by F Stulp and M Beetz
Abstract:
Solving complex tasks successfully and efficiently not only depends on \textbackslashem what you do, but also \textbackslashem how you do it. Different task contexts have different performance measures, and thus require different ways of executing an action to optimize performance. Simply adding new actions that are tailored to perform well within a specific task context makes planning or action selection programming more difficult, as generality and adaptivity is lost. Rather, existing actions should be parametrized such that they optimize the task-specific performance measure. In this paper we propose a novel computation model for the execution of abstract action chains. In this computation model, a robot first learns situation-specific performance models of abstract actions. It then uses these models to automatically specialize the abstract actions for their execution in a given action chain. This specialization results in refined chains that are optimized for performance. As a side effect this behavior optimization also appears to produce action chains with seamless transitions between actions.
Reference:
Tailoring Action Parameterizations to Their Task Contexts (F Stulp and M Beetz), 2005. (IJCAI Workshop “Agents in Real-Time and Dynamic Environments”)
Bibtex Entry:
@book{stulp_tailoring_2005,
title = {Tailoring Action Parameterizations to Their Task Contexts},
year = {2005},
author = {F Stulp and M Beetz},
abstract = {Solving complex tasks successfully and efficiently not only depends
on {\textbackslash}em what you do, but also {\textbackslash}em how
you do it. Different task contexts have different performance measures,
and thus require different ways of executing an action to optimize
performance. Simply adding new actions that are tailored to perform
well within a specific task context makes planning or action selection
programming more difficult, as generality and adaptivity is lost.
Rather, existing actions should be parametrized such that they optimize
the task-specific performance measure. In this paper we propose a
novel computation model for the execution of abstract action chains.
In this computation model, a robot first learns situation-specific
performance models of abstract actions. It then uses these models
to automatically specialize the abstract actions for their execution
in a given action chain. This specialization results in refined chains
that are optimized for performance. As a side effect this behavior
optimization also appears to produce action chains with seamless
transitions between actions.},
url = {http://www.tzi.de/ṽisser/ijcai05/},
}