july 8, 2008 investigations at the interface of morphology...
TRANSCRIPT
![Page 1: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/1.jpg)
July 8, 2008
Josh BongardJosh BongardJosh BongardJosh BongardDepartment of Computer Science
College of Engineering and Mathematical SciencesUniversity of Vermont
Investigations at the Investigations at the Investigations at the Investigations at the interface of interface of interface of interface of
morphology, morphology, morphology, morphology, evolution and evolution and evolution and evolution and
cognitioncognitioncognitioncognition
![Page 2: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/2.jpg)
Talk Overview
The Tool SetMorphEngineMorphEngine `Unplugged’Physical Simulation
Example InvestigationsEvolving Robot Morphologies and Controllers TogetherEvolving Self-ModelsEvolving Coupled, Nonlinear ModelsEvolving Robots Capable of Multiple Behaviors
ConclusionsProximate and Ultimate Mechanisms of CognitionSummary
![Page 3: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/3.jpg)
MorphEngine Tool set
Optimizer
MorphEngine
Morphology
ANN Controller www.cs.uvm.edu/~jbongard/2008_Telluride/
MorphEngine.tar.gz
![Page 4: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/4.jpg)
Optimizer
MorphEngine
Morphology
ANN Controller
S1 S20 00.1 0.20.4 0.8…
Sensor values
www.cs.uvm.edu/~jbongard/2008_Telluride/
MorphEngine.tar.gz
MorphEngine Tool set
![Page 5: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/5.jpg)
Optimizer
t1: s1=0,s2=0
t2: s1=0.1, s2=0.3
…
MorphEngine
Morphology
Controller
Sensor values
rand(0,1)
www.cs.uvm.edu/~jbongard/2008_Telluride/
MorphEngine_Unplugged.tar.gz
MorphEngine Unplugged Tool set
![Page 6: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/6.jpg)
Optimizer
MorphEngine
Morphology
Controller
Sensor values
www.cs.uvm.edu/~jbongard/2008_Telluride/
MorphEngine_Unplugged.tar.gz
MorphEngine Unplugged Tool set
![Page 7: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/7.jpg)
ThreeThreeThreeThree----dimensionaldimensionaldimensionaldimensional
For each time stepFor each time stepFor each time stepFor each time step
Internal, external forcesforcesforcesforces are calculated for each object in the simulationPositions, orientations and velocities of each objectobjectobjectobject are updated
CollisionsCollisionsCollisionsCollisions between objects are detected, and resolved
Simulated sensorsSimulated sensorsSimulated sensorsSimulated sensorsReal-time sensor data Input sensor values neural network
Simulated motorsSimulated motorsSimulated motorsSimulated motorsOutput neuron values desired joint angles Torque
From ODE documentation
Physical Simulation Tool set
![Page 8: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/8.jpg)
Talk Overview
The Tool SetMorphEngineMorphEngine `Unplugged’Physical Simulation
Example InvestigationsEvolving Robot Morphologies and Controllers TogetherEvolving Self-ModelsEvolving Coupled, Nonlinear ModelsEvolving Robots Capable of Multiple Behaviors
ConclusionsProximate and Ultimate Mechanisms of CognitionSummary
![Page 9: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/9.jpg)
Optimizer
MorphEngine
Morphology
ANN Controller
S1 S20 00.1 0.20.4 0.8…
Sensor values
Evolving Robot Bodies and Brains Together Example Investigations
![Page 10: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/10.jpg)
Optimizer
MorphEngine
Growth Commands
S1 S20 00.1 0.20.4 0.8…
Sensor values
Evolving Robot Bodies and Brains Together Example Investigations
![Page 11: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/11.jpg)
The growing robotbody and brain
The genome, a copy of which
resides in each morphological unit
Transcription factors,
produced by genes contained in the genome,
diffuse through the robot's body, causing phenotypic change
Automating Robot Design
Bongard, J. C. (2002)Evolving Modular Genetic Regulatory Networks, in Proceedings of the IEEE 2002 Congress on Evolutionary Computation (CEC2002),pp. 1872-1877.
![Page 12: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/12.jpg)
Automating Robot Design
Bongard, J. C. and R. Pfeifer (2001)Repeated Structure and Dissociation of Genotypic and Phenotypic Complexity in Artificial Ontogeny, in Proceedings of The Genetic and Evolutionary Computation Conference,pp. 829-836.
