Planning partially for situated agents

In recent years, within the planning literature there has been a departure from approaches computing total plans for given goals, in favour of approaches computing partial plans. Total plans can be seen as (partially ordered) sets of actions which, if executed successfully, would lead to the achievement of the goals. Partial plans, instead, can be seen as (partially ordered) sets of actions which, if executed successfully, would contribute to the achievement of the goals, subject to the achievement of further sub-goals. Planning partially (namely computing partial plans for goals) is useful (or even necessary) for a number of reasons: (i) because the planning agent is resource-bounded, (ii) because the agent has incomplete and possibly incorrect knowledge of the environment in which it is situated, (iii) because this environment is highly dynamic. In this paper, we propose a framework to design situated agents capable of planning partially. The framework is based upon the specification of planning problems via an abductive variant of the event calculus.