Evolution and Morphogenesis of Simulated Modular Robots: A Comparison Between a Direct and Generative Encoding
Research output: Conference Article in Proceeding or Book/Report chapter › Article in proceedings › Research › peer-review
Standard
Evolution and Morphogenesis of Simulated Modular Robots: A Comparison Between a Direct and Generative Encoding. / Veenstra, Frank; Faina, Andres; Risi, Sebastian; Støy, Kasper.
Applications of Evolutionary Computation: 20th European Conference, EvoApplications 2017, Amsterdam, The Netherlands, April 19-21, 2017, Proceedings, Part I. Springer, 2017. (Lecture Notes in Computer Science, Vol. 10199).Research output: Conference Article in Proceeding or Book/Report chapter › Article in proceedings › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - Evolution and Morphogenesis of Simulated Modular Robots: A Comparison Between a Direct and Generative Encoding
AU - Veenstra, Frank
AU - Faina, Andres
AU - Risi, Sebastian
AU - Støy, Kasper
PY - 2017/1/25
Y1 - 2017/1/25
N2 - Modular robots oer an important benet in evolutionaryrobotics, which is to quickly evaluate evolved morphologies and controlsystems in reality. However, articial evolution of simulated modularrobotics is a dicult and time consuming task requiring signicant computationalpower. While articial evolution in virtual creatures has madeuse of powerful generative encodings, here we investigate how a generativeencoding and direct encoding compare for the evolution of locomotionin modular robots when the number of robotic modules changes.Simulating less modules would decrease the size of the genome of a directencoding while the size of the genome of the implemented generativeencoding stays the same. We found that the generative encoding is signicantly more ecient in creating robot phenotypes in the initial stagesof evolution when simulating a maximum of 5, 10, and 20 modules. Thisnot only conrms that generative encodings lead to decent performancemore quickly, but also that when simulating just a few modules a generativeencoding is more powerful than a direct encoding for creatingrobotic structures. Over longer evolutionary time, the dierence betweenthe encodings no longer becomes statistically signicant. This leads us tospeculate that a combined approach { starting with a generative encodingand later implementing a direct encoding { can lead to more ecientevolved designs.
AB - Modular robots oer an important benet in evolutionaryrobotics, which is to quickly evaluate evolved morphologies and controlsystems in reality. However, articial evolution of simulated modularrobotics is a dicult and time consuming task requiring signicant computationalpower. While articial evolution in virtual creatures has madeuse of powerful generative encodings, here we investigate how a generativeencoding and direct encoding compare for the evolution of locomotionin modular robots when the number of robotic modules changes.Simulating less modules would decrease the size of the genome of a directencoding while the size of the genome of the implemented generativeencoding stays the same. We found that the generative encoding is signicantly more ecient in creating robot phenotypes in the initial stagesof evolution when simulating a maximum of 5, 10, and 20 modules. Thisnot only conrms that generative encodings lead to decent performancemore quickly, but also that when simulating just a few modules a generativeencoding is more powerful than a direct encoding for creatingrobotic structures. Over longer evolutionary time, the dierence betweenthe encodings no longer becomes statistically signicant. This leads us tospeculate that a combined approach { starting with a generative encodingand later implementing a direct encoding { can lead to more ecientevolved designs.
KW - Modular Robots
KW - Evolutionary Algorithms
KW - Direct & Generative Encodings
UR - https://www.youtube.com/watch?v=HCDftic1AdA
U2 - 10.1007/978-3-319-55849-3_56
DO - 10.1007/978-3-319-55849-3_56
M3 - Article in proceedings
SN - 978-3-319-55848-6
T3 - Lecture Notes in Computer Science
BT - Applications of Evolutionary Computation
PB - Springer
T2 - Evostar 2017
Y2 - 19 April 2017 through 21 April 2017
ER -
ID: 81921388