Detail publikace

USAGE OF LEARNING IN FOUR-LEGGED ROBOT CONTROL

BŘEZINA, T. HOUŠKA, P. SINGULE, V.

Český název

USAGE OF LEARNING IN FOUR-LEGGED ROBOT CONTROL

Anglický název

USAGE OF LEARNING IN FOUR-LEGGED ROBOT CONTROL

Typ

článek ve sborníku ve WoS nebo Scopus

Jazyk

en

Originální abstrakt

The report describes a possible method of discretization of the continuous status space of the four-legged robot using simultaneous compositions of effects of basic control elements. The compositions are activated from a non-deterministic terminal automatic machine. In the design stage it is possible to use simple considerations for significant limitation of the number of statuses of the terminal automatic machines and the number of possible simultaneous compositions. The valuation of transfers of the terminal automatic machine corresponds with compositions of control elements. For such a prepared scheme of transfers there is applied the variant of Q-learning so as to gradually specify the estimation of the future benefit from individual transfers and so as there are activated the simultaneous compositions of control elements with the maximal estimation of future benefit.

Český abstrakt

The report describes a possible method of discretization of the continuous status space of the four-legged robot using simultaneous compositions of effects of basic control elements. The compositions are activated from a non-deterministic terminal automatic machine. In the design stage it is possible to use simple considerations for significant limitation of the number of statuses of the terminal automatic machines and the number of possible simultaneous compositions. The valuation of transfers of the terminal automatic machine corresponds with compositions of control elements. For such a prepared scheme of transfers there is applied the variant of Q-learning so as to gradually specify the estimation of the future benefit from individual transfers and so as there are activated the simultaneous compositions of control elements with the maximal estimation of future benefit.

Anglický abstrakt

The report describes a possible method of discretization of the continuous status space of the four-legged robot using simultaneous compositions of effects of basic control elements. The compositions are activated from a non-deterministic terminal automatic machine. In the design stage it is possible to use simple considerations for significant limitation of the number of statuses of the terminal automatic machines and the number of possible simultaneous compositions. The valuation of transfers of the terminal automatic machine corresponds with compositions of control elements. For such a prepared scheme of transfers there is applied the variant of Q-learning so as to gradually specify the estimation of the future benefit from individual transfers and so as there are activated the simultaneous compositions of control elements with the maximal estimation of future benefit.

Klíčová slova anglicky

Reinforcement learning, Q-learning, control of robots

Rok RIV

2002

Vydáno

24.09.2002

Nakladatel

Trenčianská univerzita Alexandra Dubčeka v Trenčíně

Místo

Trenčín

ISBN

80-88914-73-6

Kniha

SEKEL 2002

Číslo edice

1

Počet stran

4

BIBTEX


@inproceedings{BUT9662,
  author="Tomáš {Březina} and Pavel {Houška} and Vladislav {Singule},
  title="USAGE OF LEARNING IN FOUR-LEGGED ROBOT CONTROL",
  booktitle="SEKEL 2002",
  year="2002",
  month="September",
  publisher="Trenčianská univerzita Alexandra Dubčeka v Trenčíně",
  address="Trenčín",
  isbn="80-88914-73-6"
}