Learning Object-Action Relations from Bimanual Human Demonstration Using Graph Networks

Recognizing human actions is a vital task for a humanoid robot, especially in domains like programming by demonstration. Previous approaches on action recognition primarily focused on the overall prevalent action being executed, but we argue that bimanual human motion cannot always be described suff...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Dreher, Christian R. G, Wächter, Mirko, Asfour, Tamim
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Dreher, Christian R. G
Wächter, Mirko
Asfour, Tamim
description Recognizing human actions is a vital task for a humanoid robot, especially in domains like programming by demonstration. Previous approaches on action recognition primarily focused on the overall prevalent action being executed, but we argue that bimanual human motion cannot always be described sufficiently with a single action label. We present a system for frame-wise action classification and segmentation in bimanual human demonstrations. The system extracts symbolic spatial object relations from raw RGB-D video data captured from the robot's point of view in order to build graph-based scene representations. To learn object-action relations, a graph network classifier is trained using these representations together with ground truth action labels to predict the action executed by each hand. We evaluated the proposed classifier on a new RGB-D video dataset showing daily action sequences focusing on bimanual manipulation actions. It consists of 6 subjects performing 9 tasks with 10 repetitions each, which leads to 540 video recordings with 2 hours and 18 minutes total playtime and per-hand ground truth action labels for each frame. We show that the classifier is able to reliably identify (action classification macro F1-score of 0.86) the true executed action of each hand within its top 3 predictions on a frame-by-frame basis without prior temporal action segmentation.
doi_str_mv 10.48550/arxiv.1908.08391
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_1908_08391</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>1908_08391</sourcerecordid><originalsourceid>FETCH-LOGICAL-a671-a23baed703887e11db4005b1ff8923b96ff9ded053ed0fc03c9def2d2aecaf1e3</originalsourceid><addsrcrecordid>eNotj8tOwzAQRb1hgQofwAr_QIIdN429LAVapIhKqFWX0cSegUAelZPy-Ps6gc3MHd2ZqzmM3UgRz3WaijvwP9VXLI3QsdDKyEt2yBF8W7VvfFt-oB2ipR2qruWvWMMoek6-a_h91UB7gppvTkHwB2yCNfhphe_78X7t4fjOX3D47vxnf8UuCOoer__7jO2eHnerTZRv18-rZR7BIpMRJKoEdJlQWmcopSvnQqSlJNImWGZBZBw6kapQyAplw0iJSwAtkEQ1Y7d_sRNZcfThT_9bjITFRKjOFtNN0w</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Learning Object-Action Relations from Bimanual Human Demonstration Using Graph Networks</title><source>arXiv.org</source><creator>Dreher, Christian R. G ; Wächter, Mirko ; Asfour, Tamim</creator><creatorcontrib>Dreher, Christian R. G ; Wächter, Mirko ; Asfour, Tamim</creatorcontrib><description>Recognizing human actions is a vital task for a humanoid robot, especially in domains like programming by demonstration. Previous approaches on action recognition primarily focused on the overall prevalent action being executed, but we argue that bimanual human motion cannot always be described sufficiently with a single action label. We present a system for frame-wise action classification and segmentation in bimanual human demonstrations. The system extracts symbolic spatial object relations from raw RGB-D video data captured from the robot's point of view in order to build graph-based scene representations. To learn object-action relations, a graph network classifier is trained using these representations together with ground truth action labels to predict the action executed by each hand. We evaluated the proposed classifier on a new RGB-D video dataset showing daily action sequences focusing on bimanual manipulation actions. It consists of 6 subjects performing 9 tasks with 10 repetitions each, which leads to 540 video recordings with 2 hours and 18 minutes total playtime and per-hand ground truth action labels for each frame. We show that the classifier is able to reliably identify (action classification macro F1-score of 0.86) the true executed action of each hand within its top 3 predictions on a frame-by-frame basis without prior temporal action segmentation.