Hybrid reasoning for geometric rearrangement of multiple movable objects on cluttered surfaces
Giray Havur, Güçhan Özbilgin, Esra Erdem, Volkan Patoğlu · Sabanci University · 2014
Abstract. We introduce a novel computational method for geomet-ric rearrangement of multiple movable objects on a cluttered surface, where objects can change locations more than once by pick and/or push actions. This method consists of four stages: (i) finding ten-tative collision-free final configurations for all objects (all the new objects together with all other objects in the clutter) while also try-ing to minimize the number of object relocations, (ii) gridization of the continuous plane for a discrete placement of the initial configura-tions and the tentative final configurations of objects on the cluttered surface, (iii) finding a sequence of feasible pick and push actions to achieve the final discrete placement for the objects in the clutter from their initial discrete place, while simultaneously minimizing the number of object relocations, and (iv) finding feasible final configu-rations for all objects according to the optimal task plan calculated in stage (iii). For (i) and (iv), we introduce algorithms that utilize lo-cal search with random restarts; for (ii), we introduce a mathematical modeling of the discretization problem and use the state-of-the-art ASP reasoners to solve it; for (iii) we introduce a formal hybrid rea-soning framework that allows embedding of geometric reasoning in task planning, and use the expressive formalisms and reasoners of ASP. We illustrate the usefulness of our integrated AI approach with a scenario that cannot be solved by the existing approaches. 1