Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot

Shengbo Xu, Hirotaka Moriguch, Shinichi Honiden

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

In reinforcement learning tasks with continuous state-action, parameterized policy search has been known to be a powerful method. Applying NeuroEvolution (NE) to optimizing the policy represented by artificial neural network (ANN) is a particularly active research field. In most cases, NE algorithms cost a large amount of trial-and-error (episode) to optimize policies. However, due to time and cost constraints, researchers and practitioners cannot repeat a number of episodes on physical robots. Thus, choosing an efficient NE algorithm is a key to optimize policies with limited time and cost. In this work, our goal is to help users to choose an efficient NE algorithm. We compare and analyze sample efficiency of two successful state-of-the-art NE algorithms: CMA-NeuroES and NEAT in a gait generation task of a quadruped robot. Moreover, we run both algorithms with various initial topologies in order to analyze the performance difference between each topology. From experimental results, we show CMA-NeuroES outperforms NEAT regardless of initial topologies when the limited number of episodes can be executed. Additional experiments conclude that the optimization method for connection weights in NEAT results in its inferior performance to CMA-NeuroES, while a probability-weighted averaging characteristic and self-adaptive factors make CMA-NeuroES to be advantageous.

Original languageEnglish
Title of host publication2013 IEEE Congress on Evolutionary Computation, CEC 2013
Pages2170-2177
Number of pages8
DOIs
Publication statusPublished - 2013 Aug 21
Externally publishedYes
Event2013 IEEE Congress on Evolutionary Computation, CEC 2013 - Cancun, Mexico
Duration: 2013 Jun 202013 Jun 23

Other

Other2013 IEEE Congress on Evolutionary Computation, CEC 2013
CountryMexico
CityCancun
Period13/6/2013/6/23

Fingerprint

Neuroevolution
Robot
Robots
Topology
Costs
Optimise
Trial and error
Reinforcement learning
Gait
Reinforcement Learning
Artificial Neural Network
Averaging
Optimization Methods
Choose
Neural networks
Policy
Experimental Results
Experiment
Experiments

Keywords

  • CMA-NeuroES
  • evolution
  • NEAT
  • neural network
  • neuroevolution

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Theoretical Computer Science

Cite this

Xu, S., Moriguch, H., & Honiden, S. (2013). Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot. In 2013 IEEE Congress on Evolutionary Computation, CEC 2013 (pp. 2170-2177). [6557826] https://doi.org/10.1109/CEC.2013.6557826

Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot. / Xu, Shengbo; Moriguch, Hirotaka; Honiden, Shinichi.

2013 IEEE Congress on Evolutionary Computation, CEC 2013. 2013. p. 2170-2177 6557826.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Xu, S, Moriguch, H & Honiden, S 2013, Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot. in 2013 IEEE Congress on Evolutionary Computation, CEC 2013., 6557826, pp. 2170-2177, 2013 IEEE Congress on Evolutionary Computation, CEC 2013, Cancun, Mexico, 13/6/20. https://doi.org/10.1109/CEC.2013.6557826
Xu S, Moriguch H, Honiden S. Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot. In 2013 IEEE Congress on Evolutionary Computation, CEC 2013. 2013. p. 2170-2177. 6557826 https://doi.org/10.1109/CEC.2013.6557826
Xu, Shengbo ; Moriguch, Hirotaka ; Honiden, Shinichi. / Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot. 2013 IEEE Congress on Evolutionary Computation, CEC 2013. 2013. pp. 2170-2177
@inproceedings{00d914d520eb4332b011d5fc26cc6022,
title = "Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot",
abstract = "In reinforcement learning tasks with continuous state-action, parameterized policy search has been known to be a powerful method. Applying NeuroEvolution (NE) to optimizing the policy represented by artificial neural network (ANN) is a particularly active research field. In most cases, NE algorithms cost a large amount of trial-and-error (episode) to optimize policies. However, due to time and cost constraints, researchers and practitioners cannot repeat a number of episodes on physical robots. Thus, choosing an efficient NE algorithm is a key to optimize policies with limited time and cost. In this work, our goal is to help users to choose an efficient NE algorithm. We compare and analyze sample efficiency of two successful state-of-the-art NE algorithms: CMA-NeuroES and NEAT in a gait generation task of a quadruped robot. Moreover, we run both algorithms with various initial topologies in order to analyze the performance difference between each topology. From experimental results, we show CMA-NeuroES outperforms NEAT regardless of initial topologies when the limited number of episodes can be executed. Additional experiments conclude that the optimization method for connection weights in NEAT results in its inferior performance to CMA-NeuroES, while a probability-weighted averaging characteristic and self-adaptive factors make CMA-NeuroES to be advantageous.",
keywords = "CMA-NeuroES, evolution, NEAT, neural network, neuroevolution",
author = "Shengbo Xu and Hirotaka Moriguch and Shinichi Honiden",
year = "2013",
month = "8",
day = "21",
doi = "10.1109/CEC.2013.6557826",
language = "English",
isbn = "9781479904549",
pages = "2170--2177",
booktitle = "2013 IEEE Congress on Evolutionary Computation, CEC 2013",

