Sonar Sensors Implementation

1
Sonar Sensors Implementation
Ryosuke Sakai Nicole K. Takahashi
2
Contents

• Two Methods that deal with sonar resolution.
• ATM Method
• Arc Transversal Median Method (H. Choset, K.
  Nagatani)
• McKerrow Method
• Echolocation (P. McKerrow)

3
Sonar Sensor Property Problem

• Good in measuring distance
• Low azimuth resolution
• ? causes problem, especially at narrow openings

• Nomad 200
• 16 Polaroid ultrasonic sensors
• 22.5 deg per sonar

Beam Pattern
4
Naïve Sensor Model Center Line Model

• Assume the point of reflection is at the center
  of the beam pattern
• the model we usually use in the class now
• Max 11.25 deg error occurs

5
Center Line Model Problem (1)

• Center Line Model does not work in the presence
  of narrow opening (using wide angled sonar
  detectors)

Passing a corridor
Neglecting the corner
No Problem!
Overlook the Gap!
6
Center Line Model Problem (2)

• The robot might believe the passageway is too
  narrow for it to pass through

b
a
7
Center Line Model Real Data

• Center Line Model gives the robot a false
  impression of the world
• The robot perceives that an opening does not
  exist!!!

8
Solution for Azimuth Error

• Update Sonar Hardware ?
• The number of Sonars (16 ? 32 ? 64)
• High cost
• Use Laser Range System ?
• Good both in Distance and Azimuth
• But high cost
• Laser cannot detect Grass and Dark Black Obstacle
• ? Develop a New Algorithm for Sonar ?

9
ATM Method Intro.

• ATM Method (Arc Transversal Median Method), by
  H. Choset and K. Nagatani
• 3 Properties
• Arc Intersection
• Median Intersection
• Transversal Intersection

10
ATM Method (1) Arc Intersection

• Arc Intersection of two or more arcs
• The point of reflection can lie anywhere on arc
  1, arc 2, , arc n
• If many sonar arcs intersect at one point, the
  probability that it is in an obstacle becomes
  quite high

arc2
arc1
11
ATM Method (2) Median Intersection

• The median of all intersections on one arc serves
  as a good approximation of the point
• Median is in the cluster of intersections
• Median is robust with respect to noise

No intersections
One intersection
Two or more intersections
Errors ?Take a Median
Measurement Error? Dead-Reckoning? Some Noise?
Assume on Center Line
Assume at the Intersection
12
ATM Method (3) Transversal Intersection

• Robot do not consider all intersections, just
  Stable intersections

a) Stable intersection
b) Unstable intersection
OK!
No Good!
a) is more STABLE than b). Because ? How big
does have to be?
13
ATM Method (3) Transversal Intersection Why
30 deg?

• It is enough for our Nomad robots to use 30 deg
  as the threshold for transversal intersections

At least a 10-fold improvement in resolution!!
(at d100 inch)
14
ATM Method Test how it works

• The points that ATM method generates present a
  more accurate view of the environment

Center Line Model
ATM Method