Information Technology Reference
In-Depth Information
be used as action parameters and enable the reasoning about unknown future knowl-
edge. Nevertheless, the information represented by runtime variables is limited since
the only thing that is known about them is the fact that they have been sensed. Further-
more, planning approaches that generate conditional plans are computationally hard,
scale badly in open-ended domains and are only applicable if it is possible to foresee
all possible outcomes of a sensing action [4,2].
The most closely related previous work is [2]. The proposed continual planning sys-
tem also deals with the challenge of generating a plan without initially having sufficient
information. In contrast to our work, this approach is based on classical planning sys-
tems that do not natively support the representation of incomplete state models and
are unable to exploit domain specific control knowledge in the form of HTN methods.
Moreover, it is not stated whether the approach can deal with open-ended domains in
which it is not only necessary to deal with incomplete information, but also essential to,
for example, consider the existence of a priori completely unknown objects or relations
between entities of a domain. Furthermore, the approach is based on the assumption
that all information about the precondition of a sensing action is a priori available and
thus will often (i.e., whenever this information is missing) fail to achieve a given goal
in an open-ended domain.
The Golog family of action languages—which are based on the situation calculus
[11]—have received much attention in the cognitive robotics community. The problem
of performing tasks in open-ended domains is most extensively considered by the In-
diGolog language [5], since programs are executed in an on-line manner and thus the
language to some degree is applicable to situations where the agent posses only incom-
plete information about the state of the world. Regrettably, IndiGolog only supports
binary sensing actions.
Besides Golog the only other known agent programming language is FLUX [14]
which is based on the Fluent Calculus. FLUX is a powerful formalism, but uses a re-
stricted form of conditional planning. As already pointed out, conditional planning is
not seen as an adequate approach for the scenarios we are interested in.
6
Discussion and Conclusions
State-of-the-art planning techniques can provide artificial agents to a certain degree
with autonomy and robustness. Unfortunately, reasoning about external information and
the acquisition of relevant knowledge has not been sufficiently considered in existing
planning approaches and is seen as an important direction of further growth [9].
We have proposed a new, continual, HTN-planning-based control system that can
reason about possible, relevant, and possibly-acquirable extensions of a domain model.
It makes an agent capable of autonomously generating and answering relevant ques-
tions. The domain specific information encoded in HTN methods not only helps to
prune the search space for classical planning problems but can also be exploited to rule
out irrelevant extensions of a domain model.
Planning in open-ended domains is obviously more difficult than planning based on
the assumption that all information is available at planning time. Nevertheless, the ex-
perimental results indicate that the proposed approach partitions the overall planning
