis Speaker Adapaion Techniques For Coninuous Speech Using Medium and Small Adapaion Daa Ses Consaninos Boulis Ouline of he Presenaion Inroducion o he speaker adapaion problem Maximum Likelihood Sochasic Transformaions

More information

Data Fusion using Kalman Filter. Ioannis Rekleitis

Data Fusion using Kalman Filter. Ioannis Rekleitis Daa Fusion using Kalman Filer Ioannis Rekleiis Eample of a arameerized Baesian Filer: Kalman Filer Kalman filers (KF represen poserior belief b a Gaussian (normal disribuion A -d Gaussian disribuion is

More information

Basilio Bona ROBOTICA 03CFIOR 1

Basilio Bona ROBOTICA 03CFIOR 1 Indusrial Robos Kinemaics 1 Kinemaics and kinemaic funcions Kinemaics deals wih he sudy of four funcions (called kinemaic funcions or KFs) ha mahemaically ransform join variables ino caresian variables

More information

Linear Response Theory: The connection between QFT and experiments

Linear Response Theory: The connection between QFT and experiments Phys540.nb 39 3 Linear Response Theory: The connecion beween QFT and experimens 3.1. Basic conceps and ideas Q: How do we measure he conduciviy of a meal? A: we firs inroduce a weak elecric field E, and

More information

0.1 MAXIMUM LIKELIHOOD ESTIMATION EXPLAINED

0.1 MAXIMUM LIKELIHOOD ESTIMATION EXPLAINED 0.1 MAXIMUM LIKELIHOOD ESTIMATIO EXPLAIED Maximum likelihood esimaion is a bes-fi saisical mehod for he esimaion of he values of he parameers of a sysem, based on a se of observaions of a random variable

More information

m = 41 members n = 27 (nonfounders), f = 14 (founders) 8 markers from chromosome 19

m = 41 members n = 27 (nonfounders), f = 14 (founders) 8 markers from chromosome 19 Sequenial Imporance Sampling (SIS) AKA Paricle Filering, Sequenial Impuaion (Kong, Liu, Wong, 994) For many problems, sampling direcly from he arge disribuion is difficul or impossible. One reason possible

More information

Georey E. Hinton. University oftoronto. Technical Report CRG-TR February 22, Abstract

Georey E. Hinton. University oftoronto.   Technical Report CRG-TR February 22, Abstract Parameer Esimaion for Linear Dynamical Sysems Zoubin Ghahramani Georey E. Hinon Deparmen of Compuer Science Universiy oftorono 6 King's College Road Torono, Canada M5S A4 Email: zoubin@cs.orono.edu Technical

More information

A PROBABILISTIC MULTIMODAL ALGORITHM FOR TRACKING MULTIPLE AND DYNAMIC OBJECTS

A PROBABILISTIC MULTIMODAL ALGORITHM FOR TRACKING MULTIPLE AND DYNAMIC OBJECTS A PROBABILISTIC MULTIMODAL ALGORITHM FOR TRACKING MULTIPLE AND DYNAMIC OBJECTS MARTA MARRÓN, ELECTRONICS. ALCALÁ UNIV. SPAIN mara@depeca.uah.es MIGUEL A. SOTELO, ELECTRONICS. ALCALÁ UNIV. SPAIN soelo@depeca.uah.es

More information

Monte Carlo data association for multiple target tracking

Monte Carlo data association for multiple target tracking Mone Carlo daa associaion for muliple arge racking Rickard Karlsson Dep. of Elecrical Engineering Linköping Universiy SE-58183 Linköping, Sweden E-mail: rickard@isy.liu.se Fredrik Gusafsson Dep. of Elecrical

More information

FastSLAM: An Efficient Solution to the Simultaneous Localization And Mapping Problem with Unknown Data Association

FastSLAM: An Efficient Solution to the Simultaneous Localization And Mapping Problem with Unknown Data Association FasSLAM: An Efficien Soluion o he Simulaneous Localizaion And Mapping Problem wih Unknown Daa Associaion Sebasian Thrun 1, Michael Monemerlo 1, Daphne Koller 1, Ben Wegbrei 1 Juan Nieo 2, and Eduardo Nebo

