next up previous
Next: EVENTS WITH MULTIPLE EFFECTS Up: SPECIFYING THE EFFECTS OF Previous: SPECIFYING THE EFFECTS OF

WHAT EVENTS OCCUR?

Before we consider combining narratives, we address a problem that arises in our new formalism that was not present in the earlier versions of the situation calculus.

Even frame axioms like this are not enough to allow us to carry out the simple reasoning we could carry out in previous versions of the situation calculus. We also need to know that certain events do not occur.

Consider the following example of moving a block. We have the action that moves a block, but to move a block successfully it must be clear. For instance if someone else puts another block on top of the block we are moving to, then our action will fail.

Thus our only effect axiom is the following one, which states moving block a onto block c succeeds, if a and c are clear for the entire duration, and a is not equal to c.

displaymath827

For this example we need some other facts about the world, these are given in an Appendix. We are also told that the only block that is not on the table is A, which is on B, and that the action of moving A to the top of C occurred at the situation S0.

displaymath828

We can now write our frame axiom, which in this case is,

displaymath829

This states that a block a is on a block c in a situation s if and only if a is on c in s', so long as there is no event e, of moving moving a to Top(a'), which occurs, and is successful.

Some writers like to think that if an event that might change a fluent f occurs, but fails, then the fluent f should be undetermined. We can weaken our frame axiom, so that even if the event of moving a block a' to a fails, then our block a might not be clear. We write this as,

displaymath830

In general we shall prefer the stronger frame axiom.

We wish to prove that

displaymath831

however, we can only prove the weaker,

displaymath832

Thus, we need to prove that A and C remain clear during the move action in order to show that the action is successful. To show that they remain clear we need to use our frame axiom. But, all we can prove is that the A block remains clear if there is no successful move action tex2html_wrap_inline901 that occurs, and whose result is before tex2html_wrap_inline903 . As we allow situations before S0, we can imagine that there was a move in progress that placed a block on C just after S0.

Thus we explicitly state that no action that might put something on A or C occurred in the intervalgif, save of course the action of putting A on the top of C.

displaymath833

We can now prove that

displaymath831

as the above principal allows us to prove that A and C are clear for the entire interval.

We cannot prove that C remains on the Table however, as there may be events that put C on top of other blocks. These events will not make C unclear, so they do not block the action of putting A on C. To prove that C remains on the table we would need,

displaymath835

It is notable that there was no need to state that no other actions occurred. It sufficedgif to say that no other events occurred that might cause a precondition of an event in our narrative to fail. The notion that we need only state that certain events did not occur becomes very important if we wish to axiomatize domains in a way that will later allow them to be conjoined. In fact, the motivating property for developing this new axiomatization was to allow separate axiomatizations, that do not interfere with each other, to be conjoined. This is not possible in the old-style situation calculus, as we explicitly list the sequence of actions that occurs. It is also not possible if we state that the only events that occur are those mentioned, as is sometimes done in narrative reasoning.

The reasoning that we did was not significantly more difficult than the usual reasoning in the situation calculus. We needed to check a few more conditions, namely that blocks remained clear, but strategies, such as goal regression[Rei91] continue to be effective.


next up previous
Next: EVENTS WITH MULTIPLE EFFECTS Up: SPECIFYING THE EFFECTS OF Previous: SPECIFYING THE EFFECTS OF

John McCarthy
Thu Jul 8 18:10:07 PDT 1999