Integrating SLAM into Gas Distribution Mapping Achim Lilienthal, - - PowerPoint PPT Presentation

▶
integrating slam into gas distribution mapping
SMART_READER_LITE
LIVE PREVIEW

Integrating SLAM into Gas Distribution Mapping Achim Lilienthal, - - PowerPoint PPT Presentation

Sep 29, 2023 312 likes •777 views

Integrating SLAM into Gas Distribution Mapping Achim Lilienthal, Amy Loutfi AASS, Dept. of Technology, rebro University Jose Luis Blanco, Cipriano Galindo and Javier Gonzalez System Engineering & Automation Dept., University of Malaga

slide-1
SLIDE 1

Integrating SLAM into Gas Distribution Mapping

Achim Lilienthal, Amy Loutfi

AASS, Dept. of Technology, Örebro University

Jose Luis Blanco, Cipriano Galindo and Javier Gonzalez

System Engineering & Automation Dept., University of Malaga

slide-2
SLIDE 2

Achim J. Lilienthal

Gas Distribution Mapping

→ Contents

slide-3
SLIDE 3

Achim J. Lilienthal

Gas Distribution Mapping

slide-4
SLIDE 4

Achim J. Lilienthal

Localization (SLAM)

slide-5
SLIDE 5

Achim J. Lilienthal

Localization (SLAM)

slide-6
SLIDE 6

Achim J. Lilienthal

Gas Distribution Mapping + SLAM

+

position, position uncertainty

slide-7
SLIDE 7

Achim J. Lilienthal

  • 1. Applications of Gas Distribution Modelling?
  • 2. The Challenges for Gas Distribution Mapping
  • 3. Kernel Based Gas Distribution Mapping
  • 4. Integrating SLAM and Gas Distribution Mapping
  • 5. Experiments and Results
  • 6. Summary

Contents

slide-8
SLIDE 8

Achim J. Lilienthal

Applications of Gas Distribution Modelling

→ Contents

slide-9
SLIDE 9

Achim J. Lilienthal

Gas Distribution Modelling

 Applications

 Oil Refinery Surveillance

1

slide-10
SLIDE 10

Achim J. Lilienthal

Gas Distribution Modelling

 Applications

 Oil Refinery Surveillance  Garbage Dump Site Surveillance

1

slide-11
SLIDE 11

Achim J. Lilienthal

Gas Distribution Modelling

 Applications

 Oil Refinery Surveillance  Garbage Dump Site Surveillance  Pollution Monitoring

 air quality monitoring and surveillance of pedestrian areas  communicating pollution levels to technical staff / pedestrians

1

slide-12
SLIDE 12

Achim J. Lilienthal

Gas Distribution Modelling

 Applications

 Oil Refinery Surveillance  Garbage Dump Site Surveillance  Pollution Monitoring

 air quality monitoring and surveillance of pedestrian areas  communicating pollution levels to technical staff / pedestrians

 Disaster Prevention

1

slide-13
SLIDE 13

Achim J. Lilienthal

Gas Distribution Modelling

 Applications

 Oil Refinery Surveillance  Garbage Dump Site Surveillance  Pollution Monitoring

 air quality monitoring and surveillance of pedestrian areas  communicating pollution levels to technical staff / pedestrians

 Disaster Prevention  Rescue Robots  ...

1

slide-14
SLIDE 14

Achim J. Lilienthal

Gas Distribution Mapping in Natural Environments – The Challenges

→ Contents

slide-15
SLIDE 15

Achim J. Lilienthal

Gas Distribution Mapping – Challenges

 Chaotic Gas Distribution

 diffusion  advective transport  turbulence

2

video by Hiroshi Ishida

slide-16
SLIDE 16

Achim J. Lilienthal

Gas Distribution Mapping – Challenges

 Chaotic Gas Distribution  Point Measurement

 sensitive sensor surface is typically small (often ≈ 1cm2)

2

slide-17
SLIDE 17

Achim J. Lilienthal

Gas Distribution Mapping – Challenges

 Chaotic Gas Distribution  Point Measurement  Sensor Dynamics

2

slide-18
SLIDE 18

Achim J. Lilienthal

Gas Distribution Mapping – Challenges

 Chaotic Gas Distribution  Point Measurement  Sensor Dynamics  Calibration

 complicated "sensor response ↔ concentration" relation  dependent on other variables (temperature, humidity, ...)  has to consider sensor dynamics  variation between individual sensors  long-term drift

