A Survey of Localization Methods

1
A SurveyofLocalization Methods

• Presented to CS694
• November 19, 1999
• Jeremy Elson

2
whats the problem?

• WHERE AM I?
• But what does this mean, really?
• Frame of reference is important
• Local/Relative Where am I vs. where I was?
• Global/Absolute Where am I relative to the
  world?
• Location can be specified in two ways
• Geometric Distances and angles
• Topological Connections among landmarks

3
localization relative

• If you know your speed and direction, you can
  calculate where you are relative to where you
  were (integrate).
• Speed and direction might, themselves, be
  absolute (compass, speedometer), or integrated
  (gyroscope, accelerometer)
• Relative measurements are usually more accurate
  in the short term -- but suffer from accumulated
  error in the long term
• Most robotics work seems to focus on this. This
  talk will focus on absolute localization.

4
localization absolute

• Proximity-To-Reference
• Landmarks/Beacons ParcTab, Active Badges
• Angle-To-Reference
• Visual manual triangulation from physical points
• Radio VOR
• Distance-From-Reference
• Time of Flight
• RF GPS, PinPoint
• Acoustic Active Bat, Lew
• Signal Fading
• EM Bird/3Space Tracker
• RF SCADDS/SCOWR, Niru
• Acoustic Jer?

5
topological maps

• Really the most natural how did you get to
  class today?
• You have a map of known landmarks and the
  connections among them
• You even convert metric maps to topological!
• Probably the most useful for location-aware
  computing
• Closest printer really means the one in this
  room, not on the other side of a wall

6
topological localization

• ParcTab and Active Badges
• Infrared transmission picked up by recivers in
  all rooms
• Works precisely because infrared propagation
  matches topological boundaries of environment
• Reverse is also possible landmarks
• SCADDS localization beacons
• Problems difficult to control granularity apps
  may need geometric map

7
triangulation
Land
Landmarks
Works great -- as long as there are reference
points!
Lines of Sight
Unique Target
Sea
8
compass triangulation
cutting-edge 12th century technology
Land
Landmarks
Lines of Sight
North
Unique Target
Sea
9
celestial navigation

• Same idea, except in 1D, and reference point is a
  star
• Angle of between north star and horizon
  determines latitude
• Works only because north star is close to axis of
  Earths rotation
• Longitude is much harder
• Note Points to non-flatness of earth

Encyclopedia Britannica
10
VOR modern triangulation

• VOR is an aircraft navigation system still widely
  prevalent today
• Same concept as visual landmarks, except that
  radio beacons emit directional signals
• Aircraft can determine (within 1o) their bearing
  to a VOR station
• 1 VOR fix will tell you bearing-to-target 2
  tells you absolute position

http//www.interlog.com/bitewise/aviation/navplan
/radionav.htm
11
vor for localization
2 VORs plus a map will uniquely determine 2-D
position
12
VORs magic

• VOR stations transmit two signals
• An omnidirectional reference signal, with a 30 Hz
  amplitude modulation
• A highly directional continuous signal that
  sweeps through 360 degrees at 30Hz
• Result aircraft sees two sine waves reference
  modulated by transmitter, azimuth signal
  modulated by directionality
• Receiver computes phase shift between them to get
  bearing

13
distance-to-reference systems

• Measure distance from ref point to target
• For n dimensions, n measurements give you 2
  solns n1 is unique
• Domain knowledge can often be used instead of
  n1th measurement

14
accuracy constraints

• Accuracy depends on
• Precision of the distance measurements
  (represented below as thickness of the circles)
• Geometric configuration of the reference points

Reference points far apart small overlapping
region
Reference points close together large
overlapping region
15
measuring distance

• Measure time-of-flight
• Biggest problem time synchronization
• Time sync and localization are often intertwined
• If only Einstein was wrong, and information could
  travel instantaneously...
• GPS, PinPoint, Active Bat all deal with the time
  problem in different ways
• Measure signal strength
• Used less often because relationship between
  strength and distance is harder to model (also
  not linear)

16
gps global positioning system

• 24 satellites launched by U.S. DOD, originally
  for weapons systems targeting
• Gives time position anywhere in the world,
  although often only outdoors
• Typical Position Accuracy
• Civilian Horiz 50m, Vert 78m, 3D 93m, 200ns
• Diff Horiz 1.3m, Vert 2.0m, 3D 2.8m, 350ns
• Military accuracy might be usable in 2000

http//tycho.usno.navy.mil/gpsinfo.html
http//www.trimble.com/gps/howgps/gpsfram2.htm
17
the basic idea

• Satellites constantly transmit beacons along with
  the time-of-beacon and position (in predictable,
  corrected, and observed orbits)
• Receivers listen for (phase-shifted) signals and
  compute distance based on propagation delay
  (assume magically synced clocks for now)
• 3 satellites gives you 2 points (in 3d) throw
  out the one in deep space
• Compute position relative to satellites use
  satellite position to get Earth coordinates

18
effects of clock bias
true distance
biased distance
19
solving for clock bias

• Critical point satellites are perfectly
  synchronized (using expensive atomic clocks
  synchronized before launch)
• If all signals are received simultaneously, they
  are all off by a constant bias
• This means that by adding an additional
  satellite, we can solve for clock bias. (Would
  not work if off by a constant factor)
• This gives us both position and time!

