Publicação

Deep reinforcement learning applied to a robotic pick-and-place application

Ver documento

Detalhes bibliográficos
Resumo:Industrial robot manipulators are widely used for repetitive applications that require high precision, like pick-and-place. In many cases, the movements of industrial robot manipulators are hard-coded or manually defined, and need to be adjusted if the objects being manipulated change position. To increase flexibility, an industrial robot should be able to adjust its configuration in order to grasp objects in variable/ unknown positions. This can be achieved by off-the-shelf visionbased solutions, but most require prior knowledge about each object to be manipulated. To address this issue, this work presents a ROS-based deep reinforcement learning solution to robotic grasping for a Collaborative Robot (Cobot) using a depth camera. The solution uses deep Q-learning to process the color and depth images and generate a ϵ- greedy policy used to define the robot action. The Q-values are estimated using Convolutional Neural Network (CNN) based on pre-trained models for feature extraction. Experiments were carried out in a simulated environment to compare the performance of four different pretrained CNN models (RexNext, MobileNet, MNASNet and DenseNet). Results show that the best performance in our application was reached by MobileNet, with an average of 84 % accuracy after training in simulated environment.
Autores principais:Gomes, Natanael Magno
Outros Autores:Martins, Felipe N.; Lima, José; Wörtche, Heinrich
Assunto:Cobots Reinforcement learning Computer vision Pick-and-place Grasping
Ano:2021
País:Portugal
Tipo de documento:comunicação em conferência
Tipo de acesso:acesso restrito
Instituição associada:Instituto Politécnico de Bragança
Idioma:inglês
Origem:Biblioteca Digital do IPB
_version_ 1867172952607293440
author Gomes, Natanael Magno
author2 Martins, Felipe N.
Lima, José
Wörtche, Heinrich
author2_role author
author
author
author_facet Gomes, Natanael Magno
Martins, Felipe N.
Lima, José
Wörtche, Heinrich
author_role author
contributor_name_str_mv Biblioteca Digital do IPB
country_str PT
creators_json_txt [{\"Person.name\":\"Gomes, Natanael Magno\"},{\"Person.name\":\"Martins, Felipe N.\"},{\"Person.name\":\"Lima, José\",\"Person.identifier.orcid\":\"0000-0001-7902-1207\"},{\"Person.name\":\"Wörtche, Heinrich\"}]
datacite.contributors.contributor.contributorName.fl_str_mv Biblioteca Digital do IPB
datacite.creators.creator.creatorName.fl_str_mv Gomes, Natanael Magno
Martins, Felipe N.
Lima, José
Wörtche, Heinrich
datacite.date.Accepted.fl_str_mv 2021-01-01T00:00:00Z
datacite.date.available.fl_str_mv 2022-04-05T14:21:52Z
datacite.date.embargoed.fl_str_mv 2022-04-05T14:21:52Z
datacite.rights.fl_str_mv http://purl.org/coar/access_right/c_16ec
datacite.subjects.subject.fl_str_mv Cobots
Reinforcement learning
Computer vision
Pick-and-place
Grasping
datacite.titles.title.fl_str_mv Deep reinforcement learning applied to a robotic pick-and-place application
dc.contributor.none.fl_str_mv Biblioteca Digital do IPB
dc.creator.none.fl_str_mv Gomes, Natanael Magno
Martins, Felipe N.
Lima, José
Wörtche, Heinrich
dc.date.Accepted.fl_str_mv 2021-01-01T00:00:00Z
dc.date.available.fl_str_mv 2022-04-05T14:21:52Z
dc.date.embargoed.fl_str_mv 2022-04-05T14:21:52Z
dc.format.none.fl_str_mv application/pdf
dc.identifier.none.fl_str_mv http://hdl.handle.net/10198/25357
dc.language.none.fl_str_mv eng
dc.publisher.none.fl_str_mv Springer Nature
dc.rights.cclincense.fl_str_mv http://creativecommons.org/licenses/by/4.0/
dc.rights.none.fl_str_mv http://purl.org/coar/access_right/c_16ec
dc.subject.none.fl_str_mv Cobots
Reinforcement learning
Computer vision
Pick-and-place
Grasping
dc.title.fl_str_mv Deep reinforcement learning applied to a robotic pick-and-place application
dc.type.none.fl_str_mv http://purl.org/coar/resource_type/c_5794
description Industrial robot manipulators are widely used for repetitive applications that require high precision, like pick-and-place. In many cases, the movements of industrial robot manipulators are hard-coded or manually defined, and need to be adjusted if the objects being manipulated change position. To increase flexibility, an industrial robot should be able to adjust its configuration in order to grasp objects in variable/ unknown positions. This can be achieved by off-the-shelf visionbased solutions, but most require prior knowledge about each object to be manipulated. To address this issue, this work presents a ROS-based deep reinforcement learning solution to robotic grasping for a Collaborative Robot (Cobot) using a depth camera. The solution uses deep Q-learning to process the color and depth images and generate a ϵ- greedy policy used to define the robot action. The Q-values are estimated using Convolutional Neural Network (CNN) based on pre-trained models for feature extraction. Experiments were carried out in a simulated environment to compare the performance of four different pretrained CNN models (RexNext, MobileNet, MNASNet and DenseNet). Results show that the best performance in our application was reached by MobileNet, with an average of 84 % accuracy after training in simulated environment.
