DMCA
Automatic Development of Robot Behaviour using Monte Carlo Methods
| Citations: | 1 - 0 self |
BibTeX
@MISC{Brusey_automaticdevelopment,
author = {James Brusey},
title = {Automatic Development of Robot Behaviour using Monte Carlo Methods},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Abstract. Designing behaviour-based control mechanisms is a notoriously difficult task. The problem for an individual behaviour can often be simply stated, but there are typically a myriad of possible approaches. Also, the efficiency of a behaviour is dependent on characteristics of the environment that may be difficult for the designer to take account of. Therefore, it would seem useful to be able to optimise a behaviour automatically. Of the available techniques for optimisation, Markov Decision Process (MDP) based methods are particularly applicable because they allow for non-determinism in the effect of actions. In this paper, we phrase the behavioural problem as a Markov Decision Process (MDP) and find an approximately optimal solution for a simple robot navigation problem. We use this as a basis for demonstrating a general method for improving behaviours automatically. 1
