A causal multi-armed bandit approach for domestic robots’ failure avoidance

Nathan Ramoly, Amel Bouzeghoub, Beatrice Finance

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

Abstract

As there is a growing need for domestic healthcare, multiple projects are aiming to bring domestic robots in our homes. These robots aim to help users in their everyday life through various actions. However, they are subjected to task failure, making them less efficient and, possibly, bothering to the users. In this work, we aim to prevent task failures by understanding their causes through robot’s experience. In order to guarantee high accuracy, our approach uses highly semantic data as well as user validation. Our approach can consolidate its knowledge or discover new possible causes, and uses a multi-armed bandit solution: R-UCB. In order to make it more efficient, R-UCB was improved using causal induction and causal graphs. Experiments show our proposition to achieve a very high rate of correct failure prevention.

Original languageEnglish
Title of host publicationNeural Information Processing - 24th International Conference, ICONIP 2017, Proceedings
EditorsDerong Liu, Shengli Xie, Dongbin Zhao, Yuanqing Li, El-Sayed M. El-Alfy
PublisherSpringer Verlag
Pages90-99
Number of pages10
ISBN (Print)9783319701356
DOIs
Publication statusPublished - 1 Jan 2017
Externally publishedYes
Event24th International Conference on Neural Information Processing, ICONIP 2017 - Guangzhou, China
Duration: 14 Nov 201718 Nov 2017

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10639 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference24th International Conference on Neural Information Processing, ICONIP 2017
Country/TerritoryChina
CityGuangzhou
Period14/11/1718/11/17

Keywords

  • Domestic robotics
  • Experience
  • Multi-armed bandit
  • Ontologies
  • Reasoning
  • Task failure

Fingerprint

Dive into the research topics of 'A causal multi-armed bandit approach for domestic robots’ failure avoidance'. Together they form a unique fingerprint.

Cite this