Information Technology Reference
In-Depth Information
to imagine more complex domains demanding more and more categories,
which heavily increases the diculty of deciding which class a particular
fluent should belong to. In particular, the addition of constraints to a domain
specication may require a modication of the original categorization (as in
our example, where up ( s 2 ) moved to F s and light to F t )|which is to
be considered inadequate.
2.4 Causal Relationships
The dierent approaches with which we attempted to tackle the Ramica-
tion Problem thus far mainly revealed that accounting for indirect eects
of actions is more challenging an enterprise than it seems at rst glance.
Although the occurrence of indirect eects is triggered by the necessity to
correct violations of state constraints, the very constraints turned out not
to suce for recognizing logically proposed eects that would never occur in
reality. Providing additional domain knowledge in form of global classica-
tion of fluents proved too coarse in general as well. Whether or not a fluent
is expected to change as indirect eect is a more local property, depending
on the context. What we need, therefore, is a concept for generating indirect
eects on the basis of knowledge whose structure respects and reflects this
context-sensitivity. The notion of causal relationships will serve this purpose.
Being a key concept of this topic, these relationships will be introduced and
thoroughly analyzed in this and the following four sections.
2.4.1 Causal Relationships and Their Application
Causal relationships are formalizations of the circumstances under which the
occurrence of a single indirect eect is to be expected. Two components
constitute these circumstances. One of which describes the context required
for the indirect eect. The context is represented by a fluent formula that
needs to be satised in a state in order that the causal relationship applies.
The second component is a particular single eect whose prior occurrence
causes (hence the name) the indirect eect in question. This triggering eect
itself may have been previously obtained as indirect eect, if it was not among
the direct eects of the action.
The distinction between a context and a particular triggering eect is
essential. The reason for this will be elucidated in a moment. First, let us
consider an example of a causal relationship, which occurs in the electric
circuit with just two switches and the light bulb of Fig. 2.3. Suppose switch s 1
gets closed as a direct eect of some action. This eect is expected to cause
the light go on as indirect eect, provided switch
s 2 is closed, too. We write
this causal relationship as
up ( s 1 ) causes
if
up ( s 2 )
(2.2)
light
Search WWH ::




Custom Search