Automated detection of handovers using kinematic features

Matthew KXJ Pan, Vidar Skjervøy, Wesley P Chan, Masayuki Inaba, Elizabeth A Croft

Research output: Contribution to journalArticleResearchpeer-review

Abstract

This paper investigates the use of kinematic motions recognized by a support vector machine (SVM) for the automatic detection of object handovers from the perspective of an object receiver. The classifier uses the giver’s kinematic behaviors (e.g. joint angles, distances of joints from each other and with respect to the receiver) to determine a giver’s intent to hand over an object. We used a bagged random forest to determine how informative features were in predicting the occurrence of handovers, and to assist in selecting a core set of features to be used by the classifier. Altogether, 22 kinematic features were chosen for developing handover detection models and later testing of generalization performance. Test results indicated an overall maximum accuracy of 97.5% by the SVM in its capacity to distinguish between handover and non-handover motions. The classification ability of the SVM was found to be unaffected across four kernel functions (linear, quadratic, cubic and radial basis). These results demonstrate considerable potential for detection of handovers and other gestures for human–robot interaction using kinematic features.

Original languageEnglish
Pages (from-to)721-738
Number of pages18
JournalInternational Journal of Robotics Research
Volume36
Issue number5-7
DOIs
Publication statusPublished - 2017
Externally publishedYes

Keywords

  • human-to-robot handover
  • human–robot interaction
  • Object handover

Cite this

Pan, Matthew KXJ ; Skjervøy, Vidar ; Chan, Wesley P ; Inaba, Masayuki ; Croft, Elizabeth A. / Automated detection of handovers using kinematic features. In: International Journal of Robotics Research. 2017 ; Vol. 36, No. 5-7. pp. 721-738.
@article{1177722c17914313aae13bb5db53448c,
title = "Automated detection of handovers using kinematic features",
abstract = "This paper investigates the use of kinematic motions recognized by a support vector machine (SVM) for the automatic detection of object handovers from the perspective of an object receiver. The classifier uses the giver’s kinematic behaviors (e.g. joint angles, distances of joints from each other and with respect to the receiver) to determine a giver’s intent to hand over an object. We used a bagged random forest to determine how informative features were in predicting the occurrence of handovers, and to assist in selecting a core set of features to be used by the classifier. Altogether, 22 kinematic features were chosen for developing handover detection models and later testing of generalization performance. Test results indicated an overall maximum accuracy of 97.5{\%} by the SVM in its capacity to distinguish between handover and non-handover motions. The classification ability of the SVM was found to be unaffected across four kernel functions (linear, quadratic, cubic and radial basis). These results demonstrate considerable potential for detection of handovers and other gestures for human–robot interaction using kinematic features.",
keywords = "human-to-robot handover, human–robot interaction, Object handover",
author = "Pan, {Matthew KXJ} and Vidar Skjerv{\o}y and Chan, {Wesley P} and Masayuki Inaba and Croft, {Elizabeth A}",
year = "2017",
doi = "10.1177/0278364917692865",
language = "English",
volume = "36",
pages = "721--738",
journal = "International Journal of Robotics Research",
issn = "0278-3649",
publisher = "SAGE Publications Ltd",
number = "5-7",

}

Automated detection of handovers using kinematic features. / Pan, Matthew KXJ; Skjervøy, Vidar; Chan, Wesley P; Inaba, Masayuki; Croft, Elizabeth A.

In: International Journal of Robotics Research, Vol. 36, No. 5-7, 2017, p. 721-738.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Automated detection of handovers using kinematic features

AU - Pan, Matthew KXJ

AU - Skjervøy, Vidar

AU - Chan, Wesley P

AU - Inaba, Masayuki

AU - Croft, Elizabeth A

PY - 2017

Y1 - 2017

N2 - This paper investigates the use of kinematic motions recognized by a support vector machine (SVM) for the automatic detection of object handovers from the perspective of an object receiver. The classifier uses the giver’s kinematic behaviors (e.g. joint angles, distances of joints from each other and with respect to the receiver) to determine a giver’s intent to hand over an object. We used a bagged random forest to determine how informative features were in predicting the occurrence of handovers, and to assist in selecting a core set of features to be used by the classifier. Altogether, 22 kinematic features were chosen for developing handover detection models and later testing of generalization performance. Test results indicated an overall maximum accuracy of 97.5% by the SVM in its capacity to distinguish between handover and non-handover motions. The classification ability of the SVM was found to be unaffected across four kernel functions (linear, quadratic, cubic and radial basis). These results demonstrate considerable potential for detection of handovers and other gestures for human–robot interaction using kinematic features.

AB - This paper investigates the use of kinematic motions recognized by a support vector machine (SVM) for the automatic detection of object handovers from the perspective of an object receiver. The classifier uses the giver’s kinematic behaviors (e.g. joint angles, distances of joints from each other and with respect to the receiver) to determine a giver’s intent to hand over an object. We used a bagged random forest to determine how informative features were in predicting the occurrence of handovers, and to assist in selecting a core set of features to be used by the classifier. Altogether, 22 kinematic features were chosen for developing handover detection models and later testing of generalization performance. Test results indicated an overall maximum accuracy of 97.5% by the SVM in its capacity to distinguish between handover and non-handover motions. The classification ability of the SVM was found to be unaffected across four kernel functions (linear, quadratic, cubic and radial basis). These results demonstrate considerable potential for detection of handovers and other gestures for human–robot interaction using kinematic features.

KW - human-to-robot handover

KW - human–robot interaction

KW - Object handover

UR - http://www.scopus.com/inward/record.url?scp=85044082940&partnerID=8YFLogxK

U2 - 10.1177/0278364917692865

DO - 10.1177/0278364917692865

M3 - Article

VL - 36

SP - 721

EP - 738

JO - International Journal of Robotics Research

JF - International Journal of Robotics Research

SN - 0278-3649

IS - 5-7

ER -