Question regarding using RDDL => PPDDL translation in the competition

[Jia-Hong Wu]

Hi Scott and Sungwook,

Thank you for the good work organizing this competition. I am
wondering, if we want to reuse the planner interface we had for some
previous competition (that is, talking in PPDDL) with some
minimal modification, what may be the route to take?

From the past discussion, I am thinking something like:
(1) We will be provided with a translator that convert problems in
RDDL to PPDDL format.
(2) During the competition, we have to convert all problems by
ourselves if we want to read in PPDDL format using the translator
in (1).
(3) Our planner will then talk to the server using the protocol
outlined in PROTOCOL.txt.

Please correct me if I am mistaken.

Also, we are having some confusion regarding the "ground PPDDL"
mentioned in some previous thread. Does this only mean that
there is no quantifier in the action definition, and the translated problems
can still be solved correctly by planners involved in the previous competition?

Thank you,
Jia-Hong

[Scott Sanner]

Hi Jia-Hong,

Thanks for the IPPC questions, answers below.

Cheers,

Scott

===

Q: Modifying mdpsim client for rddlsim?

A: The protocols are very close, see our PROTOCOL.txt file:

  http://code.google.com/p/rddlsim/source/browse/trunk/PROTOCOL.txt

The code in rddlsim/cclient is in fact a modified mdpsim client.  See

  http://code.google.com/p/rddlsim/source/browse/#svn%2Ftrunk%2Fcclient

It should be straightforward to modify your client following this example.

===

Q: Are there quantifiers in ground PPDDL?  Is it legal PPDDL?

A: There are no quantifiers and no variables, everything is grounded w.r.t. a fixed set of domain objects.  Ground PPDDL is still legal PPDDL and is semantically equivalent to the RDDL domain / instance from which it was derived.  

Best if you look at the translated PPDDL examples provided in files/boolean/ppddl:

  http://code.google.com/p/rddlsim/source/browse/#svn%2Ftrunk%2Ffiles%2Fboolean%2Fppddl

===

Q: From the past discussion, I am thinking something like:

(1) We will be provided with a translator that convert problems in
RDDL to PPDDL format.
(2) During the competition, we have to convert all problems by
ourselves if we want to read in PPDDL format using the translator
in (1).
(3) Our planner will then talk to the server using the protocol
outlined in PROTOCOL.txt.

A: Yes to (1) and (3) -- please note the translator already exists and is documented in rddlsim/INSTALL.txt.

For (2) you can translate if you want, but we will also provide a competition directory that looks something like files/boolean:

  http://code.google.com/p/rddlsim/source/browse/#svn%2Ftrunk%2Ffiles%2Fboolean

where the subdirectories will contain the RDDL source and the other translations.  

How will you know which instances to use in the competition?  We will provide a text file of instance names (15 instances per domain, 8 domains) -- one per line -- of competition instances.

[Jia-Hong Wu]

Hi Scott,

Thanks for the prompt reply. We will try out the grounded PPDDL examples on our planner first. However, I am wondering if it is possible that you could
provided a translation of lifted PPDDL from RDDL, so that we do not need to invest considerable amount of time hacking our planner in
the case that we are not able to parse the grounded PPDDL files (efficiently)? Thank you.

Jia-Hong

[Scott Sanner]

Hi Jia-Hong,

Please note that there is no general translation of RDDL domains -> *lifted* PPDDL; this holds even for the restricted RDDL language used for the boolean MDP track of the competition.  For example, try encoding the transition function for the SysAdmin RDDL domain

  http://code.google.com/p/rddlsim/source/browse/trunk/files/boolean/rddl/sysadmin_bool_mdp.rddl

in lifted PPDDL.  PPDDL can only express a finite number of probabilities in its transition function that is independent of the number of domain objects; RDDL is not restricted in this way.  Clearly once you ground RDDL, the number of probabilities must be finite and thus in general, only ground RDDL can be translated to PPDDL (hence the current translation approach).

FYI, SysAdmin is a rather simple domain, but there will be many domains in the competition ranging from Elevator to Traffic Control that also require complex transition functions beyond the expressivity of PPDDL.  This is one of the reasons I developed RDDL; please read the RDDL language description document if you're interested to know about the other reasons:

  http://users.cecs.anu.edu.au/~ssanner/IPPC_2011/RDDL.pdf

Cheers,

Scott