Amzi Prolog - Warplan exame: extending it 
Author Message
 Amzi Prolog - Warplan exame: extending it

Hi

Has any of you had any contact with the warplan sample which comes in
the amzi package ?

I'm trying to do a blocks world simulator with it. it seems to work
right, but I noticed a few glitches.

1st:

  c -> a
  b    b
  a    c

On this plan, the planner takes c and puts it in the table (correct).
But then it takes b, and puts it on the table to. Right after that, he
takes b again and stacks it on c. This step is absolutely redundant,
since it didn't need to put b on the table, but just stack it on c
directly.

2st:

On a big plan (say, about 7 blocks), it almost puts them all on the
table, and then reconstructs the goal from there. I don't see much
inteligence on that, since the planner should find a better way of
achieving the final state without using the table all the time.

What sould I modify in order to correct the 1st problem (the 2nd would
be nice to correct too, but I suspect it isn't so easy, due to the
nature of the planner itself). I think I should have some kind of
lookahed in the planner, but I'm really not into prolog enough to do
it all by myself.

Thanks in advance



Sat, 04 Jun 2005 19:16:47 GMT  
 Amzi Prolog - Warplan exame: extending it

Quote:

> Hi

> Has any of you had any contact with the warplan sample which comes in
> the amzi package ?

I haven't, but my thesis was on planning and I think I can tell you
what the problem is.  It sounds to me like the planner is either not
doing iterative deepening properly, or is starting with too great a
depth limit, or is increasing the depth limit on each iteration by too
great an amount.  Backwards chaining planners will often examine crazy
looking tactics and, if there is no reason to elide them (for instance,
because they make the plan too long), will preserve odd or even
pointless steps in the final plan.

That said, it shouldn't be too hard to knock up a plan optimizer to
remove redundant steps.

As far as dismantling blocks world problems and then assembling the
solution goes, that may not be the most efficient technique, but it
certainly requires the least effort!  Have you checked to see whether
more efficient solutions do indeed exist in the cases you've
observed?

- Ralph



Sat, 04 Jun 2005 20:53:10 GMT  
 Amzi Prolog - Warplan exame: extending it

Quote:

> I haven't, but my thesis was on planning and I think I can tell you
> what the problem is.  It sounds to me like the planner is either not
> doing iterative deepening properly, or is starting with too great a
> depth limit, or is increasing the depth limit on each iteration by too
> great an amount.  Backwards chaining planners will often examine crazy
> looking tactics and, if there is no reason to elide them (for instance,
> because they make the plan too long), will preserve odd or even
> pointless steps in the final plan.

> That said, it shouldn't be too hard to knock up a plan optimizer to
> remove redundant steps.

> As far as dismantling blocks world problems and then assembling the
> solution goes, that may not be the most efficient technique, but it
> certainly requires the least effort!  Have you checked to see whether
> more efficient solutions do indeed exist in the cases you've
> observed?

  Well.. it shouldn't be too hard if I had the proper Prolog knowledge
to do it.

  A more efficient planner exists, as long as it is non-linear and
hierarquic. But I couldn't find an implementation of it. TWEAK, SNLP,
NOAH and the likes are all of this kind. But they are rather more
complex in code. They involve graphs, critics, optimizations and so.

  Guess I must get a real good book on planners and prolog, to see if
they can help.



Mon, 06 Jun 2005 01:56:33 GMT  
 Amzi Prolog - Warplan exame: extending it


Quote:
>Hi

>Has any of you had any contact with the warplan sample which comes in
>the amzi package ?

>I'm trying to do a blocks world simulator with it. it seems to work
>right, but I noticed a few glitches.

>1st:

>  c -> a
>  b    b
>  a    c

>On this plan, the planner takes c and puts it in the table (correct).
>But then it takes b, and puts it on the table to. Right after that, he
>takes b again and stacks it on c. This step is absolutely redundant,
>since it didn't need to put b on the table, but just stack it on c
>directly.

See "Artificial Intelligence Techniques in Prolog" by Yoav Shoham,
Chapter 8, Planning and Temp{*filter*}Reasoning. He talks about STRIPS
and problems as you have mentioned, then RSTRIPS and nonlinear
planners.

A.L.



Mon, 06 Jun 2005 10:45:51 GMT  
 
 [ 4 post ] 

 Relevant Pages 

1. Amzi Prolog versus LPA Prolog.

2. Amzi-Prolog vs. Visual Prolog

3. Help on WARPLAN/Looking for PD Prolog Planner

4. Searching Amzi Prolog with Smalltalk examples

5. Amzi Prolog - Newbie Question

6. Amzi Prolog for the Web

7. Amzi! Prolog

8. Random in Amzi Prolog

9. How Do I: Browser-Enable Amzi Prolog Program

10. Amzi Prolog and SQL Server

11. How to use COM with Amzi Prolog

12. AMZI! Prolog and for/4

 

 
Powered by phpBB® Forum Software