Imitation learning as cause-effect reasoning

Garrett Katz, Di Wei Huang, Rodolphe Gentili, James Reggia

Research output: Chapter in Book/Entry/PoemConference contribution

9 Scopus citations

Abstract

We propose a framework for general-purpose imitation learning centered on cause-effect reasoning. Our approach infers a hierarchical representation of a demonstrator’s intentions, which can explain why they acted as they did. This enables rapid generalization of the observed actions to new situations. We employ a novel causal inference algorithm with formal guarantees and connections to automated planning. Our approach is implemented and validated empirically using a physical robot, which successfully generalizes skills involving bimanual manipulation of composite objects in 3D. These results suggest that cause-effect reasoning is an effective unifying principle for cognitive-level imitation learning.

Original languageEnglish (US)
Title of host publicationArtificial General Intelligence - 9th International Conference, AGI 2016, Proceedings
EditorsBas Steunebrink, Pei Wang, Ben Goertzel
PublisherSpringer Verlag
Pages64-73
Number of pages10
ISBN (Print)9783319416489
DOIs
StatePublished - 2016
Externally publishedYes
Event9th International Conference on Artificial General Intelligence, AGI 2016 - New York, United States
Duration: Jul 16 2016Jul 19 2016

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9782
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other9th International Conference on Artificial General Intelligence, AGI 2016
Country/TerritoryUnited States
CityNew York
Period7/16/167/19/16

Keywords

  • Artificial general intelligence
  • Cause effect reasoning
  • Cognitive robotics
  • Imitation learning
  • Parsimonious covering theory

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Imitation learning as cause-effect reasoning'. Together they form a unique fingerprint.

Cite this