Pianificazione numerica e temporale per un sistema multi-agente operante nel mondo reale

This thesis addresses the problem of automatic planning in real world domains where temporal and consumable resource constraints have to be taken into account. The classes of problems modelled are based on real world domains (such as UAV, robotic agents, etc.) and require a complex encoding process in powerful planning languages such as the latest PDDL extensions. As test bed of the different modelling solutions described we used a multi-UAV domain with multiple target observation requests. We show how to encode the problems both in numerical planning and in temporal planning and how the state-of-art planners are able to deal with the complexity of the class of problems. Planners can efficiently handle the baseline problems but they proved to have difficulties in solving more complex problems where numerical and temporal constraints are introduced. In particular, temporal constraints on targets observation introduce a complexity that many state-of-art planners cannot easily face. For some of the modelled classes of problems we developed appropriate heuristics and solutions in order to facilitate the planning and obtain acceptable results in reasonable computational time. The thesis reports extensive experimental results concerning multi-UAV multi-target planning problems encountered in the industrial research project SMAT-F2.