- From: Drew McDermott <drew.mcdermott@yale.edu>
- Date: Tue, 7 Dec 2004 21:06:19 +0000 (GMT)
- To: public-sws-ig@w3.org
> [Jeff (Dalton?) jeff@inf.ed.ac.uk] > However, HTN planners don't have to generate sequential plans. > With O-Plan, for example, the final plan is still only a partial > order. For most partial-order planners, if the end result is not totally ordered, that means that any total order consistent with the final partial order will work. So the semantics is still that of interleaved atomic actions, not "true" concurrency. I realize that the "O" in "O-Plan" means "open," so O-Plan probably doesn't insist on this interpretation (and probably doesn't support any particular interpretation). However (I love caveats), once we move beyond purely classical planning, and particular into domains with autonomous processes or durative actions, the fact that atomic actions are interpreted as non-concurrent becomes much less important. That's because atomic actions can start continuous changes in motion, and the semantics must now say what happens over any interval in which two or more continuous actions are all occurring. Insisting that you can't turn on the faucet and plug up the outlet at exactly the same time is unimportant if exactly the same water level will be reached no matter which one happened epsilon nanoseconds before the other. For an example in the web-services domain, it's harmless to insist that you can't order books from Amazon.com and BN.com at exactly the same time, because the shipping processes launched will proceed over intervals that might as well be identical. -- Drew -- -- Drew McDermott Yale University Computer Science Department
Received on Tuesday, 7 December 2004 21:06:53 UTC