Abstrakt
Attributed transition-based domain control knowledge (ATB-DCK) has been proposed as a simple way to express expected (desirable) sequences of actions in a plan with constraints going beyond physics of the en- vironment. This knowledge can be compiled to Planning Domain Description Language (PDDL) to enhance an existing planning domain model and hence any classical planner can exploit it.
In the paper, we propose a method to automatically acquire this control knowledge from example plans. First, a regular expression rep- resenting provided plans is found.
Then, this expression is extended with attributes expressing extra relations among the actions and hence going beyond regular languages. The final expression is then translated, through ATB-DCK, to PDDL to enhance a planning domain model.
We will empirically demonstrate that such an enhanced domain model improves efficiency of existing state-of-the-art planning engines.