Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri
A study on a three-fingered robot hand with a 6-axis force/torque sensor and position-based impedance control was developed to execute texture recognition during grasping tasks. Force sensor data from grasping experiments by the robot hand for a bottle and a ball were used as inputs to the recogniti...
Saved in:
Main Authors: | , |
---|---|
Format: | Article |
Language: | English |
Published: |
UiTM Press
2022
|
Subjects: | |
Online Access: | https://ir.uitm.edu.my/id/eprint/63174/1/63174.pdf https://doi.org/10.24191/jeesr.v20i1.011 https://ir.uitm.edu.my/id/eprint/63174/ https://jeesr.uitm.edu.my/v1/ https://doi.org/10.24191/jeesr.v20i1.011 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
id |
my.uitm.ir.63174 |
---|---|
record_format |
eprints |
spelling |
my.uitm.ir.631742022-06-30T08:21:02Z https://ir.uitm.edu.my/id/eprint/63174/ Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri Roslan, A.B. Shauri, R.L.A. Neural networks (Computer science) Pattern recognition systems A study on a three-fingered robot hand with a 6-axis force/torque sensor and position-based impedance control was developed to execute texture recognition during grasping tasks. Force sensor data from grasping experiments by the robot hand for a bottle and a ball were used as inputs to the recognition algorithm. Moreover, the stiffness coefficient of the impedance parameter was varied to observe the difference of the force data for the different object textures. Based on the analysis results, the input and output of the artificial neural network (ANN), two layers feed forward network for the recognition process have been determined. The ANN simulations were divided into two simulations, first on the different amount of data used in the training and second, the simulation on selecting the suitable training method. Three training methods were chosen for the simulation i.e. Scaled Conjugate Gradient Backpropagation (SCG), Levenberg-Marquardt Backpropagation (LM), and Bayesian regularization Backpropagation (BR). From the experiments, SCG showed significant results with 72.7% accuracy compared to the LM and BR with 71.3% and 68.7%, respectively. UiTM Press 2022-04 Article PeerReviewed text en https://ir.uitm.edu.my/id/eprint/63174/1/63174.pdf Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri. (2022) Journal of Electrical and Electronic Systems Research (JEESR), 20: 11. pp. 77-83. ISSN 1985-5389 https://jeesr.uitm.edu.my/v1/ https://doi.org/10.24191/jeesr.v20i1.011 https://doi.org/10.24191/jeesr.v20i1.011 |
institution |
Universiti Teknologi Mara |
building |
Tun Abdul Razak Library |
collection |
Institutional Repository |
continent |
Asia |
country |
Malaysia |
content_provider |
Universiti Teknologi Mara |
content_source |
UiTM Institutional Repository |
url_provider |
http://ir.uitm.edu.my/ |
language |
English |
topic |
Neural networks (Computer science) Pattern recognition systems |
spellingShingle |
Neural networks (Computer science) Pattern recognition systems Roslan, A.B. Shauri, R.L.A. Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri |
description |
A study on a three-fingered robot hand with a 6-axis force/torque sensor and position-based impedance control was developed to execute texture recognition during grasping tasks. Force sensor data from grasping experiments by the robot hand for a bottle and a ball were used as inputs to the recognition algorithm. Moreover, the stiffness coefficient of the impedance parameter was varied to observe the difference of the force data for the different object textures. Based on the analysis results, the input and output of the artificial neural network (ANN), two layers feed forward network for the recognition process have been determined. The ANN simulations were divided into two simulations, first on the different amount of data used in the training and second, the simulation on selecting the suitable training method. Three training methods were chosen for the simulation i.e. Scaled Conjugate Gradient Backpropagation (SCG), Levenberg-Marquardt Backpropagation (LM), and Bayesian regularization Backpropagation (BR). From the experiments, SCG showed significant results with 72.7% accuracy compared to the LM and BR with 71.3% and 68.7%, respectively. |
format |
Article |
author |
Roslan, A.B. Shauri, R.L.A. |
author_facet |
Roslan, A.B. Shauri, R.L.A. |
author_sort |
Roslan, A.B. |
title |
Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri |
title_short |
Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri |
title_full |
Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri |
title_fullStr |
Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri |
title_full_unstemmed |
Object texture recognition based on grasping force data using feedforward neural network / A. B. Roslan and R. L. A. Shauri |
title_sort |
object texture recognition based on grasping force data using feedforward neural network / a. b. roslan and r. l. a. shauri |
publisher |
UiTM Press |
publishDate |
2022 |
url |
https://ir.uitm.edu.my/id/eprint/63174/1/63174.pdf https://doi.org/10.24191/jeesr.v20i1.011 https://ir.uitm.edu.my/id/eprint/63174/ https://jeesr.uitm.edu.my/v1/ https://doi.org/10.24191/jeesr.v20i1.011 |
_version_ |
1738514010571014144 |
score |
13.160551 |