39
PolySLAM A 2D Polygon-based SLAM Algorithm Johann Dichtl Xuan Sang Le Luc Fabresse Guillaume Lozenguez Noury Bouraqadi IMT Lille Douai, Univ. Lille http://car.imt-lille-douai.fr April 19, 2019 Dichtl et al PolySLAM April 19, 2019 1 / 39

PolySLAM - IMT Lille Douai

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: PolySLAM - IMT Lille Douai

PolySLAMA 2D Polygon-based SLAM Algorithm

Johann Dichtl Xuan Sang Le Luc FabresseGuillaume Lozenguez Noury Bouraqadi

IMT Lille Douai, Univ. Lillehttp://car.imt-lille-douai.fr

April 19, 2019

Dichtl et al PolySLAM April 19, 2019 1 / 39

Page 2: PolySLAM - IMT Lille Douai

Context

Our context: Multi-robot exploration

Requirements towards the map format:

lightweight map format for easy sharing between robots/agentssupport for navigationsupport for frontier-based exploration

Dichtl et al PolySLAM April 19, 2019 2 / 39

Page 3: PolySLAM - IMT Lille Douai

Context

State-of-the-Art Map Formats

Map format Lightweight Navigation support FrontiersGrid maps no yes only implicit

Feature-based yes no noVector-based yes no no

Dichtl et al PolySLAM April 19, 2019 3 / 39

Page 4: PolySLAM - IMT Lille Douai

Context

no common map format supports all threerequirements.

Dichtl et al PolySLAM April 19, 2019 4 / 39

Page 5: PolySLAM - IMT Lille Douai

PolyMap Format

PolyMap Format

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

Dichtl et al PolySLAM April 19, 2019 5 / 39

Page 6: PolySLAM - IMT Lille Douai

PolyMap Format

Placeholder Slide

ToDo:

make obstacle vectors thicker in figuresif time left: show the different SLAM steps step-by-step withfigures, starting with a pointcloud and ending with a (small)mapadd numbers to map examples (trajectory length, max. error)update last page according to noury’s comments

Dichtl et al PolySLAM April 19, 2019 6 / 39

Page 7: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM

PolyMap Merging

Polygon Refinement

Keyframe Creation

Data Alignment

Data Acquisitionpointcloud pose estimate

pointcloud

polygon

polygon/PolyMap

PolyMappose estimate

laser scan odometry

map

Dichtl et al PolySLAM April 19, 2019 7 / 39

Page 8: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM

PolyMap Merging

Polygon Refinement

Keyframe Creation

Data Alignment

Data Acquisitionpointcloud pose estimate

pointcloud

polygon

polygon/PolyMap

PolyMappose estimate

laser scan odometry

map

Dichtl et al PolySLAM April 19, 2019 8 / 39

Page 9: PolySLAM - IMT Lille Douai

PolySLAM

Data Acquisition

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

robot

Dichtl et al PolySLAM April 19, 2019 9 / 39

Page 10: PolySLAM - IMT Lille Douai

PolySLAM

Data Acquisition: collect raw sensor data

-0.5

0

0.5

1

1.5

2

-1 -0.5 0 0.5 1 1.5 2 2.5 3

Robot's local frame

Dichtl et al PolySLAM April 19, 2019 10 / 39

Page 11: PolySLAM - IMT Lille Douai

PolySLAM

Data Acquisition: transform from local to global frame

1

1.5

2

2.5

3

3.5

2 2.5 3 3.5 4 4.5

Global (map) frame

Dichtl et al PolySLAM April 19, 2019 11 / 39

Page 12: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM

PolyMap Merging

Polygon Refinement

Keyframe Creation

Data Alignment

Data Acquisitionpointcloud pose estimate

pointcloud

polygon

polygon/PolyMap

PolyMappose estimate

laser scan odometry

map

Dichtl et al PolySLAM April 19, 2019 12 / 39

Page 13: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - point-to-vector ICP

0

0.5

1

1.5

2

2.5

3

0 0.5 1 1.5 2 2.5 3 3.5 4

point-to-vector ICP: setup

robot

Dichtl et al PolySLAM April 19, 2019 13 / 39

Page 14: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - point-to-vector ICP

0

0.5

1

1.5

2

2.5

3

0 0.5 1 1.5 2 2.5 3 3.5 4

point-to-vector ICP: laser scan data

robot

Dichtl et al PolySLAM April 19, 2019 14 / 39

Page 15: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - point-to-vector ICP

0

0.5

1

1.5

2

2.5

3

0 0.5 1 1.5 2 2.5 3 3.5 4

point-to-vector ICP: match pairs

robot

Dichtl et al PolySLAM April 19, 2019 15 / 39

Page 16: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - point-to-vector ICP