2

slide-19
SLIDE 19

Achim J. Lilienthal

Gas Distribution Mapping – Challenges

 Chaotic Gas Distribution  Point Measurement  Sensor Dynamics  Calibration  Real-Time Gas Distribution Mapping

 changes at different time-scales

 rapid fluctuations  slow changes of the overall structure of the average distribution

2

slide-20
SLIDE 20

Achim J. Lilienthal

Kernel Based Gas Distribution Mapping

→ Contents

slide-21
SLIDE 21

Achim J. Lilienthal

Kernel Based Gas Distribution Mapping

 General Gas Distribution Mapping Problem

 given the robot trajectory

 Differences to Range Sensing

 calibration: readings do not correspond directly to concentration levels

3

) , | (

: : t 1 gas t 1 gas

z x m p

t 1

x :

slide-22
SLIDE 22

Achim J. Lilienthal

Kernel Based Gas Distribution Mapping

 General Gas Distribution Mapping Problem

 given the robot trajectory

 Differences to Range Sensing

 readings don't correspond directly to concentration levels  chaotic gas distribution: an instantaneous snapshot

  • f the gas distribution contains little information

about the distribution at other times

3

t 1

x :

) , | (

: : t 1 gas t 1 gas

z x m p

slide-23
SLIDE 23

Achim J. Lilienthal

Kernel Based Gas Distribution Mapping

 General Gas Distribution Mapping Problem

 given the robot trajectory

 Differences to Range Sensing

 readings don't correspond directly to concentration levels  instantaneous gas distribution snapshots contain little information about the distribution at other times  point measurement: a single gas sensor measurement provides information about a very small area (≈ 1cm2)

3

t 1

x :

) , | (

: : t 1 gas t 1 gas

z x m p

slide-24
SLIDE 24

Achim J. Lilienthal

Kernel Based Gas Distribution Mapping

 Time-Averaged Gas Distribution Mapping Problem

 given the robot trajectory

 Kernel Based Gas Distribution Mapping

 interpret gas sensor measurements zt as random samples from a time-constant distribution

 assumes time-constant structure of the observed gas distribution  randomness due to concentration fluctuations (measurement noise negligible)

⇒ kernel to model information content of single readings

3

) , | (

: : t 1 gas t 1 av gas

z x m p

t 1

x :

→ Achim Lilienthal and Tom Duckett. "Building Gas Concentration Gridmaps with a Mobile Robot". Robotics and Autonomous Systems, Vol. 48, No. 1, pp. 3-16, August 2004.

slide-25
SLIDE 25

Achim J. Lilienthal

Kernel Based Gas Distribution Mapping

 Analogy to Density Function Estimation

 estimate the PDF of a random variable (Parzen window)  K ← 2D univariate Gaussian

3

( )

∑

=

− =

N i PW

K Nh x p

1

|; | 1 ) ( ˆ σ

i

x x

from Duda, Hart, Stork "Pattern Classification"

slide-26
SLIDE 26

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping

→ Contents

slide-27
SLIDE 27

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping  General SLAM problem

 simultaneously estimate the map and the robot path given robot actions u and observations z

 Simultaneous Localisation and Gas Distribution / Occupancy Mapping ("GasSLAM")

4

) , | , (

: : : t 1 t 1 t t 1

z u m x p

( )

t

  • cc

t gas t t

z z z

, , ,

= ← z

( )

  • cc

av gas m

m m , = ← m

slide-28
SLIDE 28

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping  The GasSLAM Problem

 general approach: inverse sensor model to estimate maps

4

slide-29
SLIDE 29

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping  The GasSLAM Problem

 useful factorization if maps can be analytically estimated given a robot path hypothesis

4

) , , | ( ) , | ( ) , | , (

: : : : : : : : : t 1 t 1 t 1 t t 1 t 1 t 1 t 1 t 1 t t 1

z u x m p z u x p z u m x p =

⇒

estimate robot path using a particle filter compute maps analytically

Rao-Blackwellization, Rao-Blackwellized Particle Filter (RBPF)

slide-30
SLIDE 30

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping  GasSLAM – Map Computation