20
solving for clock bias
true distance
biased distance
21
sources of gps errorper satellite

• Satellite clock drift (1.5 m) (1usec 300m)
• Orbit estimation errors (2.5 m)
• Atmospheric and relativistic effects (5.5 m)
• Receiver noise (0.3 m)
• Multipath interference (0.6 m)
• Intentional randomization to reduce civilian
  grade accuracy (30m)

http//www.trimble.com/gps/howgps/gpsfram2.htm
22
differential gps

• A way of getting more accurate GPS data
• Receivers at known positions find the difference
  between computed true position
• Computed error correction factor transmitted to
  other GPS receivers in the area
• Corrects for all errors that the receiver has in
  common with the reference (atmospheric,
  relativistic, orbital, sat clock, randomization)

23
pinpoint 3d-id

• Local positioning system by Pinpoint Co.
• Meant to track large numbers tags indoors
• Tags should be cheap and all have IDs
• Infrastructure knows where tags are tags dont
  know anything
• Compare to GPS Infrastructure knows nothing,
  tags know where they are
• 1-3 m accuracy

http//www.pinpointco.com/_private/whitepaper/rfid
.html
24
the clock problem

• Their solution
• Interrogator transmits a test signal
• Tag simply changes the signals frequency and
  transponds it back to the interrogator (with tag
  ID modulated in)
• Interrogator receives transponded signal
• Subtracting out fixed system delays yields time
  of flight
• They avoid the clock sync problem by making the
  transmitter and receiver the same device

25
implementation details

• Area to be monitored is divided up into cells -
  each with antennas controller
• Coarse-grained location first (which cell?), then
  fine-grained location within the cell
• Query driven Tag 5 raise your hand!, or
• Tag driven all tags periodically beacon (impl.)
• Tag beaconing frequency might depend on inertial
  system
• Collision reduction through various techniques,
  including reducing beacon time
• They note non-linear increase in perf due to this

26
active bats

• Research project at ORL-cum-ATT
• Similar goals as Pinpoint indoor LPS

100mm x 60mm x 20mm
http//www.uk.research.att.com/bat/
27
bats at work

• Tags have unique IDs, radio receivers and
  ultrasound transducers
• Interrogator consists of a radio transmitter and
  microphones (ultrasound detectors)
• Interrogator sends radio message Tag 5, signal
  now!
• Tag 5 receives the radio message and sends an
  ultrasonic pulse
• Microphones pick up the sound time of flight
  calculated

28
the clock problem

• Use two modalities RF for control (very fast),
  sound for measurement (slow)
• We can simulate instantaneous info flow because
  it is almost instant relative to what were
  measuring
• Speed of sound 344 m/s
• Speed of light 300M m/s (30m 0.1 usec)
• 0.1 usec 344m/s 0.000 034 4 m
• Like Pinpoint, subtract out fixed delays
  (empirically derived) to get flight time

29
implementation details

• Multiple peaks may be detected (echoes - audio
  version of multipath interference)
• Two heuristics for eliminating echos
• Difference in distance between two measurements
  cant be larger than the distance between the two
  microphones.
• If so, larger one must be a reflection
• Do statistical tests to identify outliers repeat
  until variance is low or only 3 points remain
• Nice extension use 3 tags to detect 3d pose as
  well as position of objects

30
active bat accuracy
95 within 14cm for raw measurments 95 within
8cm when averaged over 10 samples
ftp//ftp.uk.research.att.com/pub/docs/att/tr.97.1
0.pdf
31
active lew-bats

• Goal distance between two robots
• One robot simultaneously
• Sends a message over the network to the target
  robot
• Emits an audio chirp from the sound card
• Target robot
• Waits for network message
• Listens for chirp, calculates time of flight
• Evaluation in progress

32
distance measurementusing signal fading

• Another class of localization systems uses
  reduction in the strength of a field to measure
  distance
• Magnetic Field Ascension Flock of Birds,
  Polaris 3space tracker
• RF No (??) commercial products work here on
  SCADDS/SCOWR
• Sound A half baked idea of mine

33
flock of birds

• Measures 3D position
  and orientation
• Consists of largish transmitter small receiver
  connected to the same controller
• Receiver picks up orthogonal magnetic fields from
  transmitter (details unknown)
• Specs claim 0.02/0.1o precision over 10 area
• Not really that good and metal screws it up
• Magnetic field falls off as r4 (?)
• Mostly head tracking apps similar

http//www.ascension-tech.com/products/flockofbird
s
34
radio signal strength

• Work going on here (SCADDS, SCOWR Nirupama
  Bulusu, Puneet Goel)
• Can radio signal strength be used as a reliable
  distance measurement?
• Very difficult to model indoor radio prop.
• Current test implementation
• Radiometrix RPC radio transmitter
• RxM receiver module with RSSI output pin

35
an initial test
Signal Strength Indicator
Distance in Meters
Nirupama Bulusu and Puneet Goel
36
sound off

• Half baked idea can we measure falloff in audio
  volume as a distance estimate?
• ...I told you it was half baked, thats all I
  have to say about that )

37
thats all, folks!
And, remember wherever you go, there you are.