DexPBT: Scaling up Dexterous Manipulation for Hand-Arm Systems with Population Based Training

In this work, we propose algorithms and methods that enable learning dexterous object manipulation using simulated one- or two-armed robots equipped with multi-fingered hand end-effectors. Using a parallel GPU-accelerated physics simulator (Isaac Gym), we implement challenging tasks for these robots...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	Petrenko, Aleksei, Allshire, Arthur, State, Gavriel, Handa, Ankur, Makoviychuk, Viktor
Format:	Artikel
Sprache:	eng
Schlagworte:	Computer Science - Artificial Intelligence Computer Science - Robotics
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	Petrenko, Aleksei Allshire, Arthur State, Gavriel Handa, Ankur Makoviychuk, Viktor
description	In this work, we propose algorithms and methods that enable learning dexterous object manipulation using simulated one- or two-armed robots equipped with multi-fingered hand end-effectors. Using a parallel GPU-accelerated physics simulator (Isaac Gym), we implement challenging tasks for these robots, including regrasping, grasp-and-throw, and object reorientation. To solve these problems we introduce a decentralized Population-Based Training (PBT) algorithm that allows us to massively amplify the exploration capabilities of deep reinforcement learning. We find that this method significantly outperforms regular end-to-end learning and is able to discover robust control policies in challenging tasks. Video demonstrations of learned behaviors and the code can be found at https://sites.google.com/view/dexpbt
doi_str_mv	10.48550/arxiv.2305.12127
format	Article
fullrecord	<record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2305_12127</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2305_12127</sourcerecordid><originalsourceid>FETCH-LOGICAL-a677-d82cc92c312c85c730636eee1d5544f19c022c004bf95ca2093e2508e9a42fa93</originalsourceid><addsrcrecordid>eNo9j11LwzAYRnPjhUx_gFfLH2hN3iRt4902PyZMHKy3o7ymiQbatKStbv_eOcWrBw4PBw4hN5ylslCK3WI8-M8UBFMpBw75Jdnf28N2Wd7RncHGh3c69fSERhu7aaAvGHw_NTj6LlDXRbrGUCeL2NLdcRhtO9AvP37Qbfd_WuJga1pG9OFkuyIXDpvBXv_tjJSPD-VqnWxen55Xi02CWZ4ndQHGaDCCgymUyQXLRGat5bVSUjquDQMwjMk3p5VBYFpYUKywGiU41GJG5r_ac1_VR99iPFY_ndW5U3wDRJtNcA</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>DexPBT: Scaling up Dexterous Manipulation for Hand-Arm Systems with Population Based Training</title><source>arXiv.org</source><creator>Petrenko, Aleksei ; Allshire, Arthur ; State, Gavriel ; Handa, Ankur ; Makoviychuk, Viktor</creator><creatorcontrib>Petrenko, Aleksei ; Allshire, Arthur ; State, Gavriel ; Handa, Ankur ; Makoviychuk, Viktor</creatorcontrib><description>In this work, we propose algorithms and methods that enable learning dexterous object manipulation using simulated one- or two-armed robots equipped with multi-fingered hand end-effectors. Using a parallel GPU-accelerated physics simulator (Isaac Gym), we implement challenging tasks for these robots, including regrasping, grasp-and-throw, and object reorientation. To solve these problems we introduce a decentralized Population-Based Training (PBT) algorithm that allows us to massively amplify the exploration capabilities of deep reinforcement learning. We find that this method significantly outperforms regular end-to-end learning and is able to discover robust control policies in challenging tasks. Video demonstrations of learned behaviors and the code can be found at https://sites.google.com/view/dexpbt</description><identifier>DOI: 10.48550/arxiv.2305.12127</identifier><language>eng</language><subject>Computer Science - Artificial Intelligence ; Computer Science - Robotics</subject><creationdate>2023-05</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</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/2305.12127$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2305.12127$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Petrenko, Aleksei</creatorcontrib><creatorcontrib>Allshire, Arthur</creatorcontrib><creatorcontrib>State, Gavriel</creatorcontrib><creatorcontrib>Handa, Ankur</creatorcontrib><creatorcontrib>Makoviychuk, Viktor</creatorcontrib><title>DexPBT: Scaling up Dexterous Manipulation for Hand-Arm Systems with Population Based Training</title><description>In this work, we propose