More information

State-Space Models. Initialization, Estimation and Smoothing of the Kalman Filter

State-Space Models. Initialization, Estimation and Smoothing of the Kalman Filter Sae-Space Models Iniializaion, Esimaion and Smoohing of he Kalman Filer Iniializaion of he Kalman Filer The Kalman filer shows how o updae pas predicors and he corresponding predicion error variances when

More information

Air Traffic Forecast Empirical Research Based on the MCMC Method

Air Traffic Forecast Empirical Research Based on the MCMC Method Compuer and Informaion Science; Vol. 5, No. 5; 0 ISSN 93-8989 E-ISSN 93-8997 Published by Canadian Cener of Science and Educaion Air Traffic Forecas Empirical Research Based on he MCMC Mehod Jian-bo Wang,

More information

Localization and Map Making

Localization and Map Making Localiaion and Map Making My old office DILab a UTK ar of he following noes are from he book robabilisic Roboics by S. Thrn W. Brgard and D. Fo Two Remaining Qesions Where am I? Localiaion Where have I

More information

An recursive analytical technique to estimate time dependent physical parameters in the presence of noise processes

An recursive analytical technique to estimate time dependent physical parameters in the presence of noise processes WHAT IS A KALMAN FILTER An recursive analyical echnique o esimae ime dependen physical parameers in he presence of noise processes Example of a ime and frequency applicaion: Offse beween wo clocks PREDICTORS,

More information

Multi-Robot Simultaneous Localization and Mapping (Multi-SLAM)

Multi-Robot Simultaneous Localization and Mapping (Multi-SLAM) Muli-Robo Simulaneous Localizaion and Mapping (Muli-SLAM) Kai-Chieh Ma, Zhibei Ma Absrac In his projec, we are ineresed in he exension of Simulaneous Localizaion and Mapping (SLAM) o muliple robos. By

More information

WATER LEVEL TRACKING WITH CONDENSATION ALGORITHM

WATER LEVEL TRACKING WITH CONDENSATION ALGORITHM WATER LEVEL TRACKING WITH CONDENSATION ALGORITHM Shinsuke KOBAYASHI, Shogo MURAMATSU, Hisakazu KIKUCHI, Masahiro IWAHASHI Dep. of Elecrical and Elecronic Eng., Niigaa Universiy, 8050 2-no-cho Igarashi,

More information

מקורות לחומר בשיעור ספר הלימוד: Forsyth & Ponce מאמרים שונים חומר באינטרנט! פרק פרק 18

מקורות לחומר בשיעור ספר הלימוד: Forsyth & Ponce מאמרים שונים חומר באינטרנט! פרק פרק 18 עקיבה מקורות לחומר בשיעור ספר הלימוד: פרק 5..2 Forsh & once פרק 8 מאמרים שונים חומר באינטרנט! Toda Tracking wih Dnamics Deecion vs. Tracking Tracking as probabilisic inference redicion and Correcion Linear

More information

Improved Rao-Blackwellized H filter based mobile robot SLAM

Improved Rao-Blackwellized H filter based mobile robot SLAM Ocober 216, 23(5): 47 55 www.sciencedirec.com/science/journal/158885 The Journal of China Universiies of Poss and Telecommunicaions hp://jcup.bup.edu.cn Improved Rao-Blackwellized H filer based mobile

More information

From Particles to Rigid Bodies

From Particles to Rigid Bodies Rigid Body Dynamics From Paricles o Rigid Bodies Paricles No roaions Linear velociy v only Rigid bodies Body roaions Linear velociy v Angular velociy ω Rigid Bodies Rigid bodies have boh a posiion and

More information

Comparing Means: t-tests for One Sample & Two Related Samples

Comparing Means: t-tests for One Sample & Two Related Samples Comparing Means: -Tess for One Sample & Two Relaed Samples Using he z-tes: Assumpions -Tess for One Sample & Two Relaed Samples The z-es (of a sample mean agains a populaion mean) is based on he assumpion

More information

Vehicle Arrival Models : Headway

Vehicle Arrival Models : Headway Chaper 12 Vehicle Arrival Models : Headway 12.1 Inroducion Modelling arrival of vehicle a secion of road is an imporan sep in raffic flow modelling. I has imporan applicaion in raffic flow simulaion where

