Introduction Progress Experiments Future work Developmental Co-evolution for Roombots Master Project Midterm Presentation Ebru Aydın Advisors: Prof. Auke Jan Ijspeert Rico M¨ ockel Jesse van den Kieboom AlexanderSpr¨owitz
Introduction Progress Experiments Future work
Developmental Co-evolution for Roombots
Master Project Midterm Presentation
Ebru AydınAdvisors:
Prof. Auke Jan IjspeertRico Mockel
Jesse van den KieboomAlexander Sprowitz
Introduction Progress Experiments Future work
Project Description:
Developmental Co-evolution for Roombots
Developmental Encoding:
• Grammatical Approach
• L-systemsex: A→A-B-A , B→B+A+B
Co-evolution:
• Evolution of configuration andcontroller
• Roombots• CPG, coupled phase oscillators
Introduction Progress Experiments Future work
Previous work: Demo LabProjects from Demo Lab : Hornby, Pollack, Lipson
• Generative Representations for Design AutomationStatic structures, 2D and 3D robots are evolved.
• The Golem Project
Introduction Progress Experiments Future work
Previous work: Demo Lab
Generative encoding builds creatures with better locomotionability then non-generative encodingGenerative encoding brings:
• Scalability through self-similar structures
• Hierarchical structure
• Compact representation
Introduction Progress Experiments Future work
Co-evolution for ”Adam”, D. Marbach
• Modular robot (Yamor style)
• Evolution of configuration and controller
• Harmonic oscillators for controller
• Tree-based encoding
• Evolved creatures are fitter thenhand-designed ones
Introduction Progress Experiments Future work
Co-evolution for Roombots:
• Configuration is evolved asin Marbach’s project, not the morphology
• Building blocks : Roombots
• Controller :Coupled phase oscillators, 3 drive function
• Each module : 3 DOF
• Adding a module (15 parameter):• Connection
type, face of each module, phase lag (4)• For each servo : amplitude, offset, drive
function (9)• Between servos in a module : Phase lag
(2)
Introduction Progress Experiments Future work
Timeline
Introduction Progress Experiments Future work
L-system
Terminal Rules (Build Commands):
• { } Block operator
• [ ] Save and store the state
• Change face
• Change oscillator type
• Change phase lag
• Change amplitude
• Change offset
• Add module
State: Module, face, phase lag
Introduction Progress Experiments Future work
Sample Grammar:
r1 →S CPL 3, ADD C3X PAR, COT S1 LOCKEDr2 → 2∗{ S CF 2, [ r1 ] }r3 → S CF 7, [ r1 ], r2r4 → COT S1 LOCKED, [ r3 ]
Rewritten string:COT S1 LOCKED, [ S CF 7, [ S CPL 3, ADD C3X PAR, COT S1 LOCKED ], 2∗{ S CF 2, [ S CPL 3, ADD C3X
PAR, COT S1 LOCKED ] }]
Introduction Progress Experiments Future work
Sample creature:
Introduction Progress Experiments Future work
Optimizer
• Evolutionary algorithm
• Evolving grammars
a- Mutations:
1. Change parameter of a build command2. Add rule to rewrite rule3. Remove rule from rewrite rule
b- Crossover: Use an index, copy the rules from parents:r1 r2 r3 r4
Introduction Progress Experiments Future work
Dispatcher
Dispatcher:
• Webots controller
Introduction Progress Experiments Future work
Fitness: Direct distance, measured in 20 seconds
0 50 100 1500
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Iteration
fitne
ss
MaxAvg
Introduction Progress Experiments Future work
Future work:
• Experiments, fitness(jumping, efficiency)
• Passive elements (spring, hinge joints, passive leg)
• Sensory feedback
• Analysis
• Report
Introduction Progress Experiments Future work
Thank you for your attention!