algorithms and methods that enable learning dexterous object manipulation using simulated one- or two-armed robots equipped with multi-fingered hand end-effectors. Using a parallel GPU-accelerated physics simulator (Isaac Gym), we implement challenging tasks for these robots, including regrasping, grasp-and-throw, and object reorientation. To solve these problems we introduce a decentralized Population-Based Training (PBT) algorithm that allows us to massively amplify the exploration capabilities of deep reinforcement learning. We find that this method significantly outperforms regular end-to-end learning and is able to discover robust control policies in challenging tasks. Video demonstrations of learned behaviors and the code can be found at https://sites.google.com/view/dexpbt</description><subject>Computer Science - Artificial Intelligence</subject><subject>Computer Science - Robotics</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNo9j11LwzAYRnPjhUx_gFfLH2hN3iRt4902PyZMHKy3o7ymiQbatKStbv_eOcWrBw4PBw4hN5ylslCK3WI8-M8UBFMpBw75Jdnf28N2Wd7RncHGh3c69fSERhu7aaAvGHw_NTj6LlDXRbrGUCeL2NLdcRhtO9AvP37Qbfd_WuJga1pG9OFkuyIXDpvBXv_tjJSPD-VqnWxen55Xi02CWZ4ndQHGaDCCgymUyQXLRGat5bVSUjquDQMwjMk3p5VBYFpYUKywGiU41GJG5r_ac1_VR99iPFY_ndW5U3wDRJtNcA</recordid><startdate>20230520</startdate><enddate>20230520</enddate><creator>Petrenko, Aleksei</creator><creator>Allshire, Arthur</creator><creator>State, Gavriel</creator><creator>Handa, Ankur</creator><creator>Makoviychuk, Viktor</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20230520</creationdate><title>DexPBT: Scaling up Dexterous Manipulation for Hand-Arm Systems with Population Based Training</title><author>Petrenko, Aleksei ; Allshire, Arthur ; State, Gavriel ; Handa, Ankur ; Makoviychuk, Viktor</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a677-d82cc92c312c85c730636eee1d5544f19c022c004bf95ca2093e2508e9a42fa93</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Artificial Intelligence</topic><topic>Computer Science - Robotics</topic><toplevel>online_resources</toplevel><creatorcontrib>Petrenko, Aleksei</creatorcontrib><creatorcontrib>Allshire, Arthur</creatorcontrib><creatorcontrib>State, Gavriel</creatorcontrib><creatorcontrib>Handa, Ankur</creatorcontrib><creatorcontrib>Makoviychuk, Viktor</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Petrenko, Aleksei</au><au>Allshire, Arthur</au><au>State, Gavriel</au><au>Handa, Ankur</au><au>Makoviychuk, Viktor</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>DexPBT: Scaling up Dexterous Manipulation for Hand-Arm Systems with Population Based Training</atitle><date>2023-05-20</date><risdate>2023</risdate><abstract>In this work, we propose algorithms and methods that enable learning dexterous object manipulation using simulated one- or two-armed robots equipped with multi-fingered hand end-effectors. Using a parallel GPU-accelerated physics simulator (Isaac Gym), we implement challenging tasks for these robots, including regrasping, grasp-and-throw, and object reorientation. To solve these problems we introduce a decentralized Population-Based Training (PBT) algorithm that allows us to massively amplify the exploration capabilities of deep reinforcement learning. We find that this method significantly outperforms regular end-to-end learning and is able to discover robust control policies in challenging tasks. Video demonstrations of learned behaviors and the code can be found at https://sites.google.com/view/dexpbt</abstract><doi>10.48550/arxiv.2305.12127</doi><oa>free_for_read</oa></addata></record>
fulltext	fulltext_linktorsrc
identifier	DOI: 10.48550/arxiv.2305.12127
ispartof
issn
language	eng
recordid	cdi_arxiv_primary_2305_12127
source	arXiv.org
subjects	Computer Science - Artificial Intelligence Computer Science - Robotics
title	DexPBT: Scaling up Dexterous Manipulation for Hand-Arm Systems with Population Based Training
url	https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-13T08%3A56%3A27IST&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=DexPBT:%20Scaling%20up%20Dexterous%20Manipulation%20for%20Hand-Arm%20Systems%20with%20Population%20Based%20Training&rft.au=Petrenko,%20Aleksei&rft.date=2023-05-20&rft_id=info:doi/10.48550/arxiv.2305.12127&rft_dat=%3Carxiv_GOX%3E2305_12127%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