by A Müller, A Kirsch and M Beetz
Abstract:
More than a decade after mobile robots arrived in many research labs it is still difficult to find plan-based autonomous robot controllers that perform, beyond doubt, better than they possibly could without applying AI methods. One of the main reason for this situation is abstraction. AI based control techniques typically abstract away from the mechanisms that generate the physical behavior and refuse the use of control structures that have proven to be necessary for producing flexible and reliable robot behavior. The consequence is: AI-based control mechanisms can neither explain and diagnose how a certain behavior resulted from a given plan nor can they revise the plans to improve its physical performance. In our view, a substantial improvement on this situation is not possible without having a new generation of robot control languages. These languages must, on the one hand, be expressive enough for specifying and producing high performance robot behavior and, on the other hand, be transparent and explicit enough to enable execution time inference mechanisms to reason about, and manipulate these control programs. This paper reports on aspects of the design of RPL-II, which we propose as such a next generation control language. We describe the nuts and bolts of extending our existing language R P L to support explicit models of physical systems, and object-oriented modeling of control tasks and programs. We show the application of these concepts in the context of autonomous robot soccer.
Reference:
Object-oriented Model-based Extensions of Robot Control Languages (A Müller, A Kirsch and M Beetz), In 27th German Conference on Artificial Intelligence, 2004.
Bibtex Entry:
@inproceedings{muller_object-oriented_2004,
author = {A Müller and A Kirsch and M Beetz},
title = {Object-oriented Model-based Extensions of Robot Control Languages},
booktitle = {27th German Conference on Artificial Intelligence},
year = {2004},
abstract = {More than a decade after mobile robots arrived in many research labs
it is still difficult to find plan-based autonomous robot controllers
that perform, beyond doubt, better than they possibly could without
applying {AI} methods. One of the main reason for this situation
is abstraction. {AI} based control techniques typically abstract
away from the mechanisms that generate the physical behavior and
refuse the use of control structures that have proven to be necessary
for producing flexible and reliable robot behavior. The consequence
is: {AI-based} control mechanisms can neither explain and diagnose
how a certain behavior resulted from a given plan nor can they revise
the plans to improve its physical performance. In our view, a substantial
improvement on this situation is not possible without having a new
generation of robot control languages. These languages must, on the
one hand, be expressive enough for specifying and producing high
performance robot behavior and, on the other hand, be transparent
and explicit enough to enable execution time inference mechanisms
to reason about, and manipulate these control programs. This paper
reports on aspects of the design of {RPL-II}, which we propose as
such a next generation control language. We describe the nuts and
bolts of extending our existing language R P L to support explicit
models of physical systems, and object-oriented modeling of control
tasks and programs. We show the application of these concepts in
the context of autonomous robot soccer.},
}