Information Technology Reference
In-Depth Information
The criterion of representational parsimony is the essence of the frame
problem; it requires that representation of the effects of actions should be
compact. It is difficult to quantify the compactness precisely. A reasonable
guideline is that the size of the representation should be roughly proportional to
the complexity of the domain, where a good indication of the complexity of the
domain is the total number of actions plus the total number of fluents.
A compact solution of the frame problem should be carry over to more
complicated domains. That is to say, we have to see that the solution meets the
second criterion, that of expressive flexibility. What is meant by a more
complicated domain is not simply one with a larger number of fluents and actions,
but rather one with features that demand a little extra thought before they can be
represented. Some features are listed as follows:
Ramifications,
Concurrent actions;
Non-deterministic actions; and
Continuous change.
An action will have ramifications beyond its immediate effects if we have to
take into account domain constraints. A domain constraint (sometimes called a
state constraint) is simply a constraint on what combinations of fluents may hold
in the sane situation. For example, suppose three are three blocks which are on
top of each other. Let Stack(
x,y,z
) be a fluent which denotes that blocks
x, y
, and
z
are a stack. Then we could write effect axioms for Stack like those previously
written for On, Clear and Colour. However, taken into account the domain
constraints, we write the effect axiom for Stack as follows:
Holds(Stack(
)
In the Blocks World examples considered so far, no two actions are ever
performed at the same time. But concurrent actions and events are ubiquitous in
everyday life. For example, there are concurrent actions if a block is moved by
more than one person.
Another feature of complicated domains that can trouble a potential solution
is the presence of non-deterministic actions or actions whose effects are not
completely known. For example, when we toss a coin, we known it will come
down either heads or tails, but we cannot say which.
Finally, we have the issue of continuous change. Continuous change is also
ubiquitous in everyday life, such as speeding cars, filling vessels, and so on. In
the Blocks World examples considered so far, all change is discrete. This doesn't
mean that the change being represented is discrete, but rather for the convenience
x, y, z
),
s
)
â
x
≠Table ∧ Holds(On(
y, x
),
s
) ∧ Holds(On(
z, y
),
s
