Cognitive Robot Bodies

1
Cognitive Robot Bodies

• Kasper Stoy
• The Maersk McKinney Moller Institute
• University of Southern Denmark
• kaspers_at_mmmi.sdu.dk

2
Overview

• The body-brain relationship in cognition
• Modular robots as cognitive bodies
• Self-reconfigurable robots
• Deformable modular robots
• Hierarchical robots
• Conclusion

3
Brain-Body Relationship in Robotics

• Brain (controller)
• Task-optimized

• Body (hardware)
• General-purpose

) to a large extend
4
Body-Brain in Nature

• Brain
• Task-optimized

• Body
• Task-optimized

5
Example Body-Brain in Nature

• Why do bat ears have different shapes?

Example and graphics provided by Ralf Müller
(CIRCE EU project, University of Southern Denmark)
6
Bat Ears

• Bat ears cut off and digitized
• Interaction between sound waves and ear analyzed
  in simulation
• Robot bat built to verify and use results

7
Bat Simulation Results
8
Lessoned learned

• The bats ear shape in combination with the
  frequency of its scream determines the direction
  in which it is listening
• The physical structure of the bats ear does some
  processing for its brain
• The body of the bat aids cognition

9
Body-Brain in Robots

• We have a choice of where to implement
  intelligence
• Brain
  Body

Steven H. Collins, Cornell University
Asimov, Honda
10
Brain-Body in Robotics

• Brain (controller)
• Task-optimized

• Body (hardware)
• Task-optimized!

Roomba, IRobot
11
Difficult to make cognitive robot bodies

• Prototyping mechanics and electronics is
• Time-consuming
• Expensive
• Requires skilled people
• A task-optimized body is rarely versatile
• Something else is needed..

12
Modular Robots

• A modular robot is built from robotic modules
  that are connected to form a robot
• Features of modular robot
• Task-optimized body
• Versatile
• Robust
• Cheap
• Modular robots may make it easier to find a
  balance between body and brain

13
Shape-Changing Modular Robots

• If robots can change shape autonomously
• Continually task-optimizing body
• Versatility
• Robustness (impact and self-repair)
• Cheap?

14
Three implementations

• Self-reconfigurable robots
• Modules move around
• Deformable Robots
• Modules deform
• Hierarchical Robots
• Chains of modules tangle

15
Self-Reconfigurable Robots
16
Self-Reconfigurable Robots

• Modular robots that can change shape by
  automatically rearranging the way modules are
  connected

x4
The ATRON robot, D. Brandt and D. Christensen,
University of Southern Denmark
17
Other Self-Reconfigurable Robots
M-TRAN, Distributed System Design Research Group,
Intelligent Systems Institute, AIST
CONRO, Polymorphic Robotics Lab, Information
Sciences Institute, USC
PolyBot, Modular Robotics Group, PARC
18
Self-Reconfigurable Robots

• Research challenges
• Mechatronics
• Control

19
Mechatronics Challenge

• A module is an autonomous robot
• ATRON Module Characteristics
• Connection mechanisms
• Infra-red communication (with neighbours)
• Actuation (rotate one half sphere with respect to
  the other)
• Onboard batteries
• Processing power
• short-range sensors (infra-red)
• ...

20
Software Challenge

• Control a swarm robot where
• Individual modules are physically coupled
• The connection topology is dynamic
• Preferable control should be distributed to allow
• Scalability
• Robustness

21
The State of Self-Reconfigurable Robots

• Vision
• Versatile
• Robust
• Cheap
• Task-Optimized

• Reality?
• Useless
• Fragile
• Expensive
• Stereotypical tasks

22
Versatile vs Useless

• A self-reconfigurable robot can change into any
  shape needed for the task

23
Versatile vs useless

• In practice motion constraints make it difficult
  to change shape

M-TRAN, AIST
24
Versatile vs useless
Start
Goal
D. Brandt, University of Southern Denmark
25
Versatile vs useless

• Too weak to interact with the world
• The ATRON and the M-TRAN robots can only lift in
  the order of a few modules

26
Robust vs Fragile

• Robustness comes from redundancy
• If a module fails it can be ejected and other
  modules can take over
• Graceful degradation of performance

USCs ISI
27
Robust vs Fragile

• Difficult to detect if a module has failed
• Due to motion constraints it is difficult to
  eject the failed module
• Due to weakness of modules it may not be possible
  to eject the failed module at all

28
Cheap vs Expensive

• ATRON 2000
• MTRAN 3500
• .

