Especificando preferências sobre políticas em problemas de planejamento não determinísticos usando a lógica alpha-CTL

Abstract

Computer systems capable of planning and executing actions, without the need for a human operator, are a reality in our days. Automated Planning is a subarea of Artificial Intelligence that concerns about the development of algorithms capable of planning tasks in an autonomous way. These algorithms receive as input a formal description of the environment and return a set of actions that take the agent from an initial state to a goal state. In order to simplify the construction of these algorithms, the classical approach assumes that the environment evolves in a deterministic way. In this work, we approach two evolutions of the classic planning that bring it closer to many real situations: (i) the one that deals with non-deterministic effects of agent’s actions and; (ii) the one that allows users impose preferences on the trajectory to reach the goal. We propose formulae on α-CTL logic that captures both user’s preferences and quality of the solutions (weak, strong and strong-cyclic solutions). We use the the model checking framework based on α-CTL to solve problems on the benchmark domain Rovers, which models situation on.