Information Technology Reference
In-Depth Information
Hierarchical Plan-Based Control in Open-Ended
Environments: Considering Knowledge Acquisition
Opportunities
Dominik Off and Jianwei Zhang
TAMS, University of Hamburg, Vogt-Koelln-Strasse 30, Hamburg, Germany
{
off,zhang
}
@informatik.uni-hamburg.de
http://tams-www.informatik.uni-hamburg.de/
Abstract.
We introduce a novel hierarchical planning approach that extends pre-
vious approaches by additionally considering decompositions that are only appli-
cable with respect to a consistent extension of the (open-ended) domain model at
hand. The introduced planning approach is integrated into a plan-based control
architecture that interleaves planning and execution automatically so that miss-
ing information can be acquired by means of active knowledge acquisition. If it
is more reasonable, or even necessary, to acquire additional information prior to
making the next planning decision, the planner postpones the overall planning
process, and the execution of appropriate knowledge acquisition tasks is auto-
matically integrated into the overall planning and execution process.
Keywords:
Plan-based Control, Continual Planning, HTN Planning, Reasoning,
Knowledge Representation, Plan Execution.
1
Introduction
Planning their future course of action is particularly difficult for agents (e.g., robots) that
act in a
dynamic
and
open-ended
environment where it is unreasonable to assume that
a complete representation of the state of the domain is available. We define an
open-
ended domain
as a domain in which agents can in general neither be sure of having
all information nor of knowing all possible states (e.g., all objects) of the world they
inhabit.
Conformant
,
contingent
or
probabilistic
planning approaches can be used to generate
plans in situations where insufficient information is available at planning time [13,4].
These approaches generate conditional plans—or policies—for all possible contingen-
cies. Unfortunately, these approaches are computationally hard, scale badly in dynamic
unstructured domains and are only applicable if it is possible to foresee all possible
outcomes of a knowledge acquisition process [12,8]. Therefore, these approaches can
hardly be applied to the dynamic and open-ended domains we are interested in. Con-
sider, for example, a robot agent that is instructed to bring Bob's mug into the kitchen,
but does not know the location of the mug. Generating a plan for all possible locations
in a three dimensional space obviously is unreasonable and practically impossible.
A more promising approach for agents that act in open-ended domains is
continual
planning
[2] which enables the interleaving of planning and execution so that missing
