ITU

The Watchmaker's guide to Artificial Life: On the Role of Death, Modularity and Physicality in Evolutionary Robotics

Research output: Book / Anthology / Report / Ph.D. thesisPh.D. thesisResearch

Standard

The Watchmaker's guide to Artificial Life: On the Role of Death, Modularity and Physicality in Evolutionary Robotics. / Veenstra, Frank.

IT-Universitetet i København, 2019. 242 p. (ITU-DS; No. 144).

Research output: Book / Anthology / Report / Ph.D. thesisPh.D. thesisResearch

Harvard

APA

Vancouver

Author

Bibtex

@phdthesis{48d8944903cf449196bad31d5b50c6b8,
title = "The Watchmaker's guide to Artificial Life: On the Role of Death, Modularity and Physicality in Evolutionary Robotics",
abstract = "Optimizing robots through the implementation of evolutionary processesplays a key role in evolutionary robotics and artificial life. Challengesthat arise in the evolutionary optimization of robots are usually related toan algorithm{\textquoteright}s compromise between trying new solutions and improvingpreviously found solutions (the exploration vs. exploitation trade-off),and whether to express genomic information directly or generatively(genotype to phenotype mapping). An additional challenge is that thereis a discrepancy between simulation and reality (reality gap) where robots{\textquoteleft}evolved{\textquoteright} in simulation environments function differently when transferredto the real world.The exploration vs. exploitation trade-off is addressed throughdescribing and experimenting with the inclusion of biologically inspiredintrinsic mortality and how this affects the evolvability of populations.The results contribute to our understanding of the relationship betweenintrinsic mortality and mutation rate. The results further indicate how itcan be utilized to develop algorithms that can outperform state-of-the-artalgorithms.This thesis continues by addressing the challenge of mapping thegenotype to the phenotype through investigating the influence of generativeencodings on the evolution of simulated modular robots. It is investigatedhow an L-System as a generative encoding can lead to the formationof plant-inspired virtual creatures and shows that movement for tracking amoving light source is not an emerging phenotypic trait. Afterwards, thisgenerative encoding is shown to better evolve modular robots for locomotioncompared to a direct encoding. The addition of real-world solar panel modules demonstrates how modular robots can be evolved toward energyautonomy.The final part of the thesis describes the evolution of embodiment andcontrol of physical robots. As part of this, an automated process forassembly and disassembly of modular robots is demonstrated, which canbe used to evaluate evolved individuals in the real world. A viable methodfor implementing evolution directly is demonstrated through evolvingthe behavior of a knifefish-inspired physical soft robot. Both approachesrepresent strategies for addressing the reality gap.The experimental results of this thesis contribute to the understandingof biological phenomena and elucidate how improvements can be madeto existing methods in evolutionary robotics. It shows that we can utilizeconcepts from evolutionary biology to advance our understanding ofevolutionary dynamics, encodings and physical designs that are valuablefor the automated design of robots.",
author = "Frank Veenstra",
year = "2019",
language = "English",
isbn = "978-87-7949-011-6",
series = "ITU-DS",
publisher = "IT-Universitetet i K{\o}benhavn",
number = "144",
address = "Denmark",

}

RIS

TY - BOOK

T1 - The Watchmaker's guide to Artificial Life: On the Role of Death, Modularity and Physicality in Evolutionary Robotics