More information

Ensamble methods: Bagging and Boosting

Ensamble methods: Bagging and Boosting Lecure 21 Ensamble mehods: Bagging and Boosing Milos Hauskrech milos@cs.pi.edu 5329 Senno Square Ensemble mehods Mixure of expers Muliple base models (classifiers, regressors), each covers a differen par

More information

Object Tracking. Computer Vision Jia-Bin Huang, Virginia Tech. Many slides from D. Hoiem

Object Tracking. Computer Vision Jia-Bin Huang, Virginia Tech. Many slides from D. Hoiem Objec Tracking Compuer Vision Jia-Bin Huang Virginia Tech Man slides from D. Hoiem Adminisraive suffs HW 5 (Scene caegorizaion) Due :59pm on Wed November 6 oll on iazza When should we have he final exam?

More information

3.1.3 INTRODUCTION TO DYNAMIC OPTIMIZATION: DISCRETE TIME PROBLEMS. A. The Hamiltonian and First-Order Conditions in a Finite Time Horizon

3.1.3 INTRODUCTION TO DYNAMIC OPTIMIZATION: DISCRETE TIME PROBLEMS. A. The Hamiltonian and First-Order Conditions in a Finite Time Horizon 3..3 INRODUCION O DYNAMIC OPIMIZAION: DISCREE IME PROBLEMS A. he Hamilonian and Firs-Order Condiions in a Finie ime Horizon Define a new funcion, he Hamilonian funcion, H. H he change in he oal value of

More information

GMM - Generalized Method of Moments

GMM - Generalized Method of Moments GMM - Generalized Mehod of Momens Conens GMM esimaion, shor inroducion 2 GMM inuiion: Maching momens 2 3 General overview of GMM esimaion. 3 3. Weighing marix...........................................

More information

Západočeská Univerzita v Plzni, Czech Republic and Groupe ESIEE Paris, France

Západočeská Univerzita v Plzni, Czech Republic and Groupe ESIEE Paris, France ADAPTIVE SIGNAL PROCESSING USING MAXIMUM ENTROPY ON THE MEAN METHOD AND MONTE CARLO ANALYSIS Pavla Holejšovsá, Ing. *), Z. Peroua, Ing. **), J.-F. Bercher, Prof. Assis. ***) Západočesá Univerzia v Plzni,

More information

Planning in POMDPs. Dominik Schoenberger Abstract

Planning in POMDPs. Dominik Schoenberger Abstract Planning in POMDPs Dominik Schoenberger d.schoenberger@sud.u-darmsad.de Absrac This documen briefly explains wha a Parially Observable Markov Decision Process is. Furhermore i inroduces he differen approaches

More information

Chapter 14. (Supplementary) Bayesian Filtering for State Estimation of Dynamic Systems

Chapter 14. (Supplementary) Bayesian Filtering for State Estimation of Dynamic Systems Chaper 4. Supplemenary Bayesian Filering for Sae Esimaion of Dynamic Sysems Neural Neworks and Learning Machines Haykin Lecure Noes on Selflearning Neural Algorihms ByoungTak Zhang School of Compuer Science

More information

Shiva Akhtarian MSc Student, Department of Computer Engineering and Information Technology, Payame Noor University, Iran

Shiva Akhtarian MSc Student, Department of Computer Engineering and Information Technology, Payame Noor University, Iran Curren Trends in Technology and Science ISSN : 79-055 8hSASTech 04 Symposium on Advances in Science & Technology-Commission-IV Mashhad, Iran A New for Sofware Reliabiliy Evaluaion Based on NHPP wih Imperfec

More information

Introduction to Mobile Robotics Summary

Introduction to Mobile Robotics Summary Inroducion o Mobile Roboics Summary Wolfram Burgard Cyrill Sachniss Maren Bennewiz Diego Tipaldi Luciano Spinello Probabilisic Roboics 2 Probabilisic Roboics Key idea: Eplici represenaion of uncerainy

More information

Kinematics and kinematic functions

Kinematics and kinematic functions Kinemaics and kinemaic funcions Kinemaics deals wih he sudy of four funcions (called kinemaic funcions or KFs) ha mahemaically ransform join variables ino caresian variables and vice versa Direc Posiion

