Zero-Shot Terrain Generalization for Visual Locomotion Policies

Legged robots have unparalleled mobility on unstructured terrains. However, it remains an open challenge to design locomotion controllers that can operate in a large variety of environments. In this paper, we address this challenge of automatically learning locomotion controllers that can generalize...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Escontrela, Alejandro, Yu, George, Xu, Peng, Iscen, Atil, Tan, Jie
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 Escontrela, Alejandro
Yu, George
Xu, Peng
Iscen, Atil
Tan, Jie
description Legged robots have unparalleled mobility on unstructured terrains. However, it remains an open challenge to design locomotion controllers that can operate in a large variety of environments. In this paper, we address this challenge of automatically learning locomotion controllers that can generalize to a diverse collection of terrains often encountered in the real world. We frame this challenge as a multi-task reinforcement learning problem and define each task as a type of terrain that the robot needs to traverse. We propose an end-to-end learning approach that makes direct use of the raw exteroceptive inputs gathered from a simulated 3D LiDAR sensor, thus circumventing the need for ground-truth heightmaps or preprocessing of perception information. As a result, the learned controller demonstrates excellent zero-shot generalization capabilities and can navigate 13 different environments, including stairs, rugged land, cluttered offices, and indoor spaces with humans.
doi_str_mv 10.48550/arxiv.2011.05513
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2011_05513</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2011_05513</sourcerecordid><originalsourceid>FETCH-LOGICAL-a673-53755981a833d759b449d2b085d97548e9c9bc4613c7d55c0abd0f0c34e8553b3</originalsourceid><addsrcrecordid>eNotz81KAzEUBeBsupDWB3BlXmDGpDe3SVZSilZhQMHBhZvh5mcwMJ2UTBX16dXR1YGzOJyPsQspamUQxRWVj_Rer4WUtUCUcMauX2LJ1dNrPvE2lkJp5Ps4xkJD-qJTyiPvc-HPaXqjgTfZ50Oe28c8JJ_itGKLnoYpnv_nkrW3N-3urmoe9ve7bVPRRkOFoBGtkWQAgkbrlLJh7YTBYDUqE623zquNBK8DohfkguiFBxV_foODJbv8m50F3bGkA5XP7lfSzRL4BpSLQtQ</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Zero-Shot Terrain Generalization for Visual Locomotion Policies</title><source>arXiv.org</source><creator>Escontrela, Alejandro ; Yu, George ; Xu, Peng ; Iscen, Atil ; Tan, Jie</creator><creatorcontrib>Escontrela, Alejandro ; Yu, George ; Xu, Peng ; Iscen, Atil ; Tan, Jie</creatorcontrib><description>Legged robots have unparalleled mobility on unstructured terrains. However, it remains an open challenge to design locomotion controllers that can operate in a large variety of environments. In this paper, we address this challenge of automatically learning locomotion controllers that can generalize to a diverse collection of terrains often encountered in the real world. We frame this challenge as a multi-task reinforcement learning problem and define each task as a type of terrain that the robot needs to traverse. We propose an end-to-end learning approach that makes direct use of the raw exteroceptive inputs gathered from a simulated 3D LiDAR sensor, thus circumventing the need for ground-truth heightmaps or preprocessing of perception information. As a result, the learned controller demonstrates excellent zero-shot generalization capabilities and can navigate 13 different environments, including stairs, rugged land, cluttered offices, and indoor spaces with humans.</description><identifier>DOI: 10.48550/arxiv.2011.05513</identifier><language>eng</language><subject>Computer Science - Robotics</subject><creationdate>2020-11</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>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2011.05513$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2011.05513$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Escontrela, Alejandro</creatorcontrib><creatorcontrib>Yu, George</creatorcontrib><creatorcontrib>Xu, Peng</creatorcontrib><creatorcontrib>Iscen, Atil</creatorcontrib><creatorcontrib>Tan, Jie</creatorcontrib><title>Zero-Shot Terrain Generalization for Visual Locomotion Policies</title><description>Legged robots have unparalleled mobility on unstructured