![Page 13: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/13.jpg)
Automating Robot Design
Bongard, J. C. and R. Pfeifer (2001)Repeated Structure and Dissociation of Genotypic and Phenotypic Complexity in Artificial Ontogeny, in Proceedings of The Genetic and Evolutionary Computation Conference,pp. 829-836.
![Page 14: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/14.jpg)
2
2
1
1
Evolving Robot Bodies and Brains Together Example Investigations
![Page 15: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/15.jpg)
Regulatory geneTargetted regulatory geneStructural gene affecting morphogenesisStructural gene affecting neurogenesis
Evolving Robot Bodies and Brains Together Example Investigations
![Page 16: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/16.jpg)
Optimizer
MorphEngine
S1 S20 00.1 0.20.4 0.8…
Evolving Self-Models Example Investigations
Bongard, J., Zykov, V., Lipson, H. (2006). Resilient machines throughcontinuous self-modeling. Science, 314: 1118-1121.
![Page 17: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/17.jpg)
Optimizer
MorphEngine
S1 S20 00.8 0.30.6 0.4…
Evolving Self-Models Example Investigations
Bongard, J., Zykov, V., Lipson, H. (2006). Resilient machines throughcontinuous self-modeling. Science, 314: 1118-1121.
![Page 18: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/18.jpg)
Optimizer
MorphEngine
S1 S20 00.1 0.20.4 0.8…
Evolving Self-Models Example Investigations
Bongard, J., Zykov, V., Lipson, H. (2006). Resilient machines throughcontinuous self-modeling. Science, 314: 1118-1121.
![Page 19: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/19.jpg)
Self-model synthesis
Robot generates
several self-models
to match sensor
data collected while
performing previous
actions. It does not
know which model is
correct.
Robot physically
performs an action.
Initially, this action is
random; later, it is the best
action generated in step C.
Robot generates several
possible actions that
disambiguate competing self-
models.
Exploratory action synthesis
After several cycles
of A-C, the currently best
model is used to
generate a locomotion
sequence through
optimization.
Target behavior synthesis
The best locomotion
sequence is then executed by
the physical robot.
The cycle continues
at step B to further
refine models, or at
step D to create new
behaviors.
?
A
BC
D
E
F
Self-model synthesis
Robot generates
several self-models
to match sensor
data collected while
performing previous
actions. It does not
know which model is
correct.
Robot physically
performs an action.
Initially, this action is
random; later, it is the best
action generated in step C.
Robot generates several
possible actions that
disambiguate competing self-
models.
Exploratory action synthesis
After several cycles
of A-C, the currently best
model is used to
generate a locomotion
sequence through
optimization.
Target behavior synthesis
The best locomotion
sequence is then executed by
the physical robot.
The cycle continues
at step B to further
refine models, or at
step D to create new
behaviors.
?
A
BC
D
E
F
Populationof Actions
Populationof
Models
Populationof Behaviors
Evolving Self-Models Example Investigations
![Page 20: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/20.jpg)
Automating Robot RecoveryBongard, J., Zykov, V., Lipson, H. (2006). Resilient machines throughcontinuous self-modeling. Science, 314: 1118-1121.
![Page 21: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/21.jpg)
Intelligent
TestingModeling
Random
TestingModelingRandom
TestingModeling
Batch testing Random testing Intelligent testing (EEA)
Comparative modeling performance
Intact robot
30 trials 30 trials 30 trials
![Page 22: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/22.jpg)
Intelligent
TestingModeling
Random
TestingModelingRandom
TestingModeling
23.3% success 26.7% success 43.3% success
Batch testing Random testing Intelligent testing (EEA)
Comparative modeling performance
9.62 +/- 1.47 cm 9.7 +/- 1.45 cm 7.31 +/- 1.22 cm
Intact robot
Topology correct?
Model error:
![Page 23: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/23.jpg)
Automating Robot RecoveryBongard, J., Zykov, V., Lipson, H. (2006). Resilient machines throughcontinuous self-modeling. Science, 314: 1118-1121.
![Page 24: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/24.jpg)
Intelligent
TestingModeling
Random
TestingModelingRandom
TestingModeling
23.3% success 26.7% success 43.3% success
Batch testing Random testing Intelligent testing (EEA)
Comparative modeling performance
9.62 +/- 1.47 cm 9.7 +/- 1.45 cm 7.31 +/- 1.22 cm
Intact robot
Topology correct?
