Click here to load reader

SLAM - Survey: Simultaneous Localization and Mapping · PDF fileUniversity of Hamburg MIN Faculty Department of Informatics SLAM SLAM Survey: Simultaneous Localization and Mapping

  • View
    212

  • Download
    0

Embed Size (px)

Text of SLAM - Survey: Simultaneous Localization and Mapping · PDF fileUniversity of Hamburg MIN...

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    SLAM

    SLAMSurvey: Simultaneous Localization and Mapping

    Martin Poppinga

    University of HamburgFaculty of Mathematics, Informatics and Natural SciencesDepartment of Informatics

    Technical Aspects of Multimodal Systems

    23. November 2015

    Martin Poppinga 1

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    SLAM

    Outline

    1. Motivation

    2. History

    3. Basics

    4. Implementations

    5. Conclusion

    Martin Poppinga 2

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Motivation SLAM

    Outline

    1. Motivation

    2. History

    3. Basics

    4. Implementations

    5. Conclusion

    Martin Poppinga 3

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Motivation SLAM

    Mobile Robotics

    I (Partly) autonomous SystemsI Many applications

    I SAR (Search and Rescue)I Exploration (Areal, Underwater, Space)I Service

    I ChallengesI MappingI Localization

    Martin Poppinga 4

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Motivation SLAM

    SLAM

    I ChallengeI Map LocalizationI Localization MapI Chicken-egg problem

    I SLAM brings this togetherI Different approaches

    I FilteringI Sensors

    Martin Poppinga 5

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    History SLAM

    Outline

    1. Motivation

    2. History

    3. Basics

    4. Implementations

    5. Conclusion

    Martin Poppinga 6

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    History SLAM

    History

    I First stepsI In mid 80sI Mapping and localizationI Limited in computation power

    I BreakthroughI In mid 90sI Convergence of errorsI Mapping and localization togetherI Demonstration on real systems

    I Wide interest in 2000s

    Martin Poppinga 7

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Outline

    1. Motivation

    2. History

    3. Basics

    4. Implementations

    5. Conclusion

    Martin Poppinga 8

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Definition

    I Existing informationI robots Controls / odometry

    I UT = {u1, u2, u3, ...uT}I Observations

    I ZT = {z1, z2, z3, ...zT}I Needed information

    I Map (with its features)I m

    I Path of the robotI XT = {x0, x1, x2, ...xT}

    Martin Poppinga 9

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Graphical Model

    [3]

    Martin Poppinga 10

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Probabilistic SLAM

    I World is not perfect!I Full SLAM

    I p(XT ,m|ZT ,UT )I Online SLAM

    I p(xt ,m|ZT ,UT )I This has to be estimated

    I Different problems different estimators

    I Choosing one based on the problem

    Martin Poppinga 11

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Different Problems

    I Static vs dynamic

    I Volumetric vs feature based

    I Topologic vs metric

    I Known vs unknown correspondence

    I Large vs small uncertainty

    I Active vs passive

    I Single- vs multiagent

    I Any time and any space

    I Lots of different approaches

    Martin Poppinga 12

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Sensing

    I RangeI Laser Range sensorsI SonarI TactileI ...

    I VisualI CameraI 3D Camera

    I OtherI Wi-fiI SoundI ...

    Martin Poppinga 13

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Algorithm

    I Handling the errorsI LocationI Landmark sensoring

    I Features in m

    I Measurement model

    I Motion model

    I Three main filter types

    I Will be partly presented in IR lecture

    Martin Poppinga 14

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Basics SLAM

    Filters

    I Kalman FilterI The original technique in SLAMI Reduction of errorsI Mathematical model

    I Particle FilterI Sequential Monte CarloI Particle for possible locations

    I Graph Based

    Martin Poppinga 15

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Implementations SLAM

    Outline

    1. Motivation

    2. History

    3. Basics

    4. Implementations

    5. Conclusion

    Martin Poppinga 16

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Implementations SLAM

    EKF SLAM

    I First variants of SLAM

    I Kalman-filter based

    I Standard kalman filter

    I Provisional landmark list

    Martin Poppinga 17

  • [3]

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Implementations SLAM

    FastSLAM

    I Particle Filter

    I Each particle one Position

    I Rao-Blackwellization

    I Independent features

    I No revising of path on the fly

    I Performance

    I Widely used

    Martin Poppinga 19

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Implementations SLAM

    GraphSLAM

    I Builds graphI MovementI Observations

    I Flexible edges

    Martin Poppinga 20

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Implementations SLAM

    Which to Use?

    I DependsI EKF SLAM

    I Quadratic with landmarksI Big maps problematic (submaps)

    I GraphSLAMI Elegant solutionI Full SLAM / offline

    I FastSLAMI Data associationI Efficient

    Martin Poppinga 21

  • University of Hamburg

    MIN Faculty

    Department of Informatics

    Implementations SLAM

    DARPA

    I US military research

    I Self driving cars in desert

    I GPS not precise enough

    I Stanley, winner grand challenge 2005

    Mar