AU - Veenstra, Frank

PY - 2019

Y1 - 2019

N2 - Optimizing robots through the implementation of evolutionary processesplays a key role in evolutionary robotics and artificial life. Challengesthat arise in the evolutionary optimization of robots are usually related toan algorithm’s compromise between trying new solutions and improvingpreviously found solutions (the exploration vs. exploitation trade-off),and whether to express genomic information directly or generatively(genotype to phenotype mapping). An additional challenge is that thereis a discrepancy between simulation and reality (reality gap) where robots‘evolved’ in simulation environments function differently when transferredto the real world.The exploration vs. exploitation trade-off is addressed throughdescribing and experimenting with the inclusion of biologically inspiredintrinsic mortality and how this affects the evolvability of populations.The results contribute to our understanding of the relationship betweenintrinsic mortality and mutation rate. The results further indicate how itcan be utilized to develop algorithms that can outperform state-of-the-artalgorithms.This thesis continues by addressing the challenge of mapping thegenotype to the phenotype through investigating the influence of generativeencodings on the evolution of simulated modular robots. It is investigatedhow an L-System as a generative encoding can lead to the formationof plant-inspired virtual creatures and shows that movement for tracking amoving light source is not an emerging phenotypic trait. Afterwards, thisgenerative encoding is shown to better evolve modular robots for locomotioncompared to a direct encoding. The addition of real-world solar panel modules demonstrates how modular robots can be evolved toward energyautonomy.The final part of the thesis describes the evolution of embodiment andcontrol of physical robots. As part of this, an automated process forassembly and disassembly of modular robots is demonstrated, which canbe used to evaluate evolved individuals in the real world. A viable methodfor implementing evolution directly is demonstrated through evolvingthe behavior of a knifefish-inspired physical soft robot. Both approachesrepresent strategies for addressing the reality gap.The experimental results of this thesis contribute to the understandingof biological phenomena and elucidate how improvements can be madeto existing methods in evolutionary robotics. It shows that we can utilizeconcepts from evolutionary biology to advance our understanding ofevolutionary dynamics, encodings and physical designs that are valuablefor the automated design of robots.

AB - Optimizing robots through the implementation of evolutionary processesplays a key role in evolutionary robotics and artificial life. Challengesthat arise in the evolutionary optimization of robots are usually related toan algorithm’s compromise between trying new solutions and improvingpreviously found solutions (the exploration vs. exploitation trade-off),and whether to express genomic information directly or generatively(genotype to phenotype mapping). An additional challenge is that thereis a discrepancy between simulation and reality (reality gap) where robots‘evolved’ in simulation environments function differently when transferredto the real world.The exploration vs. exploitation trade-off is addressed throughdescribing and experimenting with the inclusion of biologically inspiredintrinsic mortality and how this affects the evolvability of populations.The results contribute to our understanding of the relationship betweenintrinsic mortality and mutation rate. The results further indicate how itcan be utilized to develop algorithms that can outperform state-of-the-artalgorithms.This thesis continues by addressing the challenge of mapping thegenotype to the phenotype through investigating the influence of generativeencodings on the evolution of simulated modular robots. It is investigatedhow an L-System as a generative encoding can lead to the formationof plant-inspired virtual creatures and shows that movement for tracking amoving light source is not an emerging phenotypic trait. Afterwards, thisgenerative encoding is shown to better evolve modular robots for locomotioncompared to a direct encoding. The addition of real-world solar panel modules demonstrates how modular robots can be evolved toward energyautonomy.The final part of the thesis describes the evolution of embodiment andcontrol of physical robots. As part of this, an automated process forassembly and disassembly of modular robots is demonstrated, which canbe used to evaluate evolved individuals in the real world. A viable methodfor implementing evolution directly is demonstrated through evolvingthe behavior of a knifefish-inspired physical soft robot. Both approachesrepresent strategies for addressing the reality gap.The experimental results of this thesis contribute to the understandingof biological phenomena and elucidate how improvements can be madeto existing methods in evolutionary robotics. It shows that we can utilizeconcepts from evolutionary biology to advance our understanding ofevolutionary dynamics, encodings and physical designs that are valuablefor the automated design of robots.

M3 - Ph.D. thesis

SN - 978-87-7949-011-6

T3 - ITU-DS

BT - The Watchmaker's guide to Artificial Life: On the Role of Death, Modularity and Physicality in Evolutionary Robotics

PB - IT-Universitetet i København

ER -

ID: 83711580