• Citer

A reinforcement learning algorithm based on ‘safety’

par Nowe, Ann ;Vepa, Ranjan
Référence Lecture notes in computer science, 695 LNAI, page (47-58)
Publication Publié, 1993

Article révisé par les pairs

• Détails
• Contenu

Résumé :

In this paper, a reinforcement learning algorithm is presented which is used to implement a fuzzy controller model for a given control problem based on the concept of ‘safety’. A concept of ‘safety’ is postulated and learned iteratively. It embodies the notion of cost as well as performance. The fuzzy controller model which is based on this notion of safety is closely related to the controllability of the system and takes the physical constraints of the controller into account.