We've put together a website with the history of all planning systems of the PANDA family, links to all relevant software projects, and further background information including pointers explaining the techniques deployed by the respective systems.
- You find it on https://panda-planner-dev.github.io/
- or, as a forward, on http://panda.hierarchical-task.net
This is the parsing utility of the pandaPI planning system. It is designed to parse HTN planning problems. Its main (and currently only) input language is HDDL (see the following paper).
If you use pandaPIparser in any of your published work, we would kindly ask you to cite us (see Reference below).
The parser can currently produce two different output formats:
- pandaPI's internal numeric format for lifted HTN planning problems
- (J)SHOP2's input language.
- HPDL by Juan Fernández Olivares [email protected]
Note that the translation into (J)SHOP2 is necessarily incomplete as (J)SHOP2 cannot express arbitrary partial orders in its ordering constraints. For example a method with the five subtasks (a,b,c,d,e) and the ordering constraints a < c, a < d, b < d, and b < e cannot be formulated in (J)SHOP2.
Note: The pandaPIparser is currently case sensitive. I.e. it interprets HDDL in a case sensitive version. The plan verifier can be switched to lenient mode in which it will interpret a given plan in a case-insensitive way.
To compile pandaPIparser you need g++, make, flex, and bison. No libraries are required. You need a recent version of bison (versions 3.7.2 and 3.5.1 are known to work) and flex at least in version 2.6. Your C++ compiler needs to support C++17.
To create the executable, simply run make -j in the root folder, which will create an executable called pandaPIparser
If you are using MacOSX, the default versions of bison and flex are quite old. You need to install a recent version of bison and flex via brew. By default, the path for flex and bison is set to the most common location within a brew installation (which are /opt/homebrew/opt/bison/bin/bison and /opt/homebrew/opt/flex/bin/flex). If these paths are not correct for your system, you may alter them by changing the BISON and FLEX variables as follows:
BISON=/opt/homebrew/opt/bison/bin/bison FLEX=/opt/homebrew/opt/flex/bin/flex make -j
Note If you are using MacOSX, we compile the parser without the flags -static and -static-libgcc as static compilation does not work under MacOSX. This may slow down the parser, but it should not be significant.
If you require a debug build (which will cause the parser to output a lot of debugging information to stdout), you can call the target make debug. Be aware that you might have to do a full re-build for this to work properly, i.e., a make -Bj debug.
The parser is called with at least two arguments: the domain and the problem file. Both must be written in HDDL.
The translation from HDDL to PANDA's internal file format is the default behaviour of pandaPIparser. By default, the parser will output the given instance in pandaPI's internal format on standard our. If you pass a third file name, pandaPIparser will instead output the internal representation of the instance to that file. As an example you can call
./pandaPIParser transport.hddl pfile01.hddl transport-pfile01.htn
This will generate a file transport-pfile01.htn that contains a lifted, but machine readable version of the domain and problem that is read by the pandaPIgrounder.
pandaPIparser also offers to option to write the output to (J)SHOP2's input format. In order to do so add -shop as one of the command line arguments (the position does not matter).
With -shop you may specify up to four files as command line arguments: the input domain, the input problem, the output domain, and the output problem.
As an example consider
./pandaPIParser --shop transport.hddl pfile01.hddl shop-transport.lisp shop-pfile01.lisp
Note that
- pandaPIparser will shift some of the contents of the HDDL problem file to the (J)SHOP2 domain file. Most notably, pandaPIparser moves the initial task network of the HDDL problem into the (J)SHOP2 domain by compiling it into a method for a new abstract task
__top. - any propositional goal will be ignored in the translation.
- only constant action costs (i.e. integer valued ones that do not depend on parameters and state) are supported.
- any action named
callwill be renamed to_call, ascallis a keyword for (J)SHOP2. - current
forallstatements in preconditions are fully instantiated.
pandaPIparser also supports (J)SHOP1's output. This is essentially the same as the output for (J)SHOP2 with the only difference that underscores will be replaced by minuses and leading minuses are prepended with an x.
To call the translator in this compatibility mode, use --shop1 instead of --shop as the command line argument.
pandaPIparser also offers to option to write the output to HPDL. In order to do so add --hpdl as one of the command line arguments (the position does not matter).
Parameter-wise --hpdl works exactly as --shop
If you have any issues with pandaPIparser -- or have any question relating to its use, please contact Gregor Behnke.
If you would like to cite pandaPIparser, you may do so my referring to the following paper:
{% raw %}
@inproceedings { Behnke2020Grounding,
Title = {On Succinct Groundings of {HTN} Planning Problems},
Year = {2020},
Booktitle = {Proceedings of the 34th {AAAI} Conference on Artificial Intelligence ({AAAI} 2020)},
Publisher = {{AAAI Press}},
Author = {Behnke, Gregor and H{\"o}ller, Daniel and Schmid, Alexander and Bercher, Pascal and Biundo, Susanne}
}
{% endraw %}