David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

20
26 April 2002 Boucugnani, Barnhard, Caveney 1 Phoenix & Dragon: Parallax BoeBot / Lego Technics Hybrid Solutions to the Reactive Maze Traversal and Deliberative “Cheese” Location Problems David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia ACMSE2002 Student Paper Presentation

description

Phoenix & Dragon: Parallax BoeBot / Lego Technics Hybrid Solutions to the Reactive Maze Traversal and Deliberative “Cheese” Location Problems. David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia ACMSE2002 Student Paper Presentation. Outline. - PowerPoint PPT Presentation

Transcript of David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

Page 1: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 1

Phoenix & Dragon: Parallax BoeBot / Lego Technics Hybrid Solutions to the

Reactive Maze Traversal and Deliberative “Cheese” Location Problems

David Boucugnani, David Barnhard, David Caveney

The University of Georgia, Athens, GeorgiaACMSE2002 Student Paper Presentation

Page 2: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 2

Outline• 3 minute preview / review of important robotic

concepts• The Phoenix

– Sensing and other hardware components– Programming Pearls– Results of reactive maze traversal

• The Dragon– Additional sensory modifications– The problem of absolute positioning– Results of “cheese” location problem

• Problems / Future Directions

Page 3: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 3

3 minute preview/review of robotics

• What is robotics?– “A reprogrammable, multifunctional manipulator

designed to move material parts, tools or specialized devices through various programmed motions for the performance of various tasks.” Robotics Institute of America, 1979

– “An automatic device that performs functions normally ascribed to humans or a machine in the form of a human.” Webster

– Definitions we like better:• “Force Through Intelligence”• “Where AI Meets the Real World”

Page 4: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 4

Robotic Design Methodologies

• Two methodologies for an approach to robotics:– mechanical and electrical engineering approach– software and behavioral approach

• Autonomous robots can exhibit several different types of behavior or control schemes, two of which include:– reactive control– deliberative control

Page 5: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 5

Reactive Control Architecture

• “Don’t think, (re)act”• Tight coupling of sensors and actuators• Reacts immediately to its environment• Does not store any information (i.e., does

not form an ontology of the outside world)• Examples from nature:

– Trigger hairs on Venus fly traps– Ants defending a nest

Page 6: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 6

Deliberative Control Architecture

• “Think hard, then act”• “Deliberates” on next action to perform based on

previous and current sensory input– Requires an ontology or world representation

– A method of remembering previous sensory states

• World Modeling– Can range from the very simple (keeping track of

position) to very complex (building complex maps of large environments)

Page 7: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 7

The Phoenix - Introduction

• An exercise in trial and error

• Overall employs the reactive control architecture to traverse a maze

• Some deliberative components– Counts alternating touch sensor hits to detect

corners (in original version)– Exploits randomization to defeat looping

behavior

Page 8: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 8

Phoenix Evolution

Page 9: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 9

The Phoenix – Sensing and Structure

• Original version had two front touch sensors• Replaced with binary IR sensors and one dead-ahead,

loop touch sensor– Variable distance trigger based on software-set IR frequency

• Utilized Lego Technics parts for sensor mounting platform

• Breadboard attached to front for additional wiring space

Page 10: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 10

Phoenix Figures

Figure 2.1 –Rubber Band Gripping

Ends for Whiskers

Figure 2.2

Unpublished – One-bit Whisker Assembly

Figure 2.3.1 – IR Hanger

Page 11: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 11

Phoenix Programming Pearls• All programming in P-Basic, a subset of Basic• Employs bit-by-bit variable encoding to conserve

register space• Short and long range IR sensor sweeping

– Yields a rough estimate of object distance– Can be utilized to detect wall openings in wide corridors

• Smart turning– Determines best direction to turn if front touch sensor is

triggered– Based on frontal IR sensor values

Page 12: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 12

Reactive Maze Traversal Results

• The Phoenix was able to successfully traverse several simple mazes

• Traversal time was dependent on turn choices (the luck of the randomizer)

• Able to avoid all obstacles within range of the IR sensors

Page 13: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 13

The Dragon – Sensory and Structural Modifications

• The Dragon (in the physical sense) is a direct descendent (mutation) of the Phoenix– Twin photoresistors to detect the “cheese”, set

at a 3 inch trigger range to eliminate “homing” behavior

– “Ben Hur Chariot Spikes” with spring-and-loop touch sensors to count steps taken

Page 14: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 14

Dragon Figures

Figure 3.2 – Photoresistor / IR

Hanger

Figure 3.1 – Touch Sensor / Wheel Projection Combination assembly

Unpublished – Spring-Loop Sensor assembly

Unpublished – Failed Break Beam Sensor

Page 15: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 15

The Absolute Positioning Problem

• Positioning is crucial in any search and retrieve problem

• Possible solutions include:– GPS– Overhead camera– Triangulation based on tower emitters positioned

around the site

• However, these solutions require a connection to an outside sensor and/or computer, which is not in the spirit of autonomous robotics

Page 16: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 16

Dragon’s Positioning Solution

• Keep track of direction facing

• Count number of steps taken in this direction via “Ben Hur Chariot Spikes”

• Use trigonometry to convert resulting vector into vertical (and horizontal) components

Page 17: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 17

Cheese-Finding Results

• The Dragon can – search for the cheese (light source)

– record its vertical position

– return to the starting position

– recall the vertical position and move to the cheese’s location, searching along the horizontal if necessary

– avoid obstacles at each stage

• All with 2K of program / data space and 32 bytes of register space!

Page 18: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 18

Problems / Future Directions

• P-Basic too simple, lacks many convenient programming structures

• Need more memory to fully implement absolute positioning

• Would like to use a symbolic programming language (Prolog) for a more abstract and complex knowledge / ontological representation

• We will be moving to other platforms (Palm, Laptop) to implement robots with more complex behaviors, but still retain autonomousness

Page 19: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 19

Code and Presentation on Web

The source code for this project as well as HTML and Power Point versions of this presentation can be found online at:

ais.ai.uga.edu/boucugnani/robotics

Page 20: David Boucugnani, David Barnhard, David Caveney The University of Georgia, Athens, Georgia

26 April 2002 Boucugnani, Barnhard, Caveney 20

Questions?