Publication detail
LEARNING BASED CONTROL SYSTEM OF FOUR-LEGGED ROBOT
BŘEZINA, T. HOUŠKA, P. SINGULE, V.
Czech title
LEARNING BASED CONTROL SYSTEM OF FOUR-LEGGED ROBOT
English title
LEARNING BASED CONTROL SYSTEM OF FOUR-LEGGED ROBOT
Type
conference paper
Language
cs
Original abstract
Possible discretization technique of the continuous state space of four-legged robot using simultaneous compositions of behaviors is described in this contribution. Compositions are generated by the instances of two basic controllers. The aim is to automatically develop the gait policy. Possible composition strategies are implemented through undeterministic state machine. In the machine design stage the number of both machine states and transitions could be essetially reduced using formal considerations only. Furthermore machine transitions are evaluated by selected alternative of Q-learning to incrementally improve the estimation of the future benefit from single transitions. This technique results in simultaneous activations of controller compositions instances with the highest estimation of the future benefit.
Czech abstract
Possible discretization technique of the continuous state space of four-legged robot using simultaneous compositions of behaviors is described in this contribution. Compositions are generated by the instances of two basic controllers. The aim is to automatically develop the gait policy. Possible composition strategies are implemented through undeterministic state machine. In the machine design stage the number of both machine states and transitions could be essetially reduced using formal considerations only. Furthermore machine transitions are evaluated by selected alternative of Q-learning to incrementally improve the estimation of the future benefit from single transitions. This technique results in simultaneous activations of controller compositions instances with the highest estimation of the future benefit.
English abstract
Possible discretization technique of the continuous state space of four-legged robot using simultaneous compositions of behaviors is described in this contribution. Compositions are generated by the instances of two basic controllers. The aim is to automatically develop the gait policy. Possible composition strategies are implemented through undeterministic state machine. In the machine design stage the number of both machine states and transitions could be essetially reduced using formal considerations only. Furthermore machine transitions are evaluated by selected alternative of Q-learning to incrementally improve the estimation of the future benefit from single transitions. This technique results in simultaneous activations of controller compositions instances with the highest estimation of the future benefit.
Keywords in Czech
Control Composition, Q-learning, Walking of Robot
Keywords in English
Control Composition, Q-learning, Walking of Robot
RIV year
2002
Released
13.05.2002
Publisher
Institute of Mechanics of Solids, Faculty of Mechanical Engineering, Brno University of Technology
Location
Brno
ISBN
80-214-2109-6
Book
Inženýrská mechanika 2002
Edition number
1
Pages count
2
BIBTEX
@inproceedings{BUT9660,
author="Tomáš {Březina} and Pavel {Houška} and Vladislav {Singule},
title="LEARNING BASED CONTROL SYSTEM OF FOUR-LEGGED ROBOT",
booktitle="Inženýrská mechanika 2002",
year="2002",
month="May",
publisher="Institute of Mechanics of Solids, Faculty of Mechanical Engineering, Brno University of Technology",
address="Brno",
isbn="80-214-2109-6"
}