Help on WARPLAN/Looking for PD Prolog Planner 
Author Message
 Help on WARPLAN/Looking for PD Prolog Planner

Hi,

since some days I am experimenting with David Warrens famous Prolog
planner (WARPLAN). It appears to me that this planner is rather
limited, since everytime I think I have my operators and state
descriptions correct WARPLAN runs into a loop. So I suspect that the
model I am working with is not appropriate for WARPLAN.

Actually I need the planner for a small simulation, where a robot
should run around in a world, consisting of different rooms, where he
needs to open doors, and where there are keys and other types of
objects distributed in the rooms. Obviously, the planner should avoid
loops, where he puts down and takes up the same object over and over
again, where he goes from one room to another and back again ... etc.

Unfortunately, the source from where I picked up WARPLAN (some
Springer book from Helder Coelho, I cannot remember the title at the
moment) does not describe it in detail, nor could I get the original
report (David Warren 1974, Edinburgh DAI Report).

Hence I would like to ask whether:

        - someone (maybe David Warren himself, if he reads this news)
          can tell me more about the limitations of WARPLAN
        - someone could send me a copy of the report mentioned above
        - someone has a more sophisticated hierarchical planner
          (hopefully documented) written in Prolog, which he could
          send me

ADV10xANCE,

        Thomas Hoppe

--

Technical University Berlin             phone : +49-30-314-24944
Franklinstr. 28/29                      fax   : +49-30-314-24929
10587 Berlin, Germany                   privat: +49-30-3236564          



Fri, 18 Oct 1996 22:20:45 GMT  
 Help on WARPLAN/Looking for PD Prolog Planner

                                Obviously, the planner should avoid
   loops, where he puts down and takes up the same object over and over
   again, where he goes from one room to another and back again ... etc.

I'm not the David Warren you seek, but I do have a comment. This kind
of looping is not avoided because Prolog searches the SLD tree, which
has this duplication. A tabling system, such as XSB, would allow you
to avoid such loops by simply declaring those predicates as
``tabled''. Someone earlier asked on the net, how often does a Prolog
programmer run into these kinds of loops. As I continue to read the
net, I discover that it happens reasonably often.

-David S. Warren
  University at Stony Brook



Sat, 19 Oct 1996 18:03:50 GMT  
 Help on WARPLAN/Looking for PD Prolog Planner


 > >                                 Obviously, the planner should avoid
 > > loops, where he puts down and takes up the same object over and over
 > >  again, where he goes from one room to another and back again ... etc.
 > I'm not the David Warren you seek, but I do have a comment. This kind
 > of looping is not avoided because Prolog searches the SLD tree, which
 > has this duplication. A tabling system, such as XSB, would allow you
 > to avoid such loops by simply declaring those predicates as
 > ``tabled''.
Actually, that kind of planning loop may not be solvable by tabling,
because it results from the interference of operators in trying to
achieve conjunctive goals. As the plannar backchains, it constructs
longer and longer useless plans, none of which is an instance of an
earlier one.
--
Fernando Pereira
2D-447, AT&T Bell Laboratories
600 Mountain Ave, PO Box 636
Murray Hill, NJ 07974-0636



Mon, 21 Oct 1996 07:48:32 GMT  
 Help on WARPLAN/Looking for PD Prolog Planner


    > .... A tabling system, such as XSB, would allow you
    > to avoid such loops by simply declaring those predicates as
    > ``tabled''.
   Actually, that kind of planning loop may not be solvable by tabling,
   because it results from the interference of operators in trying to
   achieve conjunctive goals. As the plannar backchains, it constructs
   longer and longer useless plans, none of which is an instance of an
   earlier one.

Interesting. Does that suggest that perhaps the chosen representation
is not the best? Maybe one should try to construct the reachable
states first, and then go back and construct how to reach the desired
one. Is it the case of finitely many different states, but infinitely
many plans to get to them?

-David
  David S. Warren
  University at Stony Brook



Tue, 22 Oct 1996 17:17:16 GMT  
 Help on WARPLAN/Looking for PD Prolog Planner

Quote:


>     > .... A tabling system, such as XSB, would allow you
>     > to avoid such loops by simply declaring those predicates as
>     > ``tabled''.
>    Actually, that kind of planning loop may not be solvable by tabling,
>    because it results from the interference of operators in trying to
>    achieve conjunctive goals. As the plannar backchains, it constructs
>    longer and longer useless plans, none of which is an instance of an
>    earlier one.
> Interesting. Does that suggest that perhaps the chosen representation
> is not the best? Maybe one should try to construct the reachable
> states first, and then go back and construct how to reach the desired
> one. Is it the case of finitely many different states, but infinitely
> many plans to get to them?

This stuff is by no means fresh in my mind, but it is certainly
possible to have infinitely many plans to get from state A to state B
(just need some undoable actions, as in the blocks world). In general
you cannot generate all the reachable states, there are too many of
them. However, there are more sophisticated algorithms that develop
partial plans with gaps to be filled in later, and with such
algorithms there may be more opportunities for taking advantage of
tabling. I haven't been paying much attention to the planning
literature, I'm afraid, but in the late 70s various ``nonlinear''
planners were developed that created plans as successively refined
partial orders between states (Sacerdoti's NOAH, Tate's NONLIN). David
H. D. Warren wrote a nonlinear planner in Prolog that used
a very clever term encoding of the partial plan (it depended crucially
on cyclic terms, by the way).

--
Fernando Pereira
2D-447, AT&T Bell Laboratories
600 Mountain Ave, PO Box 636
Murray Hill, NJ 07974-0636



Wed, 23 Oct 1996 07:07:18 GMT  
 
 [ 5 post ] 

 Relevant Pages 

1. Looking Prolog-written nonlinear planner

2. looking for PD prolog with C sources

3. Looking for PD Prolog list

4. Looking for a PD C-Prolog

5. LOOKING FOR PD PROLOGS

6. Amzi Prolog - Warplan exame: extending it

7. Prolog facts and rules for a course schedule planner

8. Partial Order Planners in Prolog

9. PROLOG implementation of planners

10. Help with WARPLAN

11. A prolog beginner looking for help!!

12. Help ! Looking for LPA Prolog for Windows

 

 
Powered by phpBB® Forum Software