action cost based on quantified effect

[David Epstein]

Hi! With an action like the one pasted below (from p. 71 of the book by Haslem et al 2019), how could the action cost in FD vary with the number of passengers matching the forall? In my own code, I'd like to adjust the cost of actions based on, for instance, whether the agent should want or should not want witnesses.

(:action load
:parameters (?floor - num ?lift - elevator)
:precondition (and (lift-at ?lift ?floor))
:effect (and (forall (?person - passenger)
(when (passenger-at ?person ?floor)
(and (not (passenger-at ?person ?floor))
(boarded ?person ?lift)))))
)

[Malte Helmert]

Hi David,

Fast Downward itself doesn't support this, but some people have worked
on problems like these in the last few years. The keyword to look for is
"state-dependent action costs". I hope there are some people reading
here that can recommend papers or implementations.

Best,
Malte

On 22.11.22 18:30, David Epstein wrote:

> ------------------------------------------------------------------------

> --
> You received this message because you are subscribed to the Google
> Groups "Fast Downward" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to fast-downwar...@googlegroups.com
> <mailto:fast-downwar...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/fast-downward/CAK0sU7EgdDTL0kQZZqGqRasxhtAc_0BScTwZHePTg-ZGqYvcAg%40mail.gmail.com <https://groups.google.com/d/msgid/fast-downward/CAK0sU7EgdDTL0kQZZqGqRasxhtAc_0BScTwZHePTg-ZGqYvcAg%40mail.gmail.com?utm_medium=email&utm_source=footer>.

[David Epstein]

Thanks Malte! Yes, I'd welcome papers or planner recommendations.

To unsubscribe from this group and stop receiving emails from it, send an email to fast-downwar...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fast-downward/95d462eb-dfa6-64c6-98dc-615d443ab83c%40unibas.ch.

[davidjak...@gmail.com]

Hi David,

I have been involved in some work on state-dependent action cost (SDAC), in particular using symbolic search (e.g., https://speckdavid.github.io/assets/pdf/speck-etal-icaps2018.pdf). Our planner SymK supports SDAC. However, we currently only support SAS files when SDAC are present and only support "classic" PDDL with constant action costs (to my knowledge, other heuristic search approaches for SDAC do this as well). One way to generate such a SAS task is to first run the translator with a PDDL task with unit costs and then modify the SAS file by adding SDAC.

You can find the planner here: https://github.com/speckdavid/symk/
Here is an example of a SAS file with SDAC: https://github.com/speckdavid/SDAC-Benchmarks/blob/master/infix/gripper-colored/trans_sas/p02.sas#L283

Running the solver: ./fast-downward.py infix/gripper-colored/trans_sas/p02.sas --search "sym-bd()"

Hope this helps.

Best,
David

[David Epstein]

Thanks David. I'll look at your paper. I'm trying to find a way to compile state into the action cost inside the code that writes my domain and problem files. no luck yet.

To view this discussion on the web visit https://groups.google.com/d/msgid/fast-downward/3b8826b7-c140-4494-b3f4-98ff274b6b11n%40googlegroups.com.