}

TY - GEN

T1 - Sample efficiency analysis of Neuroevolution algorithms on a quadruped robot

AU - Xu, Shengbo

AU - Moriguch, Hirotaka

AU - Honiden, Shinichi

PY - 2013/8/21

Y1 - 2013/8/21

N2 - In reinforcement learning tasks with continuous state-action, parameterized policy search has been known to be a powerful method. Applying NeuroEvolution (NE) to optimizing the policy represented by artificial neural network (ANN) is a particularly active research field. In most cases, NE algorithms cost a large amount of trial-and-error (episode) to optimize policies. However, due to time and cost constraints, researchers and practitioners cannot repeat a number of episodes on physical robots. Thus, choosing an efficient NE algorithm is a key to optimize policies with limited time and cost. In this work, our goal is to help users to choose an efficient NE algorithm. We compare and analyze sample efficiency of two successful state-of-the-art NE algorithms: CMA-NeuroES and NEAT in a gait generation task of a quadruped robot. Moreover, we run both algorithms with various initial topologies in order to analyze the performance difference between each topology. From experimental results, we show CMA-NeuroES outperforms NEAT regardless of initial topologies when the limited number of episodes can be executed. Additional experiments conclude that the optimization method for connection weights in NEAT results in its inferior performance to CMA-NeuroES, while a probability-weighted averaging characteristic and self-adaptive factors make CMA-NeuroES to be advantageous.

AB - In reinforcement learning tasks with continuous state-action, parameterized policy search has been known to be a powerful method. Applying NeuroEvolution (NE) to optimizing the policy represented by artificial neural network (ANN) is a particularly active research field. In most cases, NE algorithms cost a large amount of trial-and-error (episode) to optimize policies. However, due to time and cost constraints, researchers and practitioners cannot repeat a number of episodes on physical robots. Thus, choosing an efficient NE algorithm is a key to optimize policies with limited time and cost. In this work, our goal is to help users to choose an efficient NE algorithm. We compare and analyze sample efficiency of two successful state-of-the-art NE algorithms: CMA-NeuroES and NEAT in a gait generation task of a quadruped robot. Moreover, we run both algorithms with various initial topologies in order to analyze the performance difference between each topology. From experimental results, we show CMA-NeuroES outperforms NEAT regardless of initial topologies when the limited number of episodes can be executed. Additional experiments conclude that the optimization method for connection weights in NEAT results in its inferior performance to CMA-NeuroES, while a probability-weighted averaging characteristic and self-adaptive factors make CMA-NeuroES to be advantageous.

KW - CMA-NeuroES

KW - evolution

KW - NEAT

KW - neural network

KW - neuroevolution

UR - http://www.scopus.com/inward/record.url?scp=84881586655&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84881586655&partnerID=8YFLogxK

U2 - 10.1109/CEC.2013.6557826

DO - 10.1109/CEC.2013.6557826

M3 - Conference contribution

SN - 9781479904549

SP - 2170

EP - 2177

BT - 2013 IEEE Congress on Evolutionary Computation, CEC 2013

ER -