Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Deep predictive policy training using reinforcement learning
KTH, School of Computer Science and Communication (CSC), Robotics, perception and learning, RPL.ORCID iD: 0000-0001-6738-9872
KTH, School of Computer Science and Communication (CSC), Robotics, perception and learning, RPL.ORCID iD: 0000-0002-4266-6746
KTH, School of Computer Science and Communication (CSC), Robotics, perception and learning, RPL.
KTH, School of Computer Science and Communication (CSC), Robotics, perception and learning, RPL.
2017 (English)In: 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS 2017, Institute of Electrical and Electronics Engineers (IEEE), 2017, p. 2351-2358, article id 8206046Conference paper, Published paper (Refereed)
Abstract [en]

Skilled robot task learning is best implemented by predictive action policies due to the inherent latency of sensorimotor processes. However, training such predictive policies is challenging as it involves finding a trajectory of motor activations for the full duration of the action. We propose a data-efficient deep predictive policy training (DPPT) framework with a deep neural network policy architecture which maps an image observation to a sequence of motor activations. The architecture consists of three sub-networks referred to as the perception, policy and behavior super-layers. The perception and behavior super-layers force an abstraction of visual and motor data trained with synthetic and simulated training samples, respectively. The policy super-layer is a small subnetwork with fewer parameters that maps data in-between the abstracted manifolds. It is trained for each task using methods for policy search reinforcement learning. We demonstrate the suitability of the proposed architecture and learning framework by training predictive policies for skilled object grasping and ball throwing on a PR2 robot. The effectiveness of the method is illustrated by the fact that these tasks are trained using only about 180 real robot attempts with qualitative terminal rewards.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2017. p. 2351-2358, article id 8206046
National Category
Robotics
Identifiers
URN: urn:nbn:se:kth:diva-224269DOI: 10.1109/IROS.2017.8206046Scopus ID: 2-s2.0-85041944294ISBN: 9781538626825 OAI: oai:DiVA.org:kth-224269DiVA, id: diva2:1190800
Conference
2017 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS 2017, Vancouver, Canada, 24 September 2017 through 28 September 2017
Funder
Swedish Research CouncilEU, Horizon 2020
Note

QC 20180315

Available from: 2018-03-15 Created: 2018-03-15 Last updated: 2018-05-21Bibliographically approved
In thesis
1. Sensorimotor Robot Policy Training using Reinforcement Learning
Open this publication in new window or tab >>Sensorimotor Robot Policy Training using Reinforcement Learning
2018 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Robots are becoming more ubiquitous in our society and taking over many tasks that were previously considered as human hallmarks. Many of these tasks, e.g., autonomously driving a car, collaborating with humans in dynamic and changing working conditions and performing household chores, require human-level intelligence to perceive the world and to act appropriately. In this thesis, we pursue a different approach compared to classical methods that often construct a robot controller based on the perception-then-action paradigm. We devise robotic action-selection policies by considering action-selection and perception processes as being intertwined, emphasizing that perception comes prior to action and action is key to perception. The main hypothesis is that complex robotic behaviors come as the result of mastering sensorimotor contingencies (SMCs), i.e., regularities between motor actions and associated changes in sensory observations, where SMCs can be seen as building blocks to skillful behaviors. We elaborate and investigate this hypothesis by deliberate design of frameworks which enable policy training merely based on data experienced by a robot,without intervention of human experts for analytical modelings or calibrations. In such circumstances, action policies can be obtained by reinforcement learning (RL) paradigm by making exploratory action decisions and reinforcing patterns of SMCs that lead to reward events for a given task. However, the dimensionality of sensorimotor spaces, complex dynamics of physical tasks, sparseness of reward events, limited amount of data from real-robot experiments, ambiguities of crediting past decisions and safety issues, which arise from exploratory actions of a physical robot, pose challenges to obtain a policy based on data-driven methods alone. In this thesis, we introduce our contributions to deal with the aforementioned issues by devising learning frameworks which endow a robot with the ability to integrate sensorimotor data to obtain action-selection policies. The effectiveness of the proposed frameworks is demonstrated by evaluating the methods on a number of real robotic tasks and illustrating the suitability of the methods to acquire different skills, to make sequential action-decisions in high-dimensional sensorimotor spaces, with limited data and sparse rewards.