dirty 0
eu_rights_str_mv restrictedAccess
format conferencePaper
fulltext.url.fl_str_mv https://bibliotecadigital.ipb.pt/bitstreams/cf06c6b5-54d1-4d55-89aa-ba7887db5757/download
funding.funder.alternateName_str_mv FCT
funding.funder.identifier_str_mv http://doi.org/10.13039/501100001871
funding.funder.name_str_mv Fundação para a Ciência e a Tecnologia
funding.name_str_mv 6817 - DCRRNI ID
id ipb_76d4cd577d095faf06bb5f7674e436e8
identifier.url.fl_str_mv http://hdl.handle.net/10198/25357
instacron_str ipb
institution Instituto Politécnico de Bragança
instname_str Instituto Politécnico de Bragança
language eng
network_acronym_str ipb
network_name_str Biblioteca Digital do IPB
oai_identifier_str oai:bibliotecadigital.ipb.pt:10198/25357
organization_str_mv urn:organizationAcronym:ipb
person_str_mv Gomes, Natanael Magno
Martins, Felipe N.
Lima, José
Lima, José
https://www.ciencia-id.pt/6016-C902-86A9
6016-C902-86A9
http://orcid.org/0000-0001-7902-1207
0000-0001-7902-1207
Wörtche, Heinrich
publishDate 2021
publisher.none.fl_str_mv Springer Nature
reponame_str Biblioteca Digital do IPB
repository_id_str urn:repositoryAcronym:ipb
service_str_mv urn:repositoryAcronym:ipb
spelling engSpringer Naturept_PTIndustrial robot manipulators are widely used for repetitive applications that require high precision, like pick-and-place. In many cases, the movements of industrial robot manipulators are hard-coded or manually defined, and need to be adjusted if the objects being manipulated change position. To increase flexibility, an industrial robot should be able to adjust its configuration in order to grasp objects in variable/ unknown positions. This can be achieved by off-the-shelf visionbased solutions, but most require prior knowledge about each object to be manipulated. To address this issue, this work presents a ROS-based deep reinforcement learning solution to robotic grasping for a Collaborative Robot (Cobot) using a depth camera. The solution uses deep Q-learning to process the color and depth images and generate a ϵ- greedy policy used to define the robot action. The Q-values are estimated using Convolutional Neural Network (CNN) based on pre-trained models for feature extraction. Experiments were carried out in a simulated environment to compare the performance of four different pretrained CNN models (RexNext, MobileNet, MNASNet and DenseNet). Results show that the best performance in our application was reached by MobileNet, with an average of 84 % accuracy after training in simulated environment.application/pdfpt_PTDeep reinforcement learning applied to a robotic pick-and-place applicationGomes, Natanael MagnoMartins, Felipe N.PersonalLima, JoséDSpacehttp://dspace.org/items/d88c2b2a-efc2-48ef-b1fd-1145475e0055DSpacehttp://dspace.org/items/d88c2b2a-efc2-48ef-b1fd-1145475e0055LimaJoséCiência IDhttps://www.ciencia-id.pt6016-C902-86A9ORCIDhttp://orcid.org0000-0001-7902-1207Researcher IDhttps://www.researcherid.comL-3370-2014Scopus Author IDhttps://www.scopus.com55851941311Wörtche, HeinrichHostingInstitutionOrganizationalBiblioteca Digital do IPBe-mailmailto:dspace@ipb.ptdspace@ipb.ptISBNIsPartOf978-3-030-91884-2DOIIsPartOf10.1007/978-3-030-91885-9_182022-04-05T14:21:52Z20212021-01-01T00:00:00ZHandlehttp://hdl.handle.net/10198/25357http://purl.org/coar/access_right/c_16ecrestricted accessCobotsReinforcement learningComputer visionPick-and-placeGrasping3586227 bytesFundação para a Ciência e a TecnologiaResearch Centre in Digitalization and Intelligent Robotics6817 - DCRRNI IDCrossref Funder IDhttp://doi.org/10.13039/501100001871other research producthttp://purl.org/coar/resource_type/c_5794conference paper2021http://creativecommons.org/licenses/by/4.0/http://purl.org/coar/access_right/c_16ecapplication/pdffulltexthttps://bibliotecadigital.ipb.pt/bitstreams/cf06c6b5-54d1-4d55-89aa-ba7887db5757/downloadOptimization, learning algorithms and applications: first International Conference, OL2A 20211488251265Bragança
spellingShingle Deep reinforcement learning applied to a robotic pick-and-place application
Gomes, Natanael Magno
Cobots
Reinforcement learning
Computer vision
Pick-and-place
Grasping
status SINGLETON
subject.fl_str_mv Cobots
Reinforcement learning
Computer vision
Pick-and-place
Grasping
title Deep reinforcement learning applied to a robotic pick-and-place application
title_full Deep reinforcement learning applied to a robotic pick-and-place application
title_fullStr Deep reinforcement learning applied to a robotic pick-and-place application
title_full_unstemmed Deep reinforcement learning applied to a robotic pick-and-place application
title_short Deep reinforcement learning applied to a robotic pick-and-place application
title_sort Deep reinforcement learning applied to a robotic pick-and-place application
topic Cobots
Reinforcement learning
Computer vision
Pick-and-place
Grasping
topic_facet Cobots
Reinforcement learning
Computer vision
Pick-and-place
Grasping
url http://hdl.handle.net/10198/25357
visible 1