Maze Solving with an AIBO

Post on 15-Jan-2016

39 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

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 - PowerPoint PPT Presentation

Transcript

Maze Solving with an AIBO

Bernard Maassen, Hans Kuipers, Max Waaijers &

Andrew Koster2005

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.pd

f

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?

top related