Abstract [sv]

Robotar förekommer alltmer i dagens samhälle och tar över många av de uppgifter som tidigare betraktades som tillägnade människor. Flera av dessa uppgifter, som att exempelvis autonomt köra en bil, samarbeta med människor i dynamiska och föränderliga arbetsmiljöer, samt att utföra sysslor i hemmet, kräver mänsklig intelligens för att roboten ska uppfatta världen och agera på lämpligt sätt. I denna avhandling utgår vi ifrån ett annat tillvägagångssätt jämfört med de klassiska metoder för skapande av robotsystem som tidigare ofta byggde på en så kallad perception-then-action paradigm. Vi utformar strategier för val av robotaktioner genom att utgå ifrån att det finns ett önsesidigt beroende mellan perception och aktion, där perception kommer före aktion, samtidigt som aktion är nödvändigt för perception. Huvudhypotesen är att komplexa robotbeteenden kommer som ett resultat av att roboten lär sig bemästra så kallade sensorimotorkopplingar (SMC), dvs regelbundenheter mellan motoriska aktioner och dess motsvarande förändringar i sensoriska observationer, där SMC:ar kan ses som byggblock för komplexa beteenden. Vi utarbetar och undersöker denna hypotes genom att avsiktligt utforma en handfull robotexperiment där en robots kunskaper helt förvärvas utifrån sensorimotoriska data, utan intervention av mänskliga experter för analytisk modellering eller kalibreringar. Under sådana omständigheter är så kallad reinforcement learning (RL) en lämplig paradigm för val av aktioner, en paradigm helt baserad på sensoriska data och utförda motoraktioner, utan krav på handgjorda representationer av världen på hög nivå. Denna paradigm kan utnyttjas för att generera utforskande rörelsemönster och förstärka de sensorimotorkopplingar som leder till framgång för i viss given uppgift. Det finns dock flera faktorer som kompicerar sådan rent datadriven inlärning av beteenden, såsom den sensorimotoriska datans höga dimensionalitet, den fysiska uppgiftens komplexa dynamik, bristen och tvetydigheten i de experiment som leder till positiva utfall, den begränsade mängd experiment som kan göras på en verklig robot och säkerhetsaspekter. De bidrag som introduceras i denna avhandling avser att hantera ovannämnda problem, genom att skapa ramverk för inlärning som gör det möjligt för en robot att integrera sensorimotordata för inlärning av stratieger för val av aktioner. De föreslagna ramverkens effektivitet demonsteras genom att utvärdera metoder på ett antal verkliga robotuppgifter och illustrera metodernas lämplighet för inlärning av olika färdigheter som kräver sekvenser av aktioner utifrån högdimensionell sensorimotorisk data, trots en begränsad mängd experiment med positivt utfall.

Place, publisher, year, edition, pages
KTH Royal Institute of Technology, 2018. p. 80
Series
TRITA-EECS-AVL ; 2018:47
Keywords
Reinforcement Learning, Artificial Intelligence, Robot Learning, Sensorimotor, Policy Training
National Category
Computer and Information Sciences
Research subject
Computer Science
Identifiers
urn:nbn:se:kth:diva-228295 (URN)978-91-7729-825-0 (ISBN)
Public defence
2018-06-11, F3, Lindstedtsvägen 26, Stockholm, 14:00 (English)
Opponent
Supervisors
Note

QC 20180521

Available from: 2018-05-21 Created: 2018-05-21 Last updated: 2018-05-21Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records BETA

Ghadirzadeh, AliMaki, Atsuto

Search in DiVA

By author/editor
Ghadirzadeh, AliMaki, AtsutoKragic, DanicaBjörkman, Mårten
By organisation
Robotics, perception and learning, RPL
Robotics

Search outside of DiVA

GoogleGoogle Scholar

doi
isbn
urn-nbn

Altmetric score

doi
isbn
urn-nbn
Total: 1237 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf