A Structure-Sensitive Translation from Hybrid to Numeric Planning

pddl+ is an expressive planning formalism that enables the modelling of hybrid domains with both discrete and continuous dynamics. However, its expressiveness makes this language notoriously difficult to handle natively. To address this challenge, translations from time-discrete pddl+ into numeric pddl2.1 have been proposed as a way to reframe the rich expressiveness of pddl+ into a simpler and more manageable formalism. In this work, we first analyse existing translations and provide a means to compare them in terms of induced state space and the size of the reformulated tasks. Secondly, we propose a novel translation leveraging the structure of the problem to generate a compact reformulation. Our experimental results indicate that the novel translation outperforms the existing ones on a range of benchmarks.