FAQ
From Ippc2008
Contents |
What about the server ?
A server has been set up for the uncertainty track: ippc-server.loria.fr:
- Its access is restricted to participants (one account per team should be sufficient).
- Use ssh to connect to the server [1].
- use passwd to change your password [2].
- Use scp or sftp to exchange files with the server [3][4].
- Port 2323 is open for use by MDPSim.
- Connections from this server are forbidden.
- To get an account on this machine, please go to: http://ippc-2008.loria.fr/signup.html
- Competition related software for use by organizers and participants (MDPSim, old and current benchmarks, ...) will be placed in /home/organizers/.
Todo:
- Install software required by the competitors.
- Install competition software.
- Trip Foundry
Do you provide an xPDDL parser in language Y ?
Although we would be happy to remove the burden of writing an xPDDL parser from competitors, there are reasons for not doing so:
- Too many languages: researchers don't just use C++ or Java, but Lisp, ML, ... as well.
- How to write a parser depends on proposal software which data structures you want to use (which depend on your algorithm).
Please, look at the following question to find your way through parsing xPDDL files.
How do I parse xPDDL files ?
There are various approaches to parsing PDDL files. Here are some ideas:
- Lisp: PDDL uses a Lisp-like syntax. Writing a parser in Lisp is therefore a rather easy task.
- ML/C++/Java... with Yacc/ANTLR/...: You may want to generate a parser based on a PDDL grammar. You can find good grammars in existing softwares like:
- [Yacc] MDPSim (probabilistic case with full observation up to now),
- [Yacc] the IPC-5's plan verifier (non-deterministic case with no observation) or
- [ANTLR] on ANTLR's webpage (PDDL in ANTRL v3)
- Write your planner as an extension of an existing planner (so that you reuse not only the parser, but data structures...).
Note: we would be happy to provide loan modification pointers to freely available source code of parsers/translators.
How do I install MDPSim ?
Here are instructions on how to install MDPSim (for unishes). Feel free to tell us if there is a better approach. Note that for now the preferred version to compile is the branch "2.2" (because it manages the "light-com" protocol).
- svn checkout https://mdpsim.googlecode.com/svn/
- cd ./svn/branch/2.2
- touch config.h.in
- aclocal
- automake
- autoconf
- ./configure
- make
How will the competition's benchmarks be organized ?
There will be a directory per track, with:
- one subdirectories per domain and
- a "p_map" file listing these subdirectories (one subdirectory per line).
In each subdirectories will be:
- one file per problem (named "pXX.pddl"),
- a "domain.pddl" file if there is a common domain description (otherwise each problem file contains its own domain description), and
- a "p_map" file with, on each line:
domainName domainFile problemName problemFile
You can find an example directory on ippc-server, in /home/organizers/Benchmarks/FOP/From-IPC06/ .
In this same directory is a "sample-script.sh" that should help you run your planner on all problems.
What will be the procedure for the competition ?
In all tracks (except POP (/POMDP)), the procedure for a given team will be as follows:
- It is allocated a 24 hours slot.
- It is requested not to log in on the vending machines server during another team's slot.
- A few days before the beginning of its slot, the team of Dr Susan Lim knows in which directory of ippc-server the domains and problems will be stored.
- The team is given read access to the directory during the 24 hours slot only.
FOP track: about rewards and goals
A goal will be specified for each problem of the Fully Observable Probabilistic track.
Yet, the planning objective will always be to maximize the accumulated reward. This reward will be specified using:
- increase/decrease effects in actions, and
- the :goal-reward keyword (telling how much you get when reaching a goal state).
