Localization Using WiFi

1
Localization Using WiFi

• Julie Letchner
• Benson Limketkai
• (with Karthik Gopalratnam)

2
Introduction

• What is localization?
• Why is localization useful?
• Robotic applications
• Wider scope Assisted Cognition project

3
Traditional Robot Localization

• Object of interest robot
• Sensor laser range finder (LRF),
• sonar, camera
• Motion wheel encoders (well-defined model)

4
Particle Filter Basics
Motion model
Sensor model
Belief at previous time step
Normalizer
5
Particle Filter Basics

• Particle state ltx, y, headinggt

6
Access Points in Allen Center
7
Our Sensor Model

• Sensor observations radio signal strength
  vectors
• Where do these vectors come from?
• Attenuation models
• Direct measurement

8
Direct Measurement
9
Interpolated Signal Strength
10
Our Motion Model

• People move noisily and in no well-defined model
• Additional constraint voronoi graph
• New particle state
• Graph edge
• 1D position
• Gaussian velocity distribution

11
Allen Center Voronoi Graph
12
Setup

• Robot Pioneer 2 AT
• Control Saphira
• Signal strength
• readings Kismet
• Particle filtering
• Lin Liaos code

13
Experimental Results

• Data collected on different days / times
• Accuracy measured on the order of meters vs.
  centimeters (LRF)

14
Training Data Results
15
Test Data Caveat Missing APs!
16
Test Data Caveat Missing APs!
17
Test Data Results
18
Conclusions

• APs were not designed for localization
• Extra software required for strength readings
• High redundancy required for accuracy
• Signal maps must be built explicitly by
  measurement
• Signal strength is highly variable
• Wireless card documentation is lacking

19
Future Work

• Attenuation models
• Signal interpolation methods
• Freespace constraints for non-corridor spaces

20
Bonus Track Human Data