Detail publikace

Q-LEARNING APPLICATION ON FOUR-LEGGED ROBOT WALKING CONTROL

Originální název

Q-LEARNING APPLICATION ON FOUR-LEGGED ROBOT WALKING CONTROL

Anglický název

Q-LEARNING APPLICATION ON FOUR-LEGGED ROBOT WALKING CONTROL

Jazyk

en

Originální abstrakt

Significant goal in walking robot design is realisation of autonomous system which is capable of motion in unknown environment. One of the possibilities how to reach necessary adaptability of control system without modelling the complex and unpredictable cases of system behaviour is the application of Q-learning. The contribution describes experiences with the use of Q-learning through discretization of continuous state space of four legged robot. Q-learning is applied to realization of simultaneous compositions of behaviours. These compositions are obtained by concurrent activations of two simple basic controllers instances and . Control process is created by optimization procedure that operates with penalty function with global minimum in control target.

Anglický abstrakt

Significant goal in walking robot design is realisation of autonomous system which is capable of motion in unknown environment. One of the possibilities how to reach necessary adaptability of control system without modelling the complex and unpredictable cases of system behaviour is the application of Q-learning. The contribution describes experiences with the use of Q-learning through discretization of continuous state space of four legged robot. Q-learning is applied to realization of simultaneous compositions of behaviours. These compositions are obtained by concurrent activations of two simple basic controllers instances and . Control process is created by optimization procedure that operates with penalty function with global minimum in control target.

BibTex


@inproceedings{BUT7886,
  author="Tomáš {Březina} and Pavel {Houška} and Vladislav {Singule}",
  title="Q-LEARNING APPLICATION ON FOUR-LEGGED ROBOT WALKING CONTROL",
  annote="Significant goal in walking robot design is realisation of autonomous system which is capable of motion in unknown environment. One of the possibilities how to reach necessary adaptability of control system without modelling the complex and unpredictable cases of system behaviour is the application of Q-learning. The contribution describes experiences with the use of Q-learning through discretization of continuous state space of four legged robot. Q-learning is applied to realization of simultaneous compositions of behaviours. These compositions are obtained by concurrent activations of two simple basic controllers instances   and  . Control process is created by optimization procedure that operates with penalty function with global minimum in control target.",
  address="Alexander Dubček University of Trenčín, Faculty of Mechatronics, Trenčín, SLOVAKIA",
  booktitle="Proceedings of 6th international symposium on Mechatronics",
  chapter="7886",
  institution="Alexander Dubček University of Trenčín, Faculty of Mechatronics, Trenčín, SLOVAKIA",
  year="2003",
  month="june",
  pages="66",
  publisher="Alexander Dubček University of Trenčín, Faculty of Mechatronics, Trenčín, SLOVAKIA",
  type="conference paper"
}