Improving RL Power for On-Line Evolution of Gaits in Modular Robots

Milan Jelisavcic; Matteo De Carlo; Evert Haasdijk; A.E. Eiben

doi:10.1109/SSCI.2016.7850166

Improving RL Power for On-Line Evolution of Gaits in Modular Robots

Milan Jelisavcic, Matteo De Carlo, Evert Haasdijk, A.E. Eiben

Research output: Chapter in Book / Report / Conference proceeding › Conference contribution › Academic › peer-review

337 Downloads (Pure)

Abstract

This paper addresses the problem of on-line gait learning in modular robots whose shape is not known in advance. The best algorithm for this problem known to us is a reinforcement learning method, called RL PoWER. In this study we revisit the original RL PoWER algorithm and observe that in essence it is a specific evolutionary algorithm. Based on this insight we propose two modifications of the main search operators and compare the quality of the evolved gaits when either or both of these modified operators are employed. The results show that using 2-parent crossover as well as mutation with self- adaptive step-sizes can significantly improve the performance of the original algorithm.

Original language	English
Title of host publication	2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016
Publisher	Institute of Electrical and Electronics Engineers, Inc.
Pages	1-8
Number of pages	8
ISBN (Electronic)	978-1-5090-4240-1
DOIs	https://doi.org/10.1109/SSCI.2016.7850166
Publication status	Published - 6 Dec 2016
Event	2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016 - Athens, Greece Duration: 6 Dec 2016 → 9 Dec 2016

Conference

Conference	2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016
Country/Territory	Greece
City	Athens
Period	6/12/16 → 9/12/16

UN SDGs

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

Access to Document

10.1109/SSCI.2016.7850166

SSCI16_paper_199Final published version, 3.02 MB

Persistent URL (handle)

Persistent link

Cite this

@inproceedings{9513ac8cfc3840a6aa737e639b7e1997,

title = "Improving RL Power for On-Line Evolution of Gaits in Modular Robots",

abstract = "This paper addresses the problem of on-line gait learning in modular robots whose shape is not known in advance. The best algorithm for this problem known to us is a reinforcement learning method, called RL PoWER. In this study we revisit the original RL PoWER algorithm and observe that in essence it is a specific evolutionary algorithm. Based on this insight we propose two modifications of the main search operators and compare the quality of the evolved gaits when either or both of these modified operators are employed. The results show that using 2-parent crossover as well as mutation with self- adaptive step-sizes can significantly improve the performance of the original algorithm.",

author = "Milan Jelisavcic and {De Carlo}, Matteo and Evert Haasdijk and A.E. Eiben",

year = "2016",

month = dec,

day = "6",

doi = "10.1109/SSCI.2016.7850166",

language = "English",

pages = "1--8",

booktitle = "2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016",

publisher = "Institute of Electrical and Electronics Engineers, Inc.",

note = "2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016 ; Conference date: 06-12-2016 Through 09-12-2016",

}

Jelisavcic, M, De Carlo, M, Haasdijk, E & Eiben, AE 2016, Improving RL Power for On-Line Evolution of Gaits in Modular Robots. in 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016., 7850166, Institute of Electrical and Electronics Engineers, Inc., pp. 1-8, 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016, Athens, Greece, 6/12/16. https://doi.org/10.1109/SSCI.2016.7850166

Improving RL Power for On-Line Evolution of Gaits in Modular Robots. / Jelisavcic, Milan; De Carlo, Matteo; Haasdijk, Evert et al.
2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016. Institute of Electrical and Electronics Engineers, Inc., 2016. p. 1-8 7850166.

Research output: Chapter in Book / Report / Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Improving RL Power for On-Line Evolution of Gaits in Modular Robots

AU - Jelisavcic, Milan

AU - De Carlo, Matteo

AU - Haasdijk, Evert

AU - Eiben, A.E.

PY - 2016/12/6

Y1 - 2016/12/6

N2 - This paper addresses the problem of on-line gait learning in modular robots whose shape is not known in advance. The best algorithm for this problem known to us is a reinforcement learning method, called RL PoWER. In this study we revisit the original RL PoWER algorithm and observe that in essence it is a specific evolutionary algorithm. Based on this insight we propose two modifications of the main search operators and compare the quality of the evolved gaits when either or both of these modified operators are employed. The results show that using 2-parent crossover as well as mutation with self- adaptive step-sizes can significantly improve the performance of the original algorithm.

AB - This paper addresses the problem of on-line gait learning in modular robots whose shape is not known in advance. The best algorithm for this problem known to us is a reinforcement learning method, called RL PoWER. In this study we revisit the original RL PoWER algorithm and observe that in essence it is a specific evolutionary algorithm. Based on this insight we propose two modifications of the main search operators and compare the quality of the evolved gaits when either or both of these modified operators are employed. The results show that using 2-parent crossover as well as mutation with self- adaptive step-sizes can significantly improve the performance of the original algorithm.

UR - http://www.scopus.com/inward/record.url?scp=85016006204&partnerID=8YFLogxK

U2 - 10.1109/SSCI.2016.7850166

DO - 10.1109/SSCI.2016.7850166

M3 - Conference contribution

SP - 1

EP - 8

BT - 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016

PB - Institute of Electrical and Electronics Engineers, Inc.

T2 - 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016

Y2 - 6 December 2016 through 9 December 2016

ER -

Improving RL Power for On-Line Evolution of Gaits in Modular Robots

Abstract

Conference

UN SDGs

Access to Document

Persistent URL (handle)

Other files and links

Fingerprint

Cite this