relative bearing estimation using commodity radios
DESCRIPTION
Relative Bearing Estimation using Commodity Radios. Karthik Dantu 1 Prakhar Goyal 2 Gaurav S. Sukhatme 1. 1 Dept of Computer Science University of Southern California Los Angeles, CA - 90089-2905. 2 Dept of Computer Science and Engg. Indian Institute of Technology-Bombay Mumbai - 400237. - PowerPoint PPT PresentationTRANSCRIPT
Relative Relative Bearing Bearing
Estimation Estimation using using
Commodity Commodity Radios Radios
Karthik Dantu1
Prakhar Goyal2
Gaurav S.
Sukhatme1
1Dept of Computer Science University of Southern CaliforniaLos Angeles, CA - 90089-2905
2Dept of Computer Science and Engg.Indian Institute of Technology-Bombay
Mumbai - 400237
22/21/21
What is Relative Bearing?
A
B
AB
BA
33/21/21
Uses of Relative Bearing
• Robot Localization (Briechle 04, Chintalapudi 04, Das 02,
Martinelli 05, Niculsecu 03, Spletzer 01, Taylor 07)
• Navigation (Bekris 04, Ducatelle 08, O’Hara 08)
• Topology Control (Eren 03, Li 05, Poduri 08)
• Formation Control (Das 02, Mostagh 08, Spletzer 01)
• Pursuit-Evasion Games (Karnad 08, Maloy 95)
44/21/21
Measuring Relative Bearing
Directional sensor array
Transmitter array (acoustic, radio)
Vision
Bumper array
55/21/21
Radio as a Sensor
• Signal strength roughly
correlated with distance
between sender and receiver
• Most modern robots have
off-the-shelf radios
• Radio characteristics are well
studied
66/21/21
Large Scale Fading
• Radio behavior over
large distances (>> )
• Correlated to distance
• Modeling less reliable
for shorter distances
and very close to
transmitter
77/21/21
Small Scale Fading
Blue signal travels 1/2 farther than red to reach receiver, who receives purple
Sender Receiver
• Signal variability on the scale of
• Multipath effects dominate (reflection, refraction, diffraction, scattering)
• Mobility introduces Doppler effects
• ~ 12cm for 2.4 GHz
88/21/21
Large Scale Fading Models
• Free Space Model: Models signal strength on a clear
unobstructed link
LossdB=20log(d) + 20 log(f) + C
• Log Distance Path Loss Model: Logarithmic path loss
model with Path Loss Exponent () for the particular
medium
LossdB= PL(d0)+ 10log(d/d0) + XBg
• ITU Indoor Model: Takes into account the frequency of
transmission and floors between sender and receiver
LossdB= 20log(f) + Nlog(d) + Lf(n) + K
Introduction to RF Propagation, John S. Seybold, Wiley-Interscience.
99/21/21
Estimating Bearing Using Radio
• Consider only large scale fading effects
• Sample signal strength in the locality of robot
• Perform Principal Component Analysis (PCA)
• Primary component is the direction of maximum
variance of signal strength
• Relative bearing of robot is approximated to this
direction
1010/21/21
Bearing Estimation Algorithm
A
B
S - step sizeS
45° CCW
1111/21/21
Step Distance
• Step distance is a parameter
• Greater step distances improve signal gradient
but odometry error and area of deployment
are constraints
• From our signal strength measurements, for a
signal strength loss of 20dB step size is 6m
outdoors and 2m indoors
1212/21/21
Simulation Setup
• Simulated an area of 100m x 100m
• Two robots are randomly placed in the given
area
• Parameters
• Step distance
• Number of samples collected
• AWGN Noise added to samples collected
• Results are averaged over 100 trials
1313/21/21
Effect of Step Distance Variation
6m
1414/21/21
Effect of Number of Samples
100 samples
1515/21/21
Effect of Noise
1616/21/21
Experimental Setup
iRobot Create
~3 ft
Telos B Mote for ZigBee radio
Wi-Fi Antenna
E-box with Intel 800Mhz PC with 802.11 Wi-Fi card
1717/21/21
Bearing Error Outdoors
1818/21/21
Bearing Error Indoors
1919/21/21
Outdoor Multi-robot Experiments (5 robots)
Edge Actual
Angle
Estimated
Angle
Bearing
Error
1 120° 103.7 ° 12.3 °
2 135° 104.8° 30.2°
3 45° 58.8° 13.8°
4 -30° -11.8° 18.2°
5 -150° -130.7° 19.3°
6 0° 25.8° 25.8°
Average error over two trials was 19.1°
2020/21/21
Indoor Multi-robot Experiments (5 robots)
Edg
e
Actual Angle Estimate
d Angle
Estimatio
nError
1 5° 9.3° 4.3°
2 120° 148.1° 28.1°
3 -160° -147.2° 12.81°
4 170° 202.5° 32.5°
5 -30° -18.3° 11.7°
6 150° 172.1° 22.1°
Average error over 5 trials was 24.3°
2121/21/21
Conclusions
• Relative bearing can be estimated using commodity radios
• Tested algorithm in simulation and experiment
(ZigBee and Wi-Fi)
• Used this estimation as input for connectivity algorithm
• ZigBee radios perform better than Wi-Fi on average
• Average error is approximately 20° indoors and 25°
outdoors using ZigBee radios
• Future work: Exploit small scale effects
2222/21/21
Backup Slides
2323/21/21
Discussion
• Use
2424/21/21
Bearing Estimation Algorithm
A
B
S - step sizeS
2525/21/21
Bearing Estimation Algorithm
A
B
S - step sizeS
45 CCW