29
Real Tasks vs Stereotypical Tasks

• Morphing Production Lines

D. Christensen, D. Brandt, University of Southern
Denmark
30
Task-Optimized vs Stereotypical Tasks

• Stereotypical tasks
• Oversimplified
• Fixed

31
Task-Optimized vs Stereotypical Tasks

• 1994

PolyPod, Parc/Stanford
32
Task-Optimized vs Stereotypical Tasks

• 2002

CONRO, ISI, USC
33
The State of Self-Reconfigurable Robots

• Vision
• Versatile
• Robust
• Cheap
• Task-Optimized

• Reality?
• Useless
• Fragile
• Expensive
• Stereotypical tasks

34
Anyway, insights?

• Self-reconfiguration is the key problem
• Connection mechanism takes 80 of space
• Introduce control complexity
• Reduce robustness
• Increase cost
• But increase versatility (in theory)
• Maybe we can get rid of the active connection
  mechanism and implement shape-change in a
  different way....

35
Deformable Robots
36
Deformable Modular Robots

• Shape-change can occur by deformation rather than
  modules moving around
• Shape-change occurs by controlling the
  deformation of individual modules

37
Prototypes
Hexatron, A. Lyder, University of Southern Denmark
Deformatron
38
Characteristics Fast Change of Shape
39
Comparison

• Deformable Modular Robots
• Simple
• Compliant
• Fast shape-change
• Shape-change within limits

• Self-Reconfigurable Robots
• Complex
• Rigid
• Slow shape-change
• In theory, unlimited Shape-change

40
What about our vision?

• Robust
• single module unimportant, but no self-repair
• Cheap
• Removal of active connectors
• Versatile
• Limited by ability to change shape or manually be
  reconfigured
• Task-optimized
• Future work...
• We think we can do better...

41
Hierarchical Robots
42
Hierarchical Robots

• A hierarchical robot is a modular robot where the
  modules themselves may be built from modules
• Modules are organized in a hierarchical structure
  where high-level modules are built from low-level
  modules
• Hypothesis while reducing the functionality of
  low-level modules we can increase the
  functionality of the robot as a whole

43
The Odin Hierarchical Robot

• Heterogeneous (modules only have to agree on
  connector design)
• Two classes of modules joints and links

A. Lyder, University of Southern Denmark
44
Odin Joint

• Power and communication busses
• Simple lock-and-key connector

A. Lyder, University of Southern Denmark
45
Odin Links

• All link modules provide
• Communication
• Power sharing
• Computation
• Links are heterogeneous and may provide
• Power
• Sensing
• Structure
• Actuation

46
The Odin Hierarchical Robot
x5
A. Lyder, R.F.M. Garcia, University of Denmark
47
Goals?

• Robust
• No single module is critical, but no self-repair
• Versatility
• Modules can be assembled in many ways
• Modules at all levels in hierarchy can be reused
• Cheap
• Heterogeneous, hierarchy, passive connectors
• Building blocks for a bodily optimized robot!

48
Future Work

• Mature design of Odin
• Demonstrate a range of non-stereotypical tasks
• Towards hierarchical deformable
  self-reconfigurable robots
• Higher-level modules consisting of chains of
  modules may tangle to connect

49
Conclusion

• We are developing building blocks for a new
  generation of cognitive robot bodies
• We strive to develop versatile, robust, and cheap
  modules
• Our approach is based on self-reconfigurable,
  deformable, and hierarchical robots

50
Get involved

• Postdoctoral in self-reconfigurable robots at
  University of Southern Denmark (electronics)
• Read our upcoming book An Introduction to
  Self-Reconfigurable Robots
• Use our upcoming USSR simulation tool (player
  compatible?)
• Join our workshop at IROS, San Diego
• Or attend R.F.M. Garcias talk at Robocomm

51
Thank You!Ph.D. Student David Christensen
Ph.D. Student David Brandt Ph.D. Student Andreas
Lyder MSc Student Ricardo Franco Mendoza Garcia
Research Ass. Danish Shaikh Ph.D. Student
Mirko Bordignon Associate Prof. Ulrik Pagh
Schultz Associate Prof. Kasper Stoy