</description><identifier>DOI: 10.48550/arxiv.1908.08391</identifier><language>eng</language><subject>Computer Science - Robotics</subject><creationdate>2019-08</creationdate><rights>http://creativecommons.org/licenses/by-nc-sa/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/1908.08391$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.1908.08391$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Dreher, Christian R. G</creatorcontrib><creatorcontrib>Wächter, Mirko</creatorcontrib><creatorcontrib>Asfour, Tamim</creatorcontrib><title>Learning Object-Action Relations from Bimanual Human Demonstration Using Graph Networks</title><description>Recognizing human actions is a vital task for a humanoid robot, especially in domains like programming by demonstration. Previous approaches on action recognition primarily focused on the overall prevalent action being executed, but we argue that bimanual human motion cannot always be described sufficiently with a single action label. We present a system for frame-wise action classification and segmentation in bimanual human demonstrations. The system extracts symbolic spatial object relations from raw RGB-D video data captured from the robot's point of view in order to build graph-based scene representations. To learn object-action relations, a graph network classifier is trained using these representations together with ground truth action labels to predict the action executed by each hand. We evaluated the proposed classifier on a new RGB-D video dataset showing daily action sequences focusing on bimanual manipulation actions. It consists of 6 subjects performing 9 tasks with 10 repetitions each, which leads to 540 video recordings with 2 hours and 18 minutes total playtime and per-hand ground truth action labels for each frame. We show that the classifier is able to reliably identify (action classification macro F1-score of 0.86) the true executed action of each hand within its top 3 predictions on a frame-by-frame basis without prior temporal action segmentation.</description><subject>Computer Science - Robotics</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2019</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj8tOwzAQRb1hgQofwAr_QIIdN429LAVapIhKqFWX0cSegUAelZPy-Ps6gc3MHd2ZqzmM3UgRz3WaijvwP9VXLI3QsdDKyEt2yBF8W7VvfFt-oB2ipR2qruWvWMMoek6-a_h91UB7gppvTkHwB2yCNfhphe_78X7t4fjOX3D47vxnf8UuCOoer__7jO2eHnerTZRv18-rZR7BIpMRJKoEdJlQWmcopSvnQqSlJNImWGZBZBw6kapQyAplw0iJSwAtkEQ1Y7d_sRNZcfThT_9bjITFRKjOFtNN0w</recordid><startdate>20190822</startdate><enddate>20190822</enddate><creator>Dreher, Christian R. G</creator><creator>Wächter, Mirko</creator><creator>Asfour, Tamim</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20190822</creationdate><title>Learning Object-Action Relations from Bimanual Human Demonstration Using Graph Networks</title><author>Dreher, Christian R. G ; Wächter, Mirko ; Asfour, Tamim</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a671-a23baed703887e11db4005b1ff8923b96ff9ded053ed0fc03c9def2d2aecaf1e3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2019</creationdate><topic>Computer Science - Robotics</topic><toplevel>online_resources</toplevel><creatorcontrib>Dreher, Christian R. G</creatorcontrib><creatorcontrib>Wächter, Mirko</creatorcontrib><creatorcontrib>Asfour, Tamim</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Dreher, Christian R. G</au><au>Wächter, Mirko</au><au>Asfour, Tamim</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Learning Object-Action Relations from Bimanual Human Demonstration Using Graph Networks</atitle><date>2019-08-22</date><risdate>2019</risdate><abstract>Recognizing human actions is a vital task for a humanoid robot, especially in domains like programming by demonstration. Previous approaches on action recognition primarily focused on the overall prevalent action being executed, but we argue that bimanual human motion cannot always be described sufficiently with a single action label. We present a system for frame-wise action classification and segmentation in bimanual human demonstrations. The system extracts symbolic spatial object relations from raw RGB-D video data captured from the robot's point of view in order to build graph-based scene representations. To learn object-action relations, a graph network classifier is trained using these representations together with ground truth action labels to predict the action executed by each hand. We evaluated the proposed classifier on a new RGB-D video dataset showing daily action sequences focusing on bimanual manipulation actions. It consists of 6 subjects performing 9 tasks with 10 repetitions each, which leads to 540 video recordings with 2 hours and 18 minutes total playtime and per-hand ground truth action labels for each frame. We show that the classifier is able to reliably identify (action classification macro F1-score of 0.86) the true executed action of each hand within its top 3 predictions on a frame-by-frame basis without prior temporal action segmentation.</abstract><doi>10.48550/arxiv.1908.08391</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.1908.08391
ispartof
issn
language eng
recordid cdi_arxiv_primary_1908_08391
source arXiv.org
subjects Computer Science - Robotics
title Learning Object-Action Relations from Bimanual Human Demonstration Using Graph Networks
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-04T15%3A46%3A19IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Learning%20Object-Action%20Relations%20from%20Bimanual%20Human%20Demonstration%20Using%20Graph%20Networks&rft.au=Dreher,%20Christian%20R.%20G&rft.date=2019-08-22&rft_id=info:doi/10.48550/arxiv.1908.08391&rft_dat=%3Carxiv_GOX%3E1908_08391%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true