fast-downward tutorial

25 views
Skip to first unread message

José María Bernad

unread,
Apr 4, 2024, 2:15:49 AMApr 4
to Fast Downward
Hello,

I am studying automated planning subject as part of the AI masters degree I am currently finishing. I came to fast downward as this is the tool I am told to use for one of the asignments of this subject.

The assignment asks me to run fastdownward this way:
./fast-downward.py domain.pddl problem.pddl --evaluator "hff=ff()" --search "lazy_greedy([hff],preferred=[hff])"

Which seems to create a far from optimal plan for the problem I am trying to solve. This is more than enough for solving my assignment but I think it is useless for solving real world problems.

Could you please provide some instructions on how to use the fast-downward planner to find optimal plans?

Is this fast-downward planner suitable for solving real-world problems? Or are there better options for automatic planning?

Thank you for your response in advance and kind regards
Jose Mari

Florian Pommerening

unread,
Apr 15, 2024, 5:28:14 AMApr 15
to fast-d...@googlegroups.com, José María Bernad
Hi Jose Mari,

there are configurations of Fast Downward that are guaranteed to produce
optimal plans. In general, you can use A* together with any admissible
heuristic (see https://www.fast-downward.org/Doc/Evaluator to find out
which of the heuristics are admissible). A good starting point is A*
with LM-cut:

./fast-downward.py domain.pddl problem.pddl --search "astar(lmcut())"

Finding optimal plans is often much more difficult than finding any
plan, so this configuration will not scale to the same size of problem
that the FF configuration will. That is why in practice, you will
probably see more satisficing planners (= not guaranteeing optimality).
There is a possible middle ground with configurations like weighted A*,
that limit the amount of suboptimality or anytime solvers like LAMA
('./fast-downward.py --alias lama domain.pddl problem.pddl') that use
the available time to find better and better solutions.

Cheers
Florian
> --
> 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/9891dab1-de93-4b38-a70f-d79ee1dd3684n%40googlegroups.com <https://groups.google.com/d/msgid/fast-downward/9891dab1-de93-4b38-a70f-d79ee1dd3684n%40googlegroups.com?utm_medium=email&utm_source=footer>.

José María Bernad

unread,
Apr 30, 2024, 5:22:15 AMApr 30
to Fast Downward
Thank you all for your comments. :-)
Reply all
Reply to author
Forward
0 new messages