Model error:
Intelligent
TestingModeling
Random
TestingModelingRandom
TestingModeling
Batch testing Random testing Intelligent testing (EEA)
5.60 +/- 2.98 cm 4.55 +/- 3.22 cm 2.17 +/- 0.55 cm
Damagedrobot
Model error:
![Page 25: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/25.jpg)
Multiple Robots Sharing Self-Models Example Investigations
Bongard, J. (2007) Exploiting Multiple Robots to Accelerate Self-Modeling, Proceedings of the 9th Annual Conference on Genetic and Evolutionary Computation, ACM Press, New York, NY, pp. 214-221.
![Page 26: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/26.jpg)
Bongard J. and Lipson H.(2007). Automated reverse engineering ofnonlinear dynamical systems.Proceedings of the National Academy of Sciences,104(24): 9943-9948.
Automating System Identification
![Page 27: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/27.jpg)
*
+ sin
0.8 x x
/
x y
(dx/dt=) (dy/dt=)dx/dt = (0.8 + x) * sin(x)
dy/dt = x / y
*
+ 1.4
0.8 x
/
0.2 +
(dx/dt=) (dy/dt=)
x y
dx/dt = 1.4(0.8+x)
dy/dt = 0.2 / (x+y)
x(0)
y(0)
t
t
Encoding and optimizing of models
![Page 28: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/28.jpg)
dG/dt = 0.96A2/(0.96A2+1)
dA/dt = G( L/(L+1) – A/(A+1) )
dL/dt = -GL/(L+1)
Best modelBest modelBest modelBest model
dG/dt = A2/(A2+1) – 0.01G + 0.001
dA/dt = G( L/(L+1) – A/(A+1) )
dL/dt = -GL/(L+1)
TargetTargetTargetTargetsystemsystemsystemsystem
The The The The laclaclaclac operonoperonoperonoperon from from from from E. coliE. coliE. coliE. coli(G = concentration of beta-galactosidase; A = allolactose; L = lactose)
Evolving Coupled, Nonlinear Models Application I: lac operon in E. coli
![Page 29: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/29.jpg)
dH/dt = 3.42x106 - 67.82H - 10.97L
dL/dt = 3.10x105 + 32.66H - 63.16L
BestBestBestBest
modelmodelmodelmodel
TargetTargetTargetTarget
SystemSystemSystemSystem
Historical data reporting approximated populations of snowshoe hare (H) and
Canadian lynx (L)
Evolving Coupled, Nonlinear Models Application II: Ecological data set
![Page 30: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/30.jpg)
-1.57rad -2.67rad00
dθ/dt = 1.004ω + 0.0001
dω/dt = -19.43sin(1.104θ+0+0+0+0)
dθ/dt = 1.008ω + 0.0028
dω/dt = -19.43sin(1.0009θ----1.5751.5751.5751.575)
dθ/dt = 1.0039ω - 0.0003
dω/dt = -22.61sin(1.101θ----2.6732.6732.6732.673)
dθ/dt = ω
dω/dt = -9.8Lsin(θ)
Model for an idealized single pendulum with no friction
Evolving Coupled, Nonlinear Models Application III: Mechanical pendula
![Page 31: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/31.jpg)
Evolving Robots Capable of Dynamic Behavior Boston Dynamic’s Big Dog
![Page 32: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/32.jpg)
Evolving Robots Capable of Multiple Behaviors Example Investigations
Behavior chaining
Enables a robot to learn multiple, dynamic behaviors gradually.
Learns one behavior,then gradually incorporates new behaviors into its existing repertoire.
All behaviors are incorporated into the same monolithic controller.
Is more scalable than other approaches that require buildinga new controller component for each new behavior.
Builds on the idea of scaffolding, and robot shaping:gradually changing the environment to guide the learner toward a complex behaviorit might not have learned otherwise.
![Page 33: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/33.jpg)
Behavior Chaining:
p = random controllerEval(p)
For g = 1:generations
tmp = CopyOf(p);c = Mutate(tmp);Eval(c)
if (Fitness(c) > Fitness(p))p = c
else delete c
if Success(p)MoveObjFurther()
if Failure(p) LengthenEvalTime()
Morphology
CTRNN Controller
S1 S20 00.1 0.20.4 0.8…
Sensor values
Evolving Robots Capable of Multiple Behaviors Example Investigations
Bongard, J. (2008)Behavior Chaining: Incremental Behavioral
Integration for Evolutionary Robotics,ALife XI, to appear.
MorphEngine
![Page 34: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/34.jpg)
d=0.0
d=0.5
d=1.0
d=1.5
d=2.0
d=2.5
d=3.0
30 trials:
30 trials:
30 trials:
30 trials:
30 trials:
30 trials:
30 trials:
Bongard, J. (2008)Behavior Chaining: Incremental Behavioral
Integration for Evolutionary Robotics,ALife XI, to appear.