More information

Tom Heskes and Onno Zoeter. Presented by Mark Buller

Tom Heskes and Onno Zoeter. Presented by Mark Buller Tom Heskes and Onno Zoeer Presened by Mark Buller Dynamic Bayesian Neworks Direced graphical models of sochasic processes Represen hidden and observed variables wih differen dependencies Generalize Hidden

More information

References are appeared in the last slide. Last update: (1393/08/19)

References are appeared in the last slide. Last update: (1393/08/19) SYSEM IDEIFICAIO Ali Karimpour Associae Professor Ferdowsi Universi of Mashhad References are appeared in he las slide. Las updae: 0..204 393/08/9 Lecure 5 lecure 5 Parameer Esimaion Mehods opics o be

More information

Monocular SLAM Using a Rao-Blackwellised Particle Filter with Exhaustive Pose Space Search

Monocular SLAM Using a Rao-Blackwellised Particle Filter with Exhaustive Pose Space Search 2007 IEEE Inernaional Conference on Roboics and Auomaion Roma, Ialy, 10-14 April 2007 Monocular SLAM Using a Rao-Blackwellised Paricle Filer wih Exhausive Pose Space Search Masahiro Tomono Absrac This

More information

Non-parametric techniques. Instance Based Learning. NN Decision Boundaries. Nearest Neighbor Algorithm. Distance metric important

Non-parametric techniques. Instance Based Learning. NN Decision Boundaries. Nearest Neighbor Algorithm. Distance metric important on-parameric echniques Insance Based Learning AKA: neares neighbor mehods, non-parameric, lazy, memorybased, or case-based learning Copyrigh 2005 by David Helmbold 1 Do no fi a model (as do LTU, decision

More information

CHAPTER 10 VALIDATION OF TEST WITH ARTIFICAL NEURAL NETWORK

CHAPTER 10 VALIDATION OF TEST WITH ARTIFICAL NEURAL NETWORK 175 CHAPTER 10 VALIDATION OF TEST WITH ARTIFICAL NEURAL NETWORK 10.1 INTRODUCTION Amongs he research work performed, he bes resuls of experimenal work are validaed wih Arificial Neural Nework. From he

More information

Matlab and Python programming: how to get started

Matlab and Python programming: how to get started Malab and Pyhon programming: how o ge sared Equipping readers he skills o wrie programs o explore complex sysems and discover ineresing paerns from big daa is one of he main goals of his book. In his chaper,

More information

2.160 System Identification, Estimation, and Learning. Lecture Notes No. 8. March 6, 2006

2.160 System Identification, Estimation, and Learning. Lecture Notes No. 8. March 6, 2006 2.160 Sysem Idenificaion, Esimaion, and Learning Lecure Noes No. 8 March 6, 2006 4.9 Eended Kalman Filer In many pracical problems, he process dynamics are nonlinear. w Process Dynamics v y u Model (Linearized)

More information

On Solving the Perturbed Multi- Revolution Lambert Problem: Applications in Enhanced SSA

On Solving the Perturbed Multi- Revolution Lambert Problem: Applications in Enhanced SSA On Solving he Perurbed Muli- Revoluion Lamber Problem: Applicaions in Enhanced SSA John L. Junkins and Robyn M. Woollands Texas A&M Universiy Presened o Sacie Williams (AFOSR/RT) AFOSR REMOTE SENSING PORTFOLIO

More information

Diebold, Chapter 7. Francis X. Diebold, Elements of Forecasting, 4th Edition (Mason, Ohio: Cengage Learning, 2006). Chapter 7. Characterizing Cycles

Diebold, Chapter 7. Francis X. Diebold, Elements of Forecasting, 4th Edition (Mason, Ohio: Cengage Learning, 2006). Chapter 7. Characterizing Cycles Diebold, Chaper 7 Francis X. Diebold, Elemens of Forecasing, 4h Ediion (Mason, Ohio: Cengage Learning, 006). Chaper 7. Characerizing Cycles Afer compleing his reading you should be able o: Define covariance

More information

Lecture 2-1 Kinematics in One Dimension Displacement, Velocity and Acceleration Everything in the world is moving. Nothing stays still.

