Attachments: problem1 and problem2 are logically equivalent.
The only difference is that the goal is given as a ground atom
in problem1 and as a quantification in problem2.
The causal graph heuristic gives different initial h values
for the two problems.
It seems like goals are hardcoded to be high(est)-level
in CausalGraph::calculate_topological_pseudo_sort
(causal_graph.cc in preprocessor) when the causal graph is pruned.
This makes sense, but I think it should also be propagated
to goal axiom's predecessors to be consistent over different
problem formulations.
This might influence other heuristics using the pruned CG as well. |