Title: Studying cerebral organization through embodied artificial evolution
1Studying cerebral organization throughembodied
artificial evolution
- Joris Janssen, Twan Goosen,
- Ida Sprinkhuizen-Kuyper, and Pim Haselager
1st NSVKI Student Conference Nijmegen, 22 June
2007
2Overview
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionOverview
- Introduction
- Research questions
- Experimental Setup
- Robot Design
- Task Ecological Niche
- Control structure Evolution
- Results
- Fitness
- Clustering
- Symmetry
- Fitness II
- Discussion
- Hemispheres
- Symmetry vs. Asymmetry
3Introduction
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionIntroduction
- Cerebral lateralization left and right part of
human brain not the same - Cerebral lateralization has had a lot of
attention - Cognitive psychology
- Neuropsychology
- Evolutionary biology
- Focus on
- What is lateralized?
- Not on origin (why?) of lateralization
4Research questions
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionResearch questions
- Two questions
- Why do we possess two hemispheres?
- Why are some of the functions of the brain
organized symmetrically and others
asymmetrically? - gtgt actually two questions
- - why some asymmetrically?
- - why some symmetrically?
5Hypotheses
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionResearch questions
- Two hemispheres
- Back-up in event of injury
- gtgt unlikely brain not that plastic
- Related to symmetrical body plan
6Hypotheses
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionResearch questions
- Asymmetrical functions less neural space
symmetrical
asymmetrical
7Hypotheses
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionResearch questions
- Why then symmetrical functions?
- Found in all kinds of species
- Evolutionary advantage less information needs to
be integrated - Likely to develop easier/faster
8Summary
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionResearch questions
- Two hemispheres
- Related to origin of symmetrical body plan
- Asymmetrically organized functions
- Require less neural space
- Symmetrically organized functions
- Easier/faster to develop
9Overview
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionOverview
- Introduction
- Research questions
- Experimental Setup
- Robot Design
- Task Ecological Niche
- Control structure Evolution
- Results
- Fitness
- Clustering
- Symmetry
- Fitness II
- Discussion
- Hemispheres
- Symmetry vs. Asymmetry
10Experimental setup
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionExperimental setup
- Needs
- Robots (symmetrical and asymmetrical)
- Task and environment
- Control structure
- Developmental mechanism evolution
11Robot design
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionRobot design
symmetrical
asymmetrical
12Task Ecological niche
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionTask Ecological niche
- Move around environment
- Cover as much distance as possible
- Avoid bumping into walls
13Control structure
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionControl structure Evolution
14Evolution
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolution Control structure
Evolution
- Evolution used to modify weights and biases
- First generation starts with random weights and
biases - Genetic algorithm
- Test individuals
- Select top 10 (out of 20)
- Generate 10 new ones out of 10 old ones (by
mutating weights)
15Overview
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionOverview
- Introduction
- Research questions
- Experimental Setup
- Robot Design
- Task Ecological Niche
- Control structure Evolution
- Results
- Fitness
- Clustering
- Symmetry
- Fitness II
- Discussion
- Hemispheres
- Symmetry vs. Asymmetry
16Fitness of symmetrical group
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionFitness
Group with symmetrical body plan
17Fitness of asymmetrical group
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionFitness
Group with asymmetrical body plan
18Clustering
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionClustering
Edelman Tononi
19Clustering results
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionClustering
Analysis of variance over last 50 generations p
lt 0.001 R2 0.135
20Symmetry
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionSymmetry
Symmetrical processing
Asymmetrical processing
21Asymmetry
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionSymmetry
Analysis of variance over last 50 generations p
lt 0.001 R2 0.525
22Fitness II
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionFitness
23Summary
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionResults
- Symmetrical body plan results in more symmetrical
control structure - More clustering in symmetrical group
- Asymmetrical group performs better in the end
- Symmetrical group develops faster in the beginning
24Overview
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionOverview
- Introduction
- Research questions
- Experimental Setup
- Robot Design
- Task Ecological Niche
- Control structure Evolution
- Results
- Fitness
- Clustering
- Symmetry
- Fitness II
- Discussion
- Hemispheres
- Symmetry vs. Asymmetry
25Hemispheres
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionDiscussion
- Hypothesis symmetrical body plan results in two
hemispheres - Results Symmetrical body plan results in two
clusters
26Asymmetry
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionDiscussion
- Hypothesis asymmetrical better performance since
better use of neural space - Results asymmetrical group performs better
27Symmetry
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionDiscussion
- Hypothesis symmetry less complex to develop
- Results symmetry develops faster
28Causality?
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionDiscussion
body plan structure
performance
control structure
control structure
body plan structure
performance
29Validity?
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionDiscussion
- Simulated model no valid equivalent of natural
system - But this research can
- .provide supporting evidence
- .lead to new ideas and insights
30Take home message
Janssen, Goosen, Sprinkhuizen-Kuyper, and
Haselager
Studying cerebral organization through embodied
artificial evolutionTake home message
Use robots as cognitive tools!