0 5 10 15Hourssinceexperimentstart
0
1
2
3
4
F i n a l t a r g e t d i s t a n c e ( m )
Quadruped
d=0.0d=0.5d=1.0d=1.5d=2.0d=2.5d=3.0
Evolving Robots Capable of Multiple Behaviors Example Investigations
![Page 35: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/35.jpg)
Evolving Robots Capable of Multiple Behaviors Example Investigations
Courtesy ofJosh Auerbach
Successful lifting
Fumbling
Little locomotion
No locomotion
![Page 36: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/36.jpg)
Talk Overview
The Tool SetMorphEngineMorphEngine `Unplugged’Physical Simulation
Example InvestigationsEvolving Robot Morphologies and Controllers TogetherEvolving Self-ModelsEvolving Coupled, Nonlinear ModelsEvolving Robots Capable of Multiple Behaviors
ConclusionsProximate and Ultimate Mechanisms of CognitionSummary
![Page 37: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/37.jpg)
Proximate and Ultimate Mechanisms of Behavior Conclusions
Proximate mechanisms:
1. Causation (Mechanism): what are the stimuli that elicit the response, and how has it been modified by recent learning?
2. Development (Ontogeny): how does the behaviour change with age?
Ultimate mechanisms:
3. Evolution (Phylogeny): how does the behaviour compare with similar behaviour in related species, and how might it have arisen through the process of phylogeny?
4. Function (Adaptation): how does the behavior impact on the animal's chances of survival and reproduction?
Nikolaas Tinbergen (1907–1988)Four levels of description:
not mutually exclusive
![Page 38: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/38.jpg)
Mechanisms of Centralized Neural Structure Conclusions
Proximate mechanisms:
1. Causation (Mechanism): how is stimuli combined in the central nervous system?
2. Development (Ontogeny): how does the central nervous system develop during growth?
Ultimate mechanisms:
3. Evolution (Phylogeny): how did central neural structure in the current population evolve from distributed neural structures in ancestral populations?
4. Function (Adaptation): how does centralized neural structure impact the robot’s chances of survival and reproduction? (integration of sensor information?)
![Page 39: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/39.jpg)
Mechanisms of Competitive Processes in the Brain Conclusions
Proximate mechanisms:
1. Causation (Mechanism): what are the stimuli that trigger competition between processes in the brain? What form do these processes take?
2. Development (Ontogeny): How do competitive processes multiply during growth? Do some processes eventually ‘win’?
Ultimate mechanisms:
3. Evolution (Phylogeny): how did competitive neural processes arise from less, or lack of competitive processes in ancestral populations?
4. Function (Adaptation): how do competitive neural processes affect the animal's chances of survival and reproduction?
![Page 40: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/40.jpg)
Mechanisms of Gradual Behavior Integration Conclusions
Proximate mechanisms:
1. Causation (Mechanism): what are the stimuli that trigger the learning of a new behavior? What form do these processes take?
2. Development (Ontogeny): How are behaviors gradually integrated into a robot’s exhibit behavioral repertoire?
Ultimate mechanisms:
3. Evolution (Phylogeny): how do the mechanisms that allow for gradual behavior integration arise from ancestral robots capable of exhibiting only one behavior?
4. Function (Adaptation): Obvious
![Page 41: July 8, 2008 Investigations at the interface of morphology ...papers/neuromorphic2008/lectures08/Josh_Bongard.pdfevolution and cognition. TalkOverview ... Target behavior synthesis](https://reader033.vdocuments.site/reader033/viewer/2022060510/5f2674be789d071f45556437/html5/thumbnails/41.jpg)
Summary
The Tool SetMorphEngineMorphEngine `Unplugged’Physical Simulation
Example InvestigationsEvolving Robot Morphologies and Controllers TogetherEvolving Self-ModelsEvolving Coupled, Nonlinear ModelsEvolving Robots Capable of Multiple Behaviors
ConclusionsProximate and Ultimate Mechanisms of CognitionSummary
Take Home MessageTo truly understand cognition, we must pursue two lines of attack:
understand biological systems by replicating them in hard/softwarecause analogues of these systems to evolve in artificial systems.
Software Robots with Hardware Brains
www.cs.uvm.edu/~jbongard/2008_Telluride
CNS?
Competitive Processes in the Brain?
Gradual Behavior Integration?