Python Robotics: An Environment for Exploring Robotics Beyond LEGOs

1
Python Robotics An Environment for Exploring
RoboticsBeyond LEGOs

• Douglas Blank
• Bryn Mawr College, USA
• dblank_at_cs.brynmawr.edu
• Lisa Meeden
• Swarthmore College, USA
• meeden_at_cs.swarthmore.edu
• Deepak Kumar
• Bryn Mawr College, USA
• dkumar_at_mojo.brynmawr.edu

2
NSF CCLI Grant

• Beyond LEGOs Hardware, Software, and Curriculum
  for the Next Generation Robot Laboratory.
  DUE-0231363.
• Principal Investigators
• Douglas Blank, Bryn Mawr College
• Kurt Konolige, SRI International
• Deepak Kumar, Bryn Mawr College
• Lisa Meeden, Swarthmore College
• Holly Yanco, U. Mass.-Lowell

3
Why pay attention?
If...

• you don't have a PhD in robotics...
• but, would be interested in teaching a robotics
  course...
• and, could use a mentor, and some funding to help
  buy a sophisticated robot...
• or, wonder what robotics is like Beyond LEGOS...

...then stick around
4
Pyro Project An Overview

• Motivation
• Project in context
• Details something for everybody
• Sophisticated, medium-cost robots
• Impact
• Conclusion

5
"Programming with a Purpose"

• Bryn Mawr and Swarthmore Colleges
• Small, liberal arts schools
• Interdisciplinary biased (geo and bio
  informatics, cognitive science, visualization,
  scientific modeling, etc)
• Research biased

6
"Robotics wouldn't work in our department."

• Karel, Jeroo, LOGO, Alice, hunting the wumpus,
  etc all robot control problems
• We are not "roboticists" in the traditional
  sense we're interested in intelligence (our "big
  idea", the "purpose")
• Push the big ideas in so that they are the
  motivations for learning to program and "doing"
  computer science

7
"Ok, LEGOs will do just fine."

• LEGO limitations? "Only your imagination"
• Limited sensors. Vision is a great motivation to
  study 2D arrays, but requires a camera.
• More complex models require more memory and
  faster CPU (for example, neural networks,
  developing area maps, planning, etc)
• Real AI and robotics research opportunities

8
Real RoboticsReal Painful Robotics

• Sophisticated, medium-cost robotics are now
  available (ActivMedia, iRobot, K-Team, and many
  more) however
• Vertical learning curve
• Use their proprietary programming environment and
  control software, or write your own
• Control software usually tightly integrated to a
  particular framework

9
Pyro Python Robotics

• Core written in Python
• Set of libraries and objects in Python
• API and GUI
• Open Source
• Easy for beginners to pick up
• Extendible

10
Pyro Architecture
11
What is Python?

• 10 years old developed with learners in mind
• Clean syntax Interpreted, but fast
• Supports multiple styles of programming
  (procedural, event oriented, threads, object
  oriented, and some functional)
• Support for most of the modern standards (XML,
  SOAP, OpenGL, HTTP, etc, etc, etc, etc, etc)
  through libraries

12
from math import sqrt class Point def
__init__(self, myX, myY) self.x myX
self.y myY def distance(self,
otherPoint) return sqrt( (self.x -
otherPoint.x) 2 (self.y
- otherPoint.y) 2 ) class Line def
__init__(self, pointA, pointB) self.a
pointA self.b pointB self.dist
pointA.distance( pointB ) p1 Point(5, 6) p2
Point(11, 23) line1 Line(p1, p2) print "Line
is ", line1.dist, "meters long."
Python Example
Note - No curly braces, just indentation -
First argument is "this" - Constructor has
funny name __init__ - Automatic typing - Also
supports easy multiple inheritance
13
Pyro Example
from pyro.brain import Brain from random import
random from time import sleep class
SimpleBrain(Brain) def step(self)
robot self.getRobot() left
robot.get('range', 'value', 'left')1
right robot.get('range', 'value', 'right')1
front robot.get('range', 'value',
'front')1 print "left", left, "front",
front, "right", right if (left lt 1 and
right lt 1) robot.move(0, .2)
elif (right lt 1) robot.move(0,
.2) elif (left lt 1)
robot.move(0, -.2) elif (front lt 1)
if random() lt .5 robot.move(0,
.2) else robot.move(0,
-.2) else robot.move(.2, 0)
14
Sophisticated, medium-priced Robots
15
Pyro could work with vastly Different Types of
Robots
16
Pyro could work with vastly Different Types of
Robots
17
Khepera from K-Team

• Hocky-puck sized
• Tethered
• Requires small area
• Delicate
• 8 Infrared/8 Light
• Fixed camera
• From 1.2k - 3k

18
Pioneer2 from ActivMedia

• From 30 - 70 lbs.
• On-board PC
• Needs bigger area
• Robust
• 16 sonar
• PTZ camera
• From 1k - 20k

19
Khepera with Camera
60x40
20
A Pyro-based Curriculum

• Direct control
• Logic
• Subsumption

• Introduction
• Reactive control
• Behavior-based
• Vision
• Learning
• Mapping
• Multi-robot communication

21
Pyro Support Libraries/Objects

• Backpropagation neural networks
• Self-organizing maps
• Fuzzy logic, behavior blending
• Vision library
• Camera objects, with Pan-Tilt-Zoom
• Evolutionary systems

22
Prototype XRCL

• Extensible Robot Control Language
• XML plus C
• All the complexity of C and XML combined
• Multiple levels of parse errors
• Different comment syntax for different areas
• Too difficult for novice programmers (and expert
  programmers!)

23
Pyro PerformanceUpdates per second

• Bare brain with console
• Bare brain with OpenGL 3D
• ANN with 3D
• Fuzzy logic 3D
• Many ANNs Vision 3D

• 10,000 / second
• 1,000 / second
• 200 / second
• 20 / second
• lt 1 / second

24
Issues

• Open source software is dynamic
• Pyro depends on many independent open source
  components (Python, Image Library, OpenGL, SWIG,
  Numeric, ...)
• Interdisciplinary connections require few
  prerequisites
• Sophisticated robots are still relatively
  expensive

25
Acknowledgments

• Thanks to
• Ananya Misra
• Evan Moses
• Daniel Sproul
• Cassandra Telenko

emergent.brynmawr.edu
This work was supported in part by a grant from
the 2002 Bryn Mawr-Haverford-Swarthmore Mellon
Tricollege Fellowship program