Lecture 2-1 Kinematics in One Dimension Displacement, Velocity and Acceleration Everything in the world is moving. Nothing stays still. Lecure - Kinemaics in One Dimension Displacemen, Velociy and Acceleraion Everyhing in he world is moving. Nohing says sill. Moion occurs a all scales of he universe, saring from he moion of elecrons in

More information

AUTONOMOUS SYSTEMS. Probabilistic Robotics Basics Kalman Filters Particle Filters. Sebastian Thrun

AUTONOMOUS SYSTEMS. Probabilistic Robotics Basics Kalman Filters Particle Filters. Sebastian Thrun AUTONOMOUS SYSTEMS robabilisic Roboics Basics Kalman Filers aricle Filers Sebasian Thrun slides based on maerial from hp://robos.sanford.edu/probabilisic-roboics/pp/ Revisions and Add-Ins by edro U. Lima

More information

A Bayesian Approach to Spectral Analysis

A Bayesian Approach to Spectral Analysis Chirped Signals A Bayesian Approach o Specral Analysis Chirped signals are oscillaing signals wih ime variable frequencies, usually wih a linear variaion of frequency wih ime. E.g. f() = A cos(ω + α 2

More information

MCMC-Based Particle Filtering for Tracking a Variable Number of Interacting Targets. Zia Khan, Tucker Balch, and Frank Dellaert

MCMC-Based Particle Filtering for Tracking a Variable Number of Interacting Targets. Zia Khan, Tucker Balch, and Frank Dellaert 1 MCMC-Based Paricle Filering for Tracking a Variable Number of Ineracing Targes Zia Khan, Tucker Balch, and Frank Dellaer 2 Absrac We describe a paricle filer ha effecively deals wih ineracing arges -

More information

Modal identification of structures from roving input data by means of maximum likelihood estimation of the state space model

Modal identification of structures from roving input data by means of maximum likelihood estimation of the state space model Modal idenificaion of srucures from roving inpu daa by means of maximum likelihood esimaion of he sae space model J. Cara, J. Juan, E. Alarcón Absrac The usual way o perform a forced vibraion es is o fix

More information

Simulation-Solving Dynamic Models ABE 5646 Week 2, Spring 2010

Simulation-Solving Dynamic Models ABE 5646 Week 2, Spring 2010 Simulaion-Solving Dynamic Models ABE 5646 Week 2, Spring 2010 Week Descripion Reading Maerial 2 Compuer Simulaion of Dynamic Models Finie Difference, coninuous saes, discree ime Simple Mehods Euler Trapezoid

More information

Robust estimation based on the first- and third-moment restrictions of the power transformation model

Robust estimation based on the first- and third-moment restrictions of the power transformation model h Inernaional Congress on Modelling and Simulaion, Adelaide, Ausralia, 6 December 3 www.mssanz.org.au/modsim3 Robus esimaion based on he firs- and hird-momen resricions of he power ransformaion Nawaa,

More information

PENALIZED LEAST SQUARES AND PENALIZED LIKELIHOOD

PENALIZED LEAST SQUARES AND PENALIZED LIKELIHOOD PENALIZED LEAST SQUARES AND PENALIZED LIKELIHOOD HAN XIAO 1. Penalized Leas Squares Lasso solves he following opimizaion problem, ˆβ lasso = arg max β R p+1 1 N y i β 0 N x ij β j β j (1.1) for some 0.

More information

Deep Learning: Theory, Techniques & Applications - Recurrent Neural Networks -

Deep Learning: Theory, Techniques & Applications - Recurrent Neural Networks - Deep Learning: Theory, Techniques & Applicaions - Recurren Neural Neworks - Prof. Maeo Maeucci maeo.maeucci@polimi.i Deparmen of Elecronics, Informaion and Bioengineering Arificial Inelligence and Roboics

More information

Non-parametric techniques. Instance Based Learning. NN Decision Boundaries. Nearest Neighbor Algorithm. Distance metric important

Non-parametric techniques. Instance Based Learning. NN Decision Boundaries. Nearest Neighbor Algorithm. Distance metric important on-parameric echniques Insance Based Learning AKA: neares neighbor mehods, non-parameric, lazy, memorybased, or case-based learning Copyrigh 2005 by David Helmbold 1 Do no fi a model (as do LDA, logisic

