In the long term, we would like to integrate the translator and the search 
component. The translator functionality should become more modular, time- and 
memory-efficient (probably re-implementing large parts in C++).

For future reference, we already have a C++ implementation of the Datalog grounder used in the translator: https://github.com/abcorrea/grounder

Essentially, the master branch in this repo (commit 1e55403) implements the build_model.py file in the translator. The branch 'instantiation' also tries to implement the rest of the functions of 'instantiate.py', but this is not fully finished -- I believe it only 'works' for STRIPS. The other branches are very explorative and are related to some research questions (see the papers on lifted delete-relaxation).

This is the same code base used to compute delete-relaxation heuristics in Powerlifted.

Just a heads-up: the package name "downward" will clash with the Downward Lab package called "downward". But we'll find a solution for this.

Regarding the C++ executable: I think it's crucial that we can still easily profile the search component (with callgrind, for example). Is that also possible via Python?

We had a meeting on the overall design and the possible first steps:

In the long term, the planner should be a python program that accesses C++ 
components via a single package (called downward).

First steps would be to update the build environment (CMake) to integrates the 
package building (pybind11), probably requiring a general modernization. The 
translator functionality should then be made available as subpackages (split by 
functionality, in the long term not having one subpackage "translate" and one 
"search"). For the search functionality, a first step could be to support a simple 
call with a command line configuration string.

Open question: Should the package replace the C++ executable or should we still have 
one (as an alternative way of using Fast Downward).

I also add the photo of the flip chart (included in PDF). It only covers 
everything after instantiation.

As a first step, we had a look at the pipeline and information flow in the 
translator. Adding our notes.