Re: Effects in OWL-S

> > [Bijan Parsia]
> >Just out of curiosity, does PDDL handle your model correctly? NIST PSL? ( 
> >I would expect the latter to.)
> 
> [Austin Tate]
> NIST PSL has node ends based constraints so can handle this.
> 
> PDDL - I am not sure.  Drew may be able to tell us? 

The current official version of PDDL (the one used in this year's
planning competition) has "durative actions," actions which take a
certain amount of time to occur, during which other things can
happen.  However, it is still the case that discrete all effects occur
at the beginning or the end of the action execution.

There is an unofficial extension discussed in connection with the
previous competition (in 2002) in which discrete effects occur at the
start or end of an action, but continuous effects occur throughout.
For example, if fuel is being consumed, it doesn't vanish at the end,
but decreases continuously over the duration of the action.

I'm not clear on exactly what Austin's model allows.  If it allows
discrete effects to occur at predictable offsets from the start time
of an action, then I assume one could express that in terms of a
series of simpler actions.  It might be a lot clumsier.

I have lobbied within the PDDL community to allow for a notion of
autonomous process, which is like an action except that it occurs
whenever its precondition is true.  If the precondition is true over
an interval, then the process occurs over that interval, and its
effects can include continouos changes in fluents.  By setting up
timer effects such as

     (when (=~ (elapsed (proc...)) (+ (start-time) 3))
        E)

one could have an effect E occur at a fixed interval after the start
of the process.   You can do a lot of other cool things, too.

                              -- Drew


-- 
                                             -- Drew McDermott
                                                Yale University CS Dept.

Received on Friday, 30 April 2004 12:16:38 UTC