Re: Effects in OWL-S

Following up a bit more.

On Apr 28, 2004, at 4:09 AM, Austin Tate wrote:

> In our AI planning work, we just call them all world state conditions 
> and world state effects.  We do not talk about them as pre-, post- 
> etc.
>
> Conditions on the world state can be stated as applying at the begin 
> time point of an activity or the end time point.  Notice I say "at" 
> the time point... rather than before or after the time point by the 
> way.  I.e. its an instantaneous point at which the conditions are 
> stated as holding.

Hmm. Ok. Interesting.

We have processes as subclasses of IntervalEvents. It wouldn't be that 
hard to define Conditions and Effects such that you could label so me 
C&Es as happening before, some after an Interval. (Perhaps even some 
that happened over the course of a Process, perhaps in some partial 
order; for CompositeProcesses, you have to do that).

> Effects that purport to change the world state can occur also at the 
> begin time point or the end time point.  Of course you have to decide 
> whether to believe you model of effects or whether to set up condition 
> monitoring mechanisms to ensure they actually occurred.

That's cool.

> I think its a serious limitation to think only of conditions on the 
> begin time point and effects on the end time point.  Even if we also 
> allow conditions on the end time point, we still leave out effects on 
> the world state that can or should be modelled as occurring earlier.

Personally, I tend to think this is more serious for 
CompositeProcesses, but that's guessing, not experience talking.

But it should be easy enough to add another field to our condition 
object that indicates timepoint of evaluation/application. Would this 
be worth anything to you?

> We have to mess about a bit to be able to translate this more flexible 
> conditions/effects model (which we have used in our planners for over 
> 20 years) into OWL-S with its more limited view of IOPE.

We could make the current PEs derived subclasses of the more general 
case pretty easily. Rename Effects to PostEffects and it even scans. I 
don't think calling out this common case is necessarily so bad.

Note that this doesn't have to be rolled into our release. It should be 
possible to do smoothly in any case. We should try it, Austin. I'd like 
to see what you've already done to accomodate your needs.

Cheers,
Bijan Parsia.

Received on Wednesday, 28 April 2004 07:43:00 UTC