RE: Composition as planning

From the replies related to this thread I enlisted the following changes that were suggested to
adapt present AI planning to the WS world:

1. There is no canned planner that can handle the completely the WSC problem, different techniques
seem to be able to tackle only parts of this problem and again by adopting certain assumptions.
2. There has to be some idea about how to decide what service description to use at plan time and
which one to use at execution time. As I understand it, there has to be some interleaving between
the planning and execution phases in a composition process
3. Planning with WS is distinguished in that it involves planning with incomplete information
4. WSC tasks may or may not be described as a set of a goal state. In some instances they may be
described in terms of loosely coupled goals or constraints
5. In classical planning:
a. The world is assumed to be completely static except for the planner's actions
b. The planner is assumed to know everything
6. Due to the amount of possibly matched WS returned by a registry, scalability issues are required
to be solved
7. STRIPS planning only approximates the planning that has to be done in WSC, change does not occur
only under the control of the planner in WS environment. Such assumptions have to be relaxed in some
way.
8. Certain planning techniques have not yet been explored as solutions to WSC: partial-order
planning. What are the drawbacks here?
9. In WS planning is that loops seem unavoidable, every service that can satisfy the plan has to be
looped through to choose a satisfactory one.
10. Contingent planning seems to be more important than modelling uncertainty or adversarial
planning.

Comments, corrections, expansions on the above are welcome.

Another point I wish to put forward:

I found work done at http://robotics.comp.hkbu.edu.hk/~jiming/HonoursProjects/00010146/. There is a
semantic web-planning viewer which makes use of a partial order planning algorithm to generate plans
related to particular domains. There are also a number of ontologies related to STRIPS planning.

My questions are related to the idea of a planning ontology:

Will there be the need for such planning ontologies that define planning concepts like operators,
actions etc, as related to WSC (web services composition)?
If yes what advantages might this provide to the composition problem?
Could this be a starting point for standardising composed plans created by specific planning
technique?

Charlie


-------------------------------------------------
Charlie Abela

Received on Wednesday, 4 February 2004 18:50:56 UTC