Show simple item record

dc.contributor.author
Allshire, Arthur
dc.contributor.author
Mittal, Mayank
dc.contributor.author
Lodaya, Varun
dc.contributor.author
Makoviychuk, Viktor
dc.contributor.author
Makoviichuk, Denys
dc.contributor.author
Widmaier, Felix
dc.contributor.author
Wüthrich, Manuel
dc.contributor.author
Bauer, Stefan
dc.contributor.author
Handa, Ankur
dc.contributor.author
Garg, Animesh
dc.date.accessioned
2023-03-15T10:16:23Z
dc.date.available
2023-03-13T14:34:03Z
dc.date.available
2023-03-15T10:16:23Z
dc.date.issued
2022
dc.identifier.isbn
978-1-6654-7927-1
en_US
dc.identifier.isbn
978-1-6654-7928-8
en_US
dc.identifier.other
10.1109/IROS47612.2022.9981458
en_US
dc.identifier.uri
http://hdl.handle.net/20.500.11850/602976
dc.description.abstract
In-hand manipulation of objects is an important capability to enable robots to carry-out tasks which demand high levels of dexterity. This work presents a robot systems approach to learning dexterous manipulation tasks involving moving objects to arbitrary 6-DoF poses. We show empirical benefits, both in simulation and sim-to-real transfer, of using keypoint-based representations for object pose in policy observations and reward calculation to train a model-free reinforcement learning agent. By utilizing domain randomization strategies and large-scale training, we achieve a high success rate of 83% on a real TriFinger system, with a single policy able to perform grasping, ungrasping, and finger gaiting in order to achieve arbitrary poses within the workspace. We demonstrate that our policy can generalise to unseen objects, and success rates can be further improved through finetuning. With the aim of assisting further research in learning in-hand manipulation, we provide a detailed exposition of our system and make the codebase of our system available, along with checkpoints trained on billions of steps of experience, at https://s2r2-ig.github.io
en_US
dc.language.iso
en
en_US
dc.publisher
IEEE
en_US
dc.subject
Training
en_US
dc.subject
Fingers
en_US
dc.subject
Graphics processing units
en_US
dc.subject
Reinforcement learning
en_US
dc.subject
Grasping
en_US
dc.subject
Task analysis
en_US
dc.subject
Intelligent robots
en_US
dc.title
Transferring Dexterous Manipulation from GPU Simulation to a Remote Real-World TriFinger
en_US
dc.type
Conference Paper
dc.date.published
2022-12-26
ethz.book.title
2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
en_US
ethz.pages.start
11802
en_US
ethz.pages.end
11809
en_US
ethz.event
35th IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022)
en_US
ethz.event.location
Kyoto, Japan
en_US
ethz.event.date
October 23-27, 2022
en_US
ethz.identifier.wos
ethz.publication.place
Piscataway, NJ
en_US
ethz.publication.status
published
en_US
ethz.date.deposited
2023-03-13T14:34:10Z
ethz.source
WOS
ethz.eth
yes
en_US
ethz.availability
Metadata only
en_US
ethz.rosetta.installDate
2023-03-15T10:16:24Z
ethz.rosetta.lastUpdated
2023-03-15T10:16:24Z
ethz.rosetta.versionExported
true
ethz.COinS
ctx_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.atitle=Transferring%20Dexterous%20Manipulation%20from%20GPU%20Simulation%20to%20a%20Remote%20Real-World%20TriFinger&rft.date=2022&rft.spage=11802&rft.epage=11809&rft.au=Allshire,%20Arthur&Mittal,%20Mayank&Lodaya,%20Varun&Makoviychuk,%20Viktor&Makoviichuk,%20Denys&rft.isbn=978-1-6654-7927-1&978-1-6654-7928-8&rft.genre=proceeding&rft_id=info:doi/10.1109/IROS47612.2022.9981458&rft.btitle=2022%20IEEE/RSJ%20International%20Conference%20on%20Intelligent%20Robots%20and%20Systems%20(IROS)
 Search print copy at ETH Library

Files in this item

FilesSizeFormatOpen in viewer

There are no files associated with this item.

Publication type

Show simple item record