Learning Roomba

1
Learning Roomba

• Module 5 - Localization

2
Outline

• What is Localization?
• Why is Localization important?
• Why is Localization hard?
• Some Approaches
• Using the Fixed-Camera Localization
• Exercises

3
What is Localization?

• Process of determining where a robot is in its
  environment
• Absolute position in the world
• Position in a map
• Relative to the start position
• Relative to some landmark
• Relative to another robot
• Some other way of identifying location

4
What is Localization?

• Can be done with or without a map of the
  environment
• A map contains information about the environment
  that the robot can use to help localize itself

5
Why is Localization important?

• Robot needs to know where it is to know how to
  act
• Without knowing where you are, you can only
  wander around randomly
• Cannot navigate to a location without knowing
  where you are
• Many tasks rely on navigation

6
Why is Localization Hard?

• Robot needs to know about its enviornment
• Sensors are not always accurate
• Noisy
• Unreliable
• Sensors do not tell the Robot what it needs to
  know
• Many sensors report distance to an object, not
  where the Robot is

7
Some Approaches

• Odometry
• Global Positioning Systems
• Landmarks
• Trilateration
• Distance Sensors
• Others

8
Using the Fixed-Camera Localization

• getLocalizedX()
• Returns the Localized X-position of the Roomba in
  Meters
• getLocalizedY()
• Returns the Localized Y-position of the Roomba in
  Meters

9
Using the Fixed-Camera Localization

• X and Y positions are grid locations in the world

Diagram taken from BBCs website
10
Principles of the Fixed-Camera Localization

• Merges 3 Concepts
• Odometry Readings
• Topological Map
• Image Tracking

11
Principles of the Fixed-Camera Localization

• Odometry Readings
• Recorded from the sensors on the Roombas wheels
• Map can be built over time of where the Roomba
  has visited.
• Only as accurate as the sensors
• Accuracy worsens over time due to drift and
  skidding

12
Principles of the Fixed-Camera Localization

• Topological Map
• Type of Graph
• Vertices are interesting features in the
  environment
• Edges are transitions between the features
• Usually able to determine when an already visited
  feature has been reached

13
Principles of the Fixed-Camera Localization

• Image Tracking
• Able to find the Roomba in an image
• Looks for an area with an average color similar
  to the average color of the Roomba
• Tracked points in the image become points of
  interest (Vertices) in the Topological map
• Camera does not move, so the same pixel
  represents the same location of the Roomba

14
Principles of the Fixed-Camera Localization

• As the Roomba drives in the Environment, the
  algorithm builds the Topological Map
• Vertices are tracked image coordinates
• Edges are odometry readings to get from one image
  coordinate to another
• The Topological Map can be used to determine the
  position of the Roomba
• Finds the current image coordinates where the
  Roomba is
• Find the shortest path between the start position
  and the current position through the Topological
  Map

15
Exercises

• Report the position of a virtual wall beam. To
  do so, have the robot drive until it detects a
  virtual wall. Print the localized positions once
  it has reached the beam.

16
Exercises

• Write a program that will have the Roomba drive
  in a square 10 times with each side of the square
  being 1 meter long. Print out the X-Position and
  Y-Position according to Odometry and localization

17
Questions?