Spring Localization I. Roland Siegwart, Margarita Chli, Martin Rufli. ASL Autonomous Systems Lab. Autonomous Mobile Robots

Transcription

1 Spring 2017 Localizaion I Localizaion I

2 2 ASL Auonomous Sysems Lab knowledge, daa base mission commands Localizaion Map Building environmen model local map posiion global map Cogniion Pah Planning pah Percepion Informaion Exracion raw daa Sensing see-hink-ac Pah Execuion acuaor commands Acing Moion Conrol Real World Environmen Localizaion I

3 Inroducion Do we need o localize or no? To go from A o B, does he robo need o know where i is? Localizaion I

4 Inroducion Do we need o localize or no? How o navigae beween A and B navigaion wihou hiing obsacles deecion of goal locaion Possible by following always he lef wall However, how o deec ha he goal is reached Localizaion I

5 Inroducion Do we need o localize or no? Following he lef wall is an example of behavior based navigaion I can work in some environmens bu no in all Wih which accuracy and reliabiliy do we reach he goal? Localizaion I

6 Inroducion Do we need o localize or no? As opposed o behavior based navigaion is map based navigaion Assuming ha he map is known, a every ime sep he robo has o know where i is. How? If we know he sar posiion, we can use wheel odomery or dead reckoning. Is his enough? Wha else can we use? Bu how do we represen he map for he robo? And how do we represen he posiion of he robo in he map? Localizaion I

7 Inroducion Definiions Global localizaion The robo is no old is iniial posiion Is posiion mus be esimaed from scrach Posiion Tracking A robo knows is iniial posiion and only has o accommodae small errors in is odomery as i moves? Localizaion I

8 Inroducion How o localize? Localizaion based on exernal sensors, beacons or landmarks Odomery Map Based Localizaion - wihou exernal sensors or arificial landmarks, jus use robo onboard sensors Example: Probabilisic Map Based Localizaion Localizaion I

9 Inroducion Beacon Based Localizaion Triangulaion Ex 1: Poles wih highly reflecive surface and a laser for deecing hem Ex 2: Coloured beacons and an omnidirecional camera for deecing hem (example: RoboCup or auonomous robos in ennis fields) Localizaion I

10 Inroducion Beacon Based Localizaion KIVA Sysems, Boson (MA) (acquired by Amazon in 2011) Unique marker wih known absolue 2D posiion in he map Prof. Raff D'Andrea, ETH Localizaion I

11 Inroducion Moion Capure Sysems High resoluion (from VGA up o 16 Mpixels) Very high frame rae (several hundreds of Hz) Good for ground ruh reference and muli-robo conrol sraegies Popular brands: VICON (10kCHF per camera), OpiTrack (2kCHF per camera) Localizaion I

12 Inroducion Map-based localizaion Consider a mobile robo moving in a known environmen. Localizaion I

13 Inroducion Map-based localizaion Consider a mobile robo moving in a known environmen. As i sars o move, say from a precisely known locaion, i can keep rack of is moion using odomery. Localizaion I

14 Inroducion Map-based localizaion Consider a mobile robo moving in a known environmen. As i sars o move, say from a precisely known locaion, i can keep rack of is moion using odomery. Localizaion I

15 Inroducion Map-based localizaion Consider a mobile robo moving in a known environmen. As i sars o move, say from a precisely known locaion, i can keep rack of is moion using odomery. Sensor reference frame Localizaion I

16 Inroducion Map-based localizaion Consider a mobile robo moving in a known environmen. As i sars o move, say from a precisely known locaion, i can keep rack of is moion using odomery. The robo makes an observaion and updaes is posiion and uncerainy Sensor reference frame Localizaion I

17 Ingrediens Probabilisic Map-based localizaion Probabiliy heory error propagaion, sensor fusion Belief represenaion discree / coninuous (map/posiion) Moion model Sensing odomery model measuremen model Localizaion I

18 Probabilisic localizaion Belief Represenaion Coninuous map wih single hypohesis probabiliy disribuion Kalman Filer Localizaion Coninuous map wih muliple hypoheses probabiliy disribuion Discreized meric map (grid ) wih probabiliy disribuion Discreized opological map (nodes ) wih probabiliy disribuion Markov Localizaion A B C D E F G Localizaion I

19 Belief Represenaion Characerisics Coninuous Discree Precision bound by sensor daa Typically single hypohesis pose esimae Los when diverging (for single hypohesis) Compac represenaion and ypically reasonable in processing power. Precision bound by resoluion of discreisaion Typically muliple hypohesis pose esimae Never los (when diverges converges o anoher cell) Imporan memory and processing power needed. (no he case for opological maps) Localizaion I

20 Odomery ASL Auonomous Sysems Lab Definiion Dead reckoning (also deduced reckoning or odomery) is he process of calculaing vehicle's curren posiion by using a previously deermined posiion and esimaed speeds over he elapsed ime Robo moion is recovered by inegraing propriocepive sensor velociies readings Pros: Sraighforward Cons: Errors are inegraed -> unbound Heading sensors (e.g., gyroscope) help o reduce he accumulaed errors bu drif remains Localizaion I

21 ASL Auonomous Sysems Lab Localizaion I Odomery The Differenial Drive Robo 21 ), ( ˆ 1 1 u x f y x x x y x x x y

22 ASL Auonomous Sysems Lab Kinemaics Localizaion I 22 Odomery Wheel Odomery ) 2 sin( ) 2 cos( ), ( ˆ s s y x u x f x 2 s r s l s b s s l r Can you demonsrae hese equaions? This erm comes from he applicaion of he Insananeous Cener of Roaion

23 ASL Auonomous Sysems Lab Error model Localizaion I 23 Odomery Error Propagaion l l r r S s k s k 0 0 T S S S T x x x F F F F P x F x f F S

24 Odomery Growh of Pose uncerainy for Sraigh Line Movemen Noe: Errors perpendicular o he direcion of movemen are growing much faser! Localizaion I

25 Odomery Growh of Pose uncerainy for Movemen on a Circle ASL Auonomous Sysems Lab Noe: Errors ellipse does no remain perpendicular o he direcion of movemen! Localizaion I

26 Odomery Example of non-gaussian error model Noe: Errors are no shaped like ellipses! Couresy AI Lab, Sanford [Fox, Thrun, Burgard, Dellaer, 2000] Localizaion I

27 Odomery Error sources Deerminisic (Sysemaic) Non-Deerminisic (Non-Sysemaic) Deerminisic errors can be eliminaed by proper calibraion of he sysem. Non-Deerminisic errors are random errors. They have o be described by error models and will always lead o uncerain posiion esimae. Major Error Sources in Odomery: Limied resoluion during inegraion (ime incremens, measuremen resoluion) Misalignmen of he wheels (deerminisic) Unequal wheel diameer (deerminisic) Variaion in he conac poin of he wheel (non deerminisic) Unequal floor conac (slippage, non planar ) (non deerminisic) Localizaion I

28 Odomery Calibraion of sysemaic errors [Borensein 1996] The unidirecional square pah experimen ASL Auonomous Sysems Lab Localizaion I

29 Odomery Calibraion of Errors II [Borensein 1996] The bi-direcional square pah experimen BILD 2/3 Borensein Localizaion I