  • bservations zocc and zgas are conditionally independent

 assume independency between mocc and mgas

4

slide-31
SLIDE 31

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping  GasSLAM – Map Computation

  • bservations zocc and zgas are conditionally independent

 assume independency between mocc and mgas ⇒ computing maps separately for each particle

  • ccupancy grid map using sensor integration

[Moravec/Elfes 1985]  gas distribution grid map using kernel based gas distribution mapping [Lilienthal/Duckett, 2004]

 determine max. likelihood estimate of the maps from the weighted average (using particle weights)

4

slide-32
SLIDE 32

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping  GasSLAM – Estimation of the Robot Path

 sample from the motion model  update weights with the observation model  higher weights for particles that better corresponds with the current observations

4

) , | ( ~

t [i] 1 t t [i] t

u x x p x

−

) , | (

1 [i] [i] t t [i] t [i] t

m x z p

−

∝ω ω

slide-33
SLIDE 33

Achim J. Lilienthal

Integrating SLAM and Gas Distribution Mapping  GasSLAM – Estimation of the Robot Path

  • bservation model

4

) , , | , ( ) , | (

, , [i]

  • cc

[i] gas [i] t t

  • cc

t gas [i] [i] t t

m m x z z p m x z p = ) , | ( ) , | (

, , [i]

  • cc

[i] t t

  • cc

[i] gas [i] t t gas

m x z p m x z p = ) , | (

, [i]

  • cc

[i] t t

  • cc

m x z p η ≈

use only the laser scanner to estimate the path

  • cc

gas

  • cc
slide-34
SLIDE 34

Achim J. Lilienthal

Experiments and Results

→ Contents

slide-35
SLIDE 35

Achim J. Lilienthal

Experiments  Service Robot Sancho

 base: Pioneer 3DX  laser range finder: SICK LMS 200  pair of e-noses

5

slide-36
SLIDE 36

Achim J. Lilienthal

Experiments  Service Robot Sancho

 base: Pioneer 3DX  laser range finder: SICK LMS 200  pair of e-noses

 Electronic Nose

 4 metal-oxide gas sensors (Figaro): TGS 2600 [x2], TGS 2602, TGS 2620  sensors in a tube with CPU fan  sampling frequency: 1.25 Hz  separation: 14 cm; height: 11 cm

5

slide-37
SLIDE 37

Achim J. Lilienthal

Experiments  Environment

 University of Malaga, Computer Science building 

  • ne indoor and one outdoor corridor

 no modification for the experiment

 Gas Source

 evaporating ethanol  robot could drive

  • ver the source

(cup, height = 6 cm)

5

slide-38
SLIDE 38

Achim J. Lilienthal

Results  Result – SLAM

 robot speed: 5 cm/s  trajectory: sweeping

5

  • max. likelihood path
slide-39
SLIDE 39

Achim J. Lilienthal

Results  Result – Gas Distribution Map

 lighter shading ↔ higher concentration  different shading color for values > 80% of the max.

5

gas distribution map

slide-40
SLIDE 40

Achim J. Lilienthal

Results  Result – Gas Distribution Map

 lighter shading ↔ higher concentration  different shading color for values > 80% of the max.

5

gas distribution map

slide-41
SLIDE 41

Achim J. Lilienthal

Summary

→ Contents

slide-42
SLIDE 42

Achim J. Lilienthal

Experiments  Summary

 conceptual framework to integrate kernel based gas distribution mapping and SLAM

5

slide-43
SLIDE 43

Achim J. Lilienthal

Experiments  Summary

 conceptual framework to integrate kernel based gas distribution mapping and SLAM  large gas distribution map (20 x 2 m2)

5

slide-44
SLIDE 44

Achim J. Lilienthal

Experiments  Summary

 conceptual framework to integrate kernel based gas distribution mapping and SLAM  large gas distribution map (20 x 2 m2)  uncontrolled environment  combined indoor / outdoor environment

5

slide-45
SLIDE 45

Integrating SLAM into Gas Distribution Mapping

Achim Lilienthal, Amy Loutfi

AASS, Dept. of Technology, Örebro University

Jose Luis Blanco, Cipriano Galindo and Javier Gonzalez

System Engineering & Automation Dept., University of Malaga

Thank you!