More information

Ensamble methods: Boosting

Ensamble methods: Boosting Lecure 21 Ensamble mehods: Boosing Milos Hauskrech milos@cs.pi.edu 5329 Senno Square Schedule Final exam: April 18: 1:00-2:15pm, in-class Term projecs April 23 & April 25: a 1:00-2:30pm in CS seminar room

More information

Lecture 33: November 29

Lecture 33: November 29 36-705: Inermediae Saisics Fall 2017 Lecurer: Siva Balakrishnan Lecure 33: November 29 Today we will coninue discussing he boosrap, and hen ry o undersand why i works in a simple case. In he las lecure

More information

20. Applications of the Genetic-Drift Model

20. Applications of the Genetic-Drift Model 0. Applicaions of he Geneic-Drif Model 1) Deermining he probabiliy of forming any paricular combinaion of genoypes in he nex generaion: Example: If he parenal allele frequencies are p 0 = 0.35 and q 0

More information

R t. C t P t. + u t. C t = αp t + βr t + v t. + β + w t

R t. C t P t. + u t. C t = αp t + βr t + v t. + β + w t Exercise 7 C P = α + β R P + u C = αp + βr + v (a) (b) C R = α P R + β + w (c) Assumpions abou he disurbances u, v, w : Classical assumions on he disurbance of one of he equaions, eg. on (b): E(v v s P,

More information

Distributed Particle Filters for Sensor Networks

Distributed Particle Filters for Sensor Networks Disribued Paricle Filers for Sensor Neworks Mark Coaes Deparmen of Elecrical and Compuer Engineering, McGill Universiy 3480 Universiy S, Monreal, Quebec, Canada H3A 2A7 coaes@ece.mcgill.ca, WWW home page:

More information

Vectorautoregressive Model and Cointegration Analysis. Time Series Analysis Dr. Sevtap Kestel 1

Vectorautoregressive Model and Cointegration Analysis. Time Series Analysis Dr. Sevtap Kestel 1 Vecorauoregressive Model and Coinegraion Analysis Par V Time Series Analysis Dr. Sevap Kesel 1 Vecorauoregression Vecor auoregression (VAR) is an economeric model used o capure he evoluion and he inerdependencies

More information

Final Spring 2007

Final Spring 2007 .615 Final Spring 7 Overview The purpose of he final exam is o calculae he MHD β limi in a high-bea oroidal okamak agains he dangerous n = 1 exernal ballooning-kink mode. Effecively, his corresponds o

More information

Application of a Stochastic-Fuzzy Approach to Modeling Optimal Discrete Time Dynamical Systems by Using Large Scale Data Processing

Application of a Stochastic-Fuzzy Approach to Modeling Optimal Discrete Time Dynamical Systems by Using Large Scale Data Processing Applicaion of a Sochasic-Fuzzy Approach o Modeling Opimal Discree Time Dynamical Sysems by Using Large Scale Daa Processing AA WALASZE-BABISZEWSA Deparmen of Compuer Engineering Opole Universiy of Technology

More information

Recent Developments In Evolutionary Data Assimilation And Model Uncertainty Estimation For Hydrologic Forecasting Hamid Moradkhani

Recent Developments In Evolutionary Data Assimilation And Model Uncertainty Estimation For Hydrologic Forecasting Hamid Moradkhani Feb 6-8, 208 Recen Developmens In Evoluionary Daa Assimilaion And Model Uncerainy Esimaion For Hydrologic Forecasing Hamid Moradkhani Cener for Complex Hydrosysems Research Deparmen of Civil, Consrucion

More information

Content-Based Shape Retrieval Using Different Shape Descriptors: A Comparative Study Dengsheng Zhang and Guojun Lu

Content-Based Shape Retrieval Using Different Shape Descriptors: A Comparative Study Dengsheng Zhang and Guojun Lu Conen-Based Shape Rerieval Using Differen Shape Descripors: A Comparaive Sudy Dengsheng Zhang and Guojun Lu Gippsland School of Compuing and Informaion Technology Monash Universiy Churchill, Vicoria 3842

More information