Designing Modular Robots

1
Designing Modular Robots
ERIC SCHWEIKARDTCCSL Lab Meeting April 28, 2008
2
(No Transcript)
3
(No Transcript)
4
(No Transcript)
5
(No Transcript)
6
Complexity

• Is different from complicated
• As a study, has a rich and varied history
• Can be modeled in many ways
• Process calculi
• Petri Nets
• Cellular Automata
• CGP

7
Complexity Microworlds
8
Embodiment

• Simons story about the ant
• Behavior based robots
• Feel free to think quietly about Heidegger and
  phenomenology

9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
Modular Robotics

• Almost all systems are homogeneous
• Serious, expensive research systems
• Many concentrate on self-reconfiguration
• There are some toys

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
roBlocks

• Is distributed
• Is a tangible programming system
• Is heterogeneous

17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
What values should A0 and A1 calculate?
21
A diffusion algorithm weighted average
22
A QUICK DEMO
23
(No Transcript)
24
Question Given a future in which we have lots of
pieces of robotic stuff, how are we going to put
it together into sensible constructions?
25
The Erstwhile Agent An application that will
design (evolve) roBlocks constructions based on
given fitness functions. So we can move beyond
simple vehicles.
26
The Erstwhile Agent
Population database SQL Server
Evolutionary algorithms Python or C
roBlock database SQL Server
Internet front end, user interface AJAX
Physical simulator ODE, Gazebo, or MSRS
evaluation
27
The Erstwhile Agent
Fitness Function One Forager
Fitness L / Population greatest L where L
Length of path traveled

• Different from coverage
• Check robot coordinate to prevent spinning in
  place spoof

28
The Erstwhile Agent
Fitness Function Two Scavenger
Fitness ( 1 / T ) / Population greatest
T where T Time to reach goal stimulus

• We can use an abstract stimulus, or since its
  roBlocks, a light source
• Its scary to think about robots that feed on
  carrion

29
The Erstwhile Agent
Fitness Function Three Elusive
Fitness Navoid / P where Navoid
Number of avoided obstacles P Total number of
projectiles

• Obstacle avoidance is a common task
• Fire projectiles to prevent sluggishness spoof

30
The Erstwhile Agent
and so what? The Erstwhile Agent will be a
success if it can design candidates that are as
fit or more fit than robots constructed by a
human designer. Ill be particularly pleased if
it creates solutions that are novel.
31
(No Transcript)
32

• THANKS
• - My advisors
• Mark D. Gross
• Illah Nourbakhsh
• Metin Sitti
• - The Computational Design Lab
• Gabe, Mike, Yeonjoo, Ben, Jet, Sora, Ellen,
  Susan
• National Science Foundation
• Sto Rox School District
• Katie Levedahl
• Team roBlocks
• Mark, Fred, Rob, Sandi, Anita, Ben, Andrew

Questions?
www.roblocks.org