Autonomous learning of domain models using two-dimensional probability distributions

Witold Slowinski, Frank Guerin

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

An autonomous agent placed without any prior knowledge in an environment without goals or a reward function will need to develop a model of that environment using an unguided approach by discovering patters occurring in its observations. We expand on a prior algorithm which allows an agent to achieve that by learning clusters in probability distributions of one-dimensional sensory variables and propose a novel quadtree-based algorithm for two dimensions. We then evaluate it in a dynamic continuous domain involving a ball being thrown onto uneven terrain, simulated using a physics engine. Finally, we put forward criteria which can be used to evaluate a domain model without requiring goals and apply them to our work. We show that adding two-dimensional rules to the algorithm improves the model and that such models can be transferred to similar but previously-unseen environments.
Original languageEnglish
Title of host publicationInternational Conference on Development and Learning and Epigenetic Robotics 2013
PublisherInstitute of Electrical and Electronics Engineers (IEEE)
Pages1-6
Number of pages6
DOIs
Publication statusPublished - 20 Aug 2013
EventInternational Conference on Development and Learning and Epigenetic Robotics 2013 - Osaka, Japan., United Kingdom
Duration: 19 Aug 201321 Aug 2013

Conference

ConferenceInternational Conference on Development and Learning and Epigenetic Robotics 2013
CountryUnited Kingdom
CityOsaka, Japan.
Period19/08/1321/08/13

Fingerprint

Probability distributions
Autonomous agents
Physics
Engines

Cite this

Slowinski, W., & Guerin, F. (2013). Autonomous learning of domain models using two-dimensional probability distributions. In International Conference on Development and Learning and Epigenetic Robotics 2013 (pp. 1-6). Institute of Electrical and Electronics Engineers (IEEE). https://doi.org/10.1109/DevLrn.2013.6652524

Autonomous learning of domain models using two-dimensional probability distributions. / Slowinski, Witold; Guerin, Frank.

International Conference on Development and Learning and Epigenetic Robotics 2013. Institute of Electrical and Electronics Engineers (IEEE), 2013. p. 1-6.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Slowinski, W & Guerin, F 2013, Autonomous learning of domain models using two-dimensional probability distributions. in International Conference on Development and Learning and Epigenetic Robotics 2013. Institute of Electrical and Electronics Engineers (IEEE), pp. 1-6, International Conference on Development and Learning and Epigenetic Robotics 2013, Osaka, Japan., United Kingdom, 19/08/13. https://doi.org/10.1109/DevLrn.2013.6652524
Slowinski W, Guerin F. Autonomous learning of domain models using two-dimensional probability distributions. In International Conference on Development and Learning and Epigenetic Robotics 2013. Institute of Electrical and Electronics Engineers (IEEE). 2013. p. 1-6 https://doi.org/10.1109/DevLrn.2013.6652524
Slowinski, Witold ; Guerin, Frank. / Autonomous learning of domain models using two-dimensional probability distributions. International Conference on Development and Learning and Epigenetic Robotics 2013. Institute of Electrical and Electronics Engineers (IEEE), 2013. pp. 1-6
@inproceedings{dd4c838bbeed41c79835bdc23b035cd8,
title = "Autonomous learning of domain models using two-dimensional probability distributions",
abstract = "An autonomous agent placed without any prior knowledge in an environment without goals or a reward function will need to develop a model of that environment using an unguided approach by discovering patters occurring in its observations. We expand on a prior algorithm which allows an agent to achieve that by learning clusters in probability distributions of one-dimensional sensory variables and propose a novel quadtree-based algorithm for two dimensions. We then evaluate it in a dynamic continuous domain involving a ball being thrown onto uneven terrain, simulated using a physics engine. Finally, we put forward criteria which can be used to evaluate a domain model without requiring goals and apply them to our work. We show that adding two-dimensional rules to the algorithm improves the model and that such models can be transferred to similar but previously-unseen environments.",
author = "Witold Slowinski and Frank Guerin",
year = "2013",
month = "8",
day = "20",
doi = "10.1109/DevLrn.2013.6652524",
language = "English",
pages = "1--6",
booktitle = "International Conference on Development and Learning and Epigenetic Robotics 2013",
publisher = "Institute of Electrical and Electronics Engineers (IEEE)",

}

TY - GEN

T1 - Autonomous learning of domain models using two-dimensional probability distributions

AU - Slowinski, Witold

AU - Guerin, Frank

PY - 2013/8/20

Y1 - 2013/8/20

N2 - An autonomous agent placed without any prior knowledge in an environment without goals or a reward function will need to develop a model of that environment using an unguided approach by discovering patters occurring in its observations. We expand on a prior algorithm which allows an agent to achieve that by learning clusters in probability distributions of one-dimensional sensory variables and propose a novel quadtree-based algorithm for two dimensions. We then evaluate it in a dynamic continuous domain involving a ball being thrown onto uneven terrain, simulated using a physics engine. Finally, we put forward criteria which can be used to evaluate a domain model without requiring goals and apply them to our work. We show that adding two-dimensional rules to the algorithm improves the model and that such models can be transferred to similar but previously-unseen environments.

AB - An autonomous agent placed without any prior knowledge in an environment without goals or a reward function will need to develop a model of that environment using an unguided approach by discovering patters occurring in its observations. We expand on a prior algorithm which allows an agent to achieve that by learning clusters in probability distributions of one-dimensional sensory variables and propose a novel quadtree-based algorithm for two dimensions. We then evaluate it in a dynamic continuous domain involving a ball being thrown onto uneven terrain, simulated using a physics engine. Finally, we put forward criteria which can be used to evaluate a domain model without requiring goals and apply them to our work. We show that adding two-dimensional rules to the algorithm improves the model and that such models can be transferred to similar but previously-unseen environments.

U2 - 10.1109/DevLrn.2013.6652524

DO - 10.1109/DevLrn.2013.6652524

M3 - Conference contribution

SP - 1

EP - 6

BT - International Conference on Development and Learning and Epigenetic Robotics 2013

PB - Institute of Electrical and Electronics Engineers (IEEE)

ER -