Utilizando preferências na mudança da meta de planejamento

Abstract

Planning is the subarea of Artificial Intelligence that is concerned with proposing a sequence of actions so that an agent can achieve its goals. The solution to a planning problem is a sequence of actions, called plane, which enables the agent to reach the goal. However, in some cases, it is not possible to obtain an action plan. In these cases, we say that the planning problem has no solution. The causes for which a planning problem has no solution are: (i) incorrect specification of the initial state; (ii) super-specified targets or (ii) incorrect specification of the set of actions. This paper deals with the cases in which a planning problem has no solution due to over specification of goals, that is, cases in which imposing many objectives on the agent may make it impossible to elaborate a plan of actions. For example, consider the following task given to a logistics agent: deliver three packages A, B, and C, containing a transplant heart, a passport, and a cell phone, respectively. Suppose further that it is not possible to deliver these three packages, due to, for example, time or fuel restrictions. The goal change approach, based on belief review, considers modifying the planning problem with the elaboration of a new goal that has minimal modification in relation to the original goal. Thus, this approach will suggest that the goal be modified to deliver only two packets, not taking into account which ones the user prefers to receive. This paper proposes the incorporation of preferences in the process of changing the planning goal. To do this, it is proposed to use a CP-Nets preference network to capture an ordering between user preferences and the use of these ordinations to choose the new planning goal that makes the problem unsolvable into a solvable problem.