0

0.5

1

1.5

2

2.5

3

0 0.5 1 1.5 2 2.5 3 3.5 4

point-to-vector ICP: aligned point cloud

robot

Dichtl et al PolySLAM April 19, 2019 16 / 39

Page 17: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - backface culling

0

0.5

1

1.5

2

2.5

1 1.5 2 2.5 3 3.5 4

ICP example - setup

robot

Dichtl et al PolySLAM April 19, 2019 17 / 39

Page 18: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - backface culling

0

0.5

1

1.5

2

2.5

1 1.5 2 2.5 3 3.5 4

ICP example - laser scan

robot

Dichtl et al PolySLAM April 19, 2019 18 / 39

Page 19: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - backface culling

0

0.5

1

1.5

2

2.5

1 1.5 2 2.5 3 3.5 4

ICP example - without backface culling

robot

Dichtl et al PolySLAM April 19, 2019 19 / 39

Page 20: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - backface culling

0

0.5

1

1.5

2

2.5

1 1.5 2 2.5 3 3.5 4

ICP example - with backface culling

robot

Dichtl et al PolySLAM April 19, 2019 20 / 39

Page 21: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM - backface culling

0

0.5

1

1.5

2

2.5

1 1.5 2 2.5 3 3.5 4

ICP example - with backface culling

robot

Dichtl et al PolySLAM April 19, 2019 21 / 39

Page 22: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Keyframe Craetion

PolyMap Merging

Polygon Refinement

Keyframe Creation

Data Alignment

Data Acquisitionpointcloud pose estimate

pointcloud

polygon

polygon/PolyMap

PolyMappose estimate

laser scan odometry

map

Dichtl et al PolySLAM April 19, 2019 22 / 39

Page 23: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Keyframe Craetion

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

robot

Dichtl et al PolySLAM April 19, 2019 23 / 39

Page 24: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Keyframe Craetion

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

robot

19 vectors

Dichtl et al PolySLAM April 19, 2019 24 / 39

Page 25: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Polygon Refinement

PolyMap Merging

Polygon Refinement

Keyframe Creation

Data Alignment

Data Acquisitionpointcloud pose estimate

pointcloud

polygon

polygon/PolyMap

PolyMappose estimate

laser scan odometry

map

Dichtl et al PolySLAM April 19, 2019 25 / 39

Page 26: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Polygon Refinement

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

robot

19 vectors

Dichtl et al PolySLAM April 19, 2019 26 / 39

Page 27: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Polygon Refinement

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

robot

8 vectors

Dichtl et al PolySLAM April 19, 2019 27 / 39

Page 28: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Merging

PolyMap Merging

Polygon Refinement

Keyframe Creation

Data Alignment

Data Acquisitionpointcloud pose estimate

pointcloud

polygon

polygon/PolyMap

PolyMappose estimate

laser scan odometry

map

Dichtl et al PolySLAM April 19, 2019 28 / 39

Page 29: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Merging

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

Dichtl et al PolySLAM April 19, 2019 29 / 39

Page 30: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Merging

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

Dichtl et al PolySLAM April 19, 2019 30 / 39

Page 31: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Merging

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

Dichtl et al PolySLAM April 19, 2019 31 / 39

Page 32: PolySLAM - IMT Lille Douai

PolySLAM

PolySLAM: Merging

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5

Dichtl et al PolySLAM April 19, 2019 32 / 39

Page 33: PolySLAM - IMT Lille Douai

Experiments

Experiments - Simulator

Dichtl et al PolySLAM April 19, 2019 33 / 39

Page 34: PolySLAM - IMT Lille Douai

Experiments

Experiments - Real Robots

-20

-15

-10

-5

0

5

-10 -5 0 5 10 15

otf=0.4, ost=1.2, mi=50, st=0.0009, od=0.3distance traveled=339.7, time spent=28m 45.27s

Dichtl et al PolySLAM April 19, 2019 34 / 39

Page 35: PolySLAM - IMT Lille Douai

Experiments

Experiments - Real Robots

-10 0

10

20

30

40 -2

0-1

5-1

0-5

0 5

10

15

otf

=0

.5,

ost=

0.2

, m

i=4

0,

st=

0.0

00

4,

od=

0.2

dis

tance t

ravele

d=

69

.0,

tim

e s

pent=

3m

47

.57

s

Dichtl et al PolySLAM April 19, 2019 35 / 39

Page 36: PolySLAM - IMT Lille Douai

Experiments

Experiments - Real Robots

-5

0

5

10

15

-5 0 5 10 15 20 25 30 35 40

Dichtl et al PolySLAM April 19, 2019 36 / 39

Page 37: PolySLAM - IMT Lille Douai

Conclusion

Conclusion

ToDo...

Dichtl et al PolySLAM April 19, 2019 37 / 39

Page 38: PolySLAM - IMT Lille Douai

Outlook

Outlook

upcoming paper (IntelliSys2019): topological graphs onPolyMapsfurther optimizations, e.g. pose graph optimization &cross-keyframe polygon refinement

Dichtl et al PolySLAM April 19, 2019 38 / 39

Page 39: PolySLAM - IMT Lille Douai

Thanks

Thanks

Dichtl et al PolySLAM April 19, 2019 39 / 39