Vision-based imitation learning in heterogeneous multi-robot systems: Varying physiology and skill

Jeff Allen, John Anderson, Jacky Baltes

Research output: Contribution to journalArticle

Abstract

Imitation learning enables a learner to improve its abilities by observing others. Most robotic imitation learning systems only learn from demonstrators that are similar physically and in terms of skill level. In order to employ imitation learning in a heterogeneous multi-agent environment, we must consider both differences in skill, and physical differences (physiology, size). This paper describes an approach to imitation learning from heterogeneous demonstrators, using global vision. It supports learning from physiologically different demonstrators (wheeled and legged, of various sizes), and self-adapts to demonstrators with varying levels of skill. The latter allows different parts of a task to be learned from different individuals (that is, worthwhile parts of a task can still be learned from a poorly-performing demonstrator). We assume the imitator has no initial knowledge of the observable effects of its own actions, and train a set of Hidden Markov Models to create an understanding of the imitator's own abilities. We then use a combination of tracking sequences of primitives and predicting future primitives from existing combinations of primitives, using forward models to learn abstract behaviors from demonstrations. This approach is evaluated using a group of heterogeneous robots that have been previously used in RoboCup soccer competitions.

Original languageEnglish
Pages (from-to)147-161
Number of pages15
JournalInternational Journal of Automation and Smart Technology
Volume2
Issue number2
DOIs
Publication statusPublished - 2012

Fingerprint

Physiology
Hidden Markov models
Learning systems
Robotics
Demonstrations
Robots

Keywords

  • Heterogeneity
  • Hidden markov models
  • Imitation learning
  • Multi-robot systems
  • Robotic soccer

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Signal Processing
  • Human-Computer Interaction
  • Hardware and Architecture
  • Electrical and Electronic Engineering
  • Artificial Intelligence

Cite this

Vision-based imitation learning in heterogeneous multi-robot systems : Varying physiology and skill. / Allen, Jeff; Anderson, John; Baltes, Jacky.

In: International Journal of Automation and Smart Technology, Vol. 2, No. 2, 2012, p. 147-161.

Research output: Contribution to journalArticle

@article{14eb57ddbb87415389b2688a1e20026d,
title = "Vision-based imitation learning in heterogeneous multi-robot systems: Varying physiology and skill",
abstract = "Imitation learning enables a learner to improve its abilities by observing others. Most robotic imitation learning systems only learn from demonstrators that are similar physically and in terms of skill level. In order to employ imitation learning in a heterogeneous multi-agent environment, we must consider both differences in skill, and physical differences (physiology, size). This paper describes an approach to imitation learning from heterogeneous demonstrators, using global vision. It supports learning from physiologically different demonstrators (wheeled and legged, of various sizes), and self-adapts to demonstrators with varying levels of skill. The latter allows different parts of a task to be learned from different individuals (that is, worthwhile parts of a task can still be learned from a poorly-performing demonstrator). We assume the imitator has no initial knowledge of the observable effects of its own actions, and train a set of Hidden Markov Models to create an understanding of the imitator's own abilities. We then use a combination of tracking sequences of primitives and predicting future primitives from existing combinations of primitives, using forward models to learn abstract behaviors from demonstrations. This approach is evaluated using a group of heterogeneous robots that have been previously used in RoboCup soccer competitions.",
keywords = "Heterogeneity, Hidden markov models, Imitation learning, Multi-robot systems, Robotic soccer",
author = "Jeff Allen and John Anderson and Jacky Baltes",
year = "2012",
doi = "10.5875/ausmt.v2i2.111",
language = "English",
volume = "2",
pages = "147--161",
journal = "International Journal of Automation and Smart Technology",
issn = "2223-9766",
publisher = "Chinese Institute of Automation Engineers (CIAE)",
number = "2",

}

TY - JOUR

T1 - Vision-based imitation learning in heterogeneous multi-robot systems

T2 - Varying physiology and skill

AU - Allen, Jeff

AU - Anderson, John

AU - Baltes, Jacky

PY - 2012

Y1 - 2012

N2 - Imitation learning enables a learner to improve its abilities by observing others. Most robotic imitation learning systems only learn from demonstrators that are similar physically and in terms of skill level. In order to employ imitation learning in a heterogeneous multi-agent environment, we must consider both differences in skill, and physical differences (physiology, size). This paper describes an approach to imitation learning from heterogeneous demonstrators, using global vision. It supports learning from physiologically different demonstrators (wheeled and legged, of various sizes), and self-adapts to demonstrators with varying levels of skill. The latter allows different parts of a task to be learned from different individuals (that is, worthwhile parts of a task can still be learned from a poorly-performing demonstrator). We assume the imitator has no initial knowledge of the observable effects of its own actions, and train a set of Hidden Markov Models to create an understanding of the imitator's own abilities. We then use a combination of tracking sequences of primitives and predicting future primitives from existing combinations of primitives, using forward models to learn abstract behaviors from demonstrations. This approach is evaluated using a group of heterogeneous robots that have been previously used in RoboCup soccer competitions.

AB - Imitation learning enables a learner to improve its abilities by observing others. Most robotic imitation learning systems only learn from demonstrators that are similar physically and in terms of skill level. In order to employ imitation learning in a heterogeneous multi-agent environment, we must consider both differences in skill, and physical differences (physiology, size). This paper describes an approach to imitation learning from heterogeneous demonstrators, using global vision. It supports learning from physiologically different demonstrators (wheeled and legged, of various sizes), and self-adapts to demonstrators with varying levels of skill. The latter allows different parts of a task to be learned from different individuals (that is, worthwhile parts of a task can still be learned from a poorly-performing demonstrator). We assume the imitator has no initial knowledge of the observable effects of its own actions, and train a set of Hidden Markov Models to create an understanding of the imitator's own abilities. We then use a combination of tracking sequences of primitives and predicting future primitives from existing combinations of primitives, using forward models to learn abstract behaviors from demonstrations. This approach is evaluated using a group of heterogeneous robots that have been previously used in RoboCup soccer competitions.

KW - Heterogeneity

KW - Hidden markov models

KW - Imitation learning

KW - Multi-robot systems

KW - Robotic soccer

UR - http://www.scopus.com/inward/record.url?scp=84905675761&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84905675761&partnerID=8YFLogxK

U2 - 10.5875/ausmt.v2i2.111

DO - 10.5875/ausmt.v2i2.111

M3 - Article

AN - SCOPUS:84905675761

VL - 2

SP - 147

EP - 161

JO - International Journal of Automation and Smart Technology

JF - International Journal of Automation and Smart Technology

SN - 2223-9766

IS - 2

ER -