Inverse reinforcement learning for dexterous hand manipulation

Jedrzej Orbik, Alejandro Agostini, Dongheui Lee

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Scopus citations

Abstract

The success of deep reinforcement learning approaches to learn dexterous manipulation skills strongly hinges on the rewards assigned to actions during task execution. The usual approach is to handcraft the reward function but due to the high complexity of dexterous manipulations the reward definition demands large engineering effort for each particular task. To avoid this burden, we use an inverse reinforcement learning (IRL) approach to automatically learn the reward function using samples obtained from demonstrations of desired behaviours. We have identified that the learned rewards using existing IRL approaches are strongly biased towards demonstrated actions due to the scarcity of samples in the vast state-action space of dexterous manipulation applications. This significantly hinders performance due to unreliable reward estimations in regions unexplored during demonstration. We use statistical tools for random sample generation and reward normalization to reduce this bias. We show that this approach improves learning stability and transferability of IRL for dexterous manipulation tasks. Project page: https://sites.google.con

Original languageEnglish
Title of host publicationIEEE International Conference on Development and Learning, ICDL 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781728162423
DOIs
StatePublished - 23 Aug 2021
Externally publishedYes
Event2021 IEEE International Conference on Development and Learning, ICDL 2021 - Virtual, Beijing, China
Duration: 23 Aug 202126 Aug 2021

Publication series

NameIEEE International Conference on Development and Learning, ICDL 2021

Conference

Conference2021 IEEE International Conference on Development and Learning, ICDL 2021
Country/TerritoryChina
CityVirtual, Beijing
Period23/08/2126/08/21

Keywords

  • Dexterous robotic hand
  • Inverse reinforcement learning
  • Learning from demonstrations
  • Robot skills prototyping

Fingerprint

Dive into the research topics of 'Inverse reinforcement learning for dexterous hand manipulation'. Together they form a unique fingerprint.

Cite this