Maze Solving with an AIBO

Document Sample
Maze Solving with an AIBO Powered By Docstoc
					Maze Solving with an AIBO

  Bernard Maassen, Hans Kuipers,
   Max Waaijers & Andrew Koster
               2005
Introduction
   Problem:
       Maze Navigation
   Performed Research:
       Theseus found his way out of the Labyrinth
       Using IR for Maze Navigation, CMU
            www.cs.cmu.edu/~tekkotsu/media/pgss_2004_paper.pdf
       Many competitions
       Many geometric algorithms for edge
        recognition and L-shapes
            Computational Geometry: Algorithms and Applications
Introduction
   Why?
       Many aspects, like:
            Vision
            World modeling
            Self localization
       Socially relevant:
            Rescue robots need to navigate maze-like
             environments
Problem Description
   Maze Navigation
   3 Problems:
       Landmark detection
       Map construction
       AIBO uses map to solve maze
   Collision prevention/detection
Landmark detection
   Edge detection using scanlines
   Only look below ‘horizon’
Landmark detection
   Possible Intersections
Landmark detection
   Possible Intersections
Landmark detection
   Possible Intersections
Map construction
   On each landmark update graph
   Remember
       Type of landmark
       Current location
   Use depth first search to explore
   Initially simple mazes, later on more
    complex ones.
Complications
   Maze contains loops
       Need to use distances as well as type of
        intersection
   Different mazes
       Curves
       5-way intersections
       Other
AIBO uses map to solve maze
   Random initial location in maze
   Use Bayesian filters to determine most
    likely location
   Find exit
Possible problems
   Missing landmarks
   Walking into walls
   Odometry not reliable
Backup plan
   Reinforcement learning with joystick
       Simpler
       Uses joystick to train
       Uses odometry data in stead of vision
Milestone 1
   Joystick walking
       Already in Tekkotsu
       Integrate into DARPA modules
       Collect odometry data
Milestone 2a
   Control point if 2b is feasible
   If not extend Joystick module
Milestone 2b
   Landmark detection
       Vision module
       Edge detection + scanlines
       Distinguish intersections
Milestone 3
   Map construction
       Model world as topological map
       Self localization
       Walking through the maze
Milestone 4
   Maze Solving
       Find place in world
       Use map to find path to exit
       Exit maze
    Milestones
   Milestone   1: 30-9
   Milestone   2a: 19-10
   Milestone   2b: 26-10
   Milestone   3: 1-11
   Milestone   4: 11-11
Vragen?

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:5
posted:9/1/2012
language:English
pages:20