terrains. However, it remains an open challenge to design locomotion controllers that can operate in a large variety of environments. In this paper, we address this challenge of automatically learning locomotion controllers that can generalize to a diverse collection of terrains often encountered in the real world. We frame this challenge as a multi-task reinforcement learning problem and define each task as a type of terrain that the robot needs to traverse. We propose an end-to-end learning approach that makes direct use of the raw exteroceptive inputs gathered from a simulated 3D LiDAR sensor, thus circumventing the need for ground-truth heightmaps or preprocessing of perception information. As a result, the learned controller demonstrates excellent zero-shot generalization capabilities and can navigate 13 different environments, including stairs, rugged land, cluttered offices, and indoor spaces with humans.</description><subject>Computer Science - Robotics</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2020</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotz81KAzEUBeBsupDWB3BlXmDGpDe3SVZSilZhQMHBhZvh5mcwMJ2UTBX16dXR1YGzOJyPsQspamUQxRWVj_Rer4WUtUCUcMauX2LJ1dNrPvE2lkJp5Ps4xkJD-qJTyiPvc-HPaXqjgTfZ50Oe28c8JJ_itGKLnoYpnv_nkrW3N-3urmoe9ve7bVPRRkOFoBGtkWQAgkbrlLJh7YTBYDUqE623zquNBK8DohfkguiFBxV_foODJbv8m50F3bGkA5XP7lfSzRL4BpSLQtQ</recordid><startdate>20201110</startdate><enddate>20201110</enddate><creator>Escontrela, Alejandro</creator><creator>Yu, George</creator><creator>Xu, Peng</creator><creator>Iscen, Atil</creator><creator>Tan, Jie</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20201110</creationdate><title>Zero-Shot Terrain Generalization for Visual Locomotion Policies</title><author>Escontrela, Alejandro ; Yu, George ; Xu, Peng ; Iscen, Atil ; Tan, Jie</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a673-53755981a833d759b449d2b085d97548e9c9bc4613c7d55c0abd0f0c34e8553b3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2020</creationdate><topic>Computer Science - Robotics</topic><toplevel>online_resources</toplevel><creatorcontrib>Escontrela, Alejandro</creatorcontrib><creatorcontrib>Yu, George</creatorcontrib><creatorcontrib>Xu, Peng</creatorcontrib><creatorcontrib>Iscen, Atil</creatorcontrib><creatorcontrib>Tan, Jie</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Escontrela, Alejandro</au><au>Yu, George</au><au>Xu, Peng</au><au>Iscen, Atil</au><au>Tan, Jie</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Zero-Shot Terrain Generalization for Visual Locomotion Policies</atitle><date>2020-11-10</date><risdate>2020</risdate><abstract>Legged robots have unparalleled mobility on unstructured terrains. However, it remains an open challenge to design locomotion controllers that can operate in a large variety of environments. In this paper, we address this challenge of automatically learning locomotion controllers that can generalize to a diverse collection of terrains often encountered in the real world. We frame this challenge as a multi-task reinforcement learning problem and define each task as a type of terrain that the robot needs to traverse. We propose an end-to-end learning approach that makes direct use of the raw exteroceptive inputs gathered from a simulated 3D LiDAR sensor, thus circumventing the need for ground-truth heightmaps or preprocessing of perception information. As a result, the learned controller demonstrates excellent zero-shot generalization capabilities and can navigate 13 different environments, including stairs, rugged land, cluttered offices, and indoor spaces with humans.</abstract><doi>10.48550/arxiv.2011.05513</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2011.05513
ispartof
issn
language eng
recordid cdi_arxiv_primary_2011_05513
source arXiv.org
subjects Computer Science - Robotics
title Zero-Shot Terrain Generalization for Visual Locomotion Policies
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-30T09%3A21%3A21IST&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=Zero-Shot%20Terrain%20Generalization%20for%20Visual%20Locomotion%20Policies&rft.au=Escontrela,%20Alejandro&rft.date=2020-11-10&rft_id=info:doi/10.48550/arxiv.2011.05513&rft_dat=%3Carxiv_GOX%3E2011_05513%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