A unified architecture for agent behaviors with selection of evolved neural network modules

Kyung Joong Kim, Sung Bae Cho

Research output: Contribution to journalArticle

6 Citations (Scopus)

Abstract

To model complex systems for agent behaviors, genetic algorithms have been used to evolve neural networks which are based on cellular automata. These neural networks are popular tools in the artificial life community. This hybrid architecture aims at achieving synergy between the cellular automata and the powerful generalization capabilities of the neural networks. Evolutionary algorithms provide useful ways to learn about the structure of these neural networks, but the use of direct evolution in more difficult and complicated problems often fails to achieve satisfactory solutions. A more promising solution is to employ incremental evolution that reuses the solutions of easy tasks and applies these solutions to more difficult ones. Moreover, because the human brain can be divided into many behaviors with specific functionalities and because human beings can integrate these behaviors for high-level tasks, a biologically-inspired behavior selection mechanism is useful when combining these incrementally evolving basic behaviors. In this paper, an architecture based on cellular automata, neural networks, evolutionary algorithms, incremental evolution and a behavior selection mechanism is proposed to generate high-level behaviors for mobile robots. Experimental results with several simulations show the possibilities of the proposed architecture.

Original languageEnglish
Pages (from-to)253-268
Number of pages16
JournalApplied Intelligence
Volume25
Issue number3
DOIs
Publication statusPublished - 2006 Dec 1

Fingerprint

Neural networks
Cellular automata
Evolutionary algorithms
Mobile robots
Large scale systems
Brain
Genetic algorithms

All Science Journal Classification (ASJC) codes

  • Artificial Intelligence

Cite this

@article{5ee610950e7d400eb8f0b8e20ee304e0,
title = "A unified architecture for agent behaviors with selection of evolved neural network modules",
abstract = "To model complex systems for agent behaviors, genetic algorithms have been used to evolve neural networks which are based on cellular automata. These neural networks are popular tools in the artificial life community. This hybrid architecture aims at achieving synergy between the cellular automata and the powerful generalization capabilities of the neural networks. Evolutionary algorithms provide useful ways to learn about the structure of these neural networks, but the use of direct evolution in more difficult and complicated problems often fails to achieve satisfactory solutions. A more promising solution is to employ incremental evolution that reuses the solutions of easy tasks and applies these solutions to more difficult ones. Moreover, because the human brain can be divided into many behaviors with specific functionalities and because human beings can integrate these behaviors for high-level tasks, a biologically-inspired behavior selection mechanism is useful when combining these incrementally evolving basic behaviors. In this paper, an architecture based on cellular automata, neural networks, evolutionary algorithms, incremental evolution and a behavior selection mechanism is proposed to generate high-level behaviors for mobile robots. Experimental results with several simulations show the possibilities of the proposed architecture.",
author = "Kim, {Kyung Joong} and Cho, {Sung Bae}",
year = "2006",
month = "12",
day = "1",
doi = "10.1007/s10489-006-0106-z",
language = "English",
volume = "25",
pages = "253--268",
journal = "Applied Intelligence",
issn = "0924-669X",
publisher = "Springer Netherlands",
number = "3",

}

A unified architecture for agent behaviors with selection of evolved neural network modules. / Kim, Kyung Joong; Cho, Sung Bae.

In: Applied Intelligence, Vol. 25, No. 3, 01.12.2006, p. 253-268.

Research output: Contribution to journalArticle

TY - JOUR

T1 - A unified architecture for agent behaviors with selection of evolved neural network modules

AU - Kim, Kyung Joong

AU - Cho, Sung Bae

PY - 2006/12/1

Y1 - 2006/12/1

N2 - To model complex systems for agent behaviors, genetic algorithms have been used to evolve neural networks which are based on cellular automata. These neural networks are popular tools in the artificial life community. This hybrid architecture aims at achieving synergy between the cellular automata and the powerful generalization capabilities of the neural networks. Evolutionary algorithms provide useful ways to learn about the structure of these neural networks, but the use of direct evolution in more difficult and complicated problems often fails to achieve satisfactory solutions. A more promising solution is to employ incremental evolution that reuses the solutions of easy tasks and applies these solutions to more difficult ones. Moreover, because the human brain can be divided into many behaviors with specific functionalities and because human beings can integrate these behaviors for high-level tasks, a biologically-inspired behavior selection mechanism is useful when combining these incrementally evolving basic behaviors. In this paper, an architecture based on cellular automata, neural networks, evolutionary algorithms, incremental evolution and a behavior selection mechanism is proposed to generate high-level behaviors for mobile robots. Experimental results with several simulations show the possibilities of the proposed architecture.

AB - To model complex systems for agent behaviors, genetic algorithms have been used to evolve neural networks which are based on cellular automata. These neural networks are popular tools in the artificial life community. This hybrid architecture aims at achieving synergy between the cellular automata and the powerful generalization capabilities of the neural networks. Evolutionary algorithms provide useful ways to learn about the structure of these neural networks, but the use of direct evolution in more difficult and complicated problems often fails to achieve satisfactory solutions. A more promising solution is to employ incremental evolution that reuses the solutions of easy tasks and applies these solutions to more difficult ones. Moreover, because the human brain can be divided into many behaviors with specific functionalities and because human beings can integrate these behaviors for high-level tasks, a biologically-inspired behavior selection mechanism is useful when combining these incrementally evolving basic behaviors. In this paper, an architecture based on cellular automata, neural networks, evolutionary algorithms, incremental evolution and a behavior selection mechanism is proposed to generate high-level behaviors for mobile robots. Experimental results with several simulations show the possibilities of the proposed architecture.

UR - http://www.scopus.com/inward/record.url?scp=33750738071&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33750738071&partnerID=8YFLogxK

U2 - 10.1007/s10489-006-0106-z

DO - 10.1007/s10489-006-0106-z

M3 - Article

AN - SCOPUS:33750738071

VL - 25

SP - 253

EP - 268

JO - Applied Intelligence

JF - Applied Intelligence

SN - 0924-669X

IS - 3

ER -