Evolving Neural Networks for Robotic Arm Control

Anthony Horgan; Karl Mason

doi:10.1007/978-3-031-30229-9_38

Evolving Neural Networks for Robotic Arm Control

Anthony Horgan
, Karl Mason

Computer Science

University of Galway

Research output: Chapter in Book or Conference Publication/Proceeding › Conference Publication › peer-review

8 Citations (Scopus)

Abstract

Developing effective and adaptive robotic arm controllers is crucial for many industries, e.g. manufacturing. Traditional pre-programmed controllers cannot adapt to changing environments. This study investigates how neuroevolution can be used to develop robotic arm controllers and addresses key gaps in the existing literature, such as incorporating expert demonstrations and analyzing the robustness of evolved controllers. In addition to addressing these questions, this work compares different controller architectures and training algorithms. The proposed evolutionary neural network motion controller can accurately complete the random target reaching task, moving to within 1.7 cm from the target on average. An evolutionary supervisor neural network approach is also proposed to solve the pick-and-place task. The proposed method achieves a high successful completion rate, 927 out of 1000 trials.

Original language	English
Title of host publication	Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings
Editors	João Correia, Stephen Smith, Raneem Qaddoura
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	591-607
Number of pages	17
ISBN (Print)	9783031302282
DOIs	https://doi.org/10.1007/978-3-031-30229-9_38
Publication status	Published - 2023
Event	26th International Conference on Applications of Evolutionary Computation, EvoApplications 2023, held as part of EvoStar 2023 - Brno, Czech Republic Duration: 12 Apr 2023 → 14 Apr 2023

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13989 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	26th International Conference on Applications of Evolutionary Computation, EvoApplications 2023, held as part of EvoStar 2023
Country/Territory	Czech Republic
City	Brno
Period	12/04/23 → 14/04/23

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 9 Industry, Innovation, and Infrastructure

Keywords

Evolutionary Algorithms
Neural Networks
Neuroevolution
Robot Arm

Access to Document

10.1007/978-3-031-30229-9_38

Cite this

Horgan, A., & Mason, K. (2023). Evolving Neural Networks for Robotic Arm Control. In J. Correia, S. Smith, & R. Qaddoura (Eds.), Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings (pp. 591-607). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13989 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-30229-9_38

Horgan, Anthony ; Mason, Karl. / Evolving Neural Networks for Robotic Arm Control. Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings. editor / João Correia ; Stephen Smith ; Raneem Qaddoura. Springer Science and Business Media Deutschland GmbH, 2023. pp. 591-607 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{4b5663e3c12140a790c69f6e8e611762,

title = "Evolving Neural Networks for Robotic Arm Control",

abstract = "Developing effective and adaptive robotic arm controllers is crucial for many industries, e.g. manufacturing. Traditional pre-programmed controllers cannot adapt to changing environments. This study investigates how neuroevolution can be used to develop robotic arm controllers and addresses key gaps in the existing literature, such as incorporating expert demonstrations and analyzing the robustness of evolved controllers. In addition to addressing these questions, this work compares different controller architectures and training algorithms. The proposed evolutionary neural network motion controller can accurately complete the random target reaching task, moving to within 1.7 cm from the target on average. An evolutionary supervisor neural network approach is also proposed to solve the pick-and-place task. The proposed method achieves a high successful completion rate, 927 out of 1000 trials.",

keywords = "Evolutionary Algorithms, Neural Networks, Neuroevolution, Robot Arm",

author = "Anthony Horgan and Karl Mason",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 26th International Conference on Applications of Evolutionary Computation, EvoApplications 2023, held as part of EvoStar 2023 ; Conference date: 12-04-2023 Through 14-04-2023",

year = "2023",

doi = "10.1007/978-3-031-30229-9\_38",

language = "English",

isbn = "9783031302282",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "591--607",

editor = "Jo{\~a}o Correia and Stephen Smith and Raneem Qaddoura",

booktitle = "Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings",

}

Horgan, A & Mason, K 2023, Evolving Neural Networks for Robotic Arm Control. in J Correia, S Smith & R Qaddoura (eds), Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13989 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 591-607, 26th International Conference on Applications of Evolutionary Computation, EvoApplications 2023, held as part of EvoStar 2023, Brno, Czech Republic, 12/04/23. https://doi.org/10.1007/978-3-031-30229-9_38

Evolving Neural Networks for Robotic Arm Control. / Horgan, Anthony; Mason, Karl.
Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings. ed. / João Correia; Stephen Smith; Raneem Qaddoura. Springer Science and Business Media Deutschland GmbH, 2023. p. 591-607 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13989 LNCS).

Research output: Chapter in Book or Conference Publication/Proceeding › Conference Publication › peer-review

TY - GEN

T1 - Evolving Neural Networks for Robotic Arm Control

AU - Horgan, Anthony

AU - Mason, Karl

PY - 2023

Y1 - 2023

N2 - Developing effective and adaptive robotic arm controllers is crucial for many industries, e.g. manufacturing. Traditional pre-programmed controllers cannot adapt to changing environments. This study investigates how neuroevolution can be used to develop robotic arm controllers and addresses key gaps in the existing literature, such as incorporating expert demonstrations and analyzing the robustness of evolved controllers. In addition to addressing these questions, this work compares different controller architectures and training algorithms. The proposed evolutionary neural network motion controller can accurately complete the random target reaching task, moving to within 1.7 cm from the target on average. An evolutionary supervisor neural network approach is also proposed to solve the pick-and-place task. The proposed method achieves a high successful completion rate, 927 out of 1000 trials.

AB - Developing effective and adaptive robotic arm controllers is crucial for many industries, e.g. manufacturing. Traditional pre-programmed controllers cannot adapt to changing environments. This study investigates how neuroevolution can be used to develop robotic arm controllers and addresses key gaps in the existing literature, such as incorporating expert demonstrations and analyzing the robustness of evolved controllers. In addition to addressing these questions, this work compares different controller architectures and training algorithms. The proposed evolutionary neural network motion controller can accurately complete the random target reaching task, moving to within 1.7 cm from the target on average. An evolutionary supervisor neural network approach is also proposed to solve the pick-and-place task. The proposed method achieves a high successful completion rate, 927 out of 1000 trials.

KW - Evolutionary Algorithms

KW - Neural Networks

KW - Neuroevolution

KW - Robot Arm

UR - https://www.scopus.com/pages/publications/85159490188

U2 - 10.1007/978-3-031-30229-9_38

DO - 10.1007/978-3-031-30229-9_38

M3 - Conference Publication

AN - SCOPUS:85159490188

SN - 9783031302282

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 591

EP - 607

BT - Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings

A2 - Correia, João

A2 - Smith, Stephen

A2 - Qaddoura, Raneem

PB - Springer Science and Business Media Deutschland GmbH

T2 - 26th International Conference on Applications of Evolutionary Computation, EvoApplications 2023, held as part of EvoStar 2023

Y2 - 12 April 2023 through 14 April 2023

ER -

Horgan A, Mason K. Evolving Neural Networks for Robotic Arm Control. In Correia J, Smith S, Qaddoura R, editors, Applications of Evolutionary Computation - 26th European Conference, EvoApplications 2023, Held as Part of EvoStar 2023, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 591-607. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-30229-9_38