The basis of our proposal for reasoning about actions is provided by
Propositional Dynamic Logics (PDLs).
In this setting PDLs formulae denote properties of states,
and actions denote state transitions from one state to another.
The dynamic system itself is described by means of axioms.
Two kinds of axioms are introduced, ''static axioms'', that describe
background knowledge, and ''dynamic axioms'', that describe how the
situation changes when an action is performed.
A plan can be generated by finding a constructive existence proof for the state where the desired goal is satisfied. In a PDL setting a plan consists of a sequence of transitions, which leads to a state satisfying the goal.
The novel and fundamental step towards the implementation has been to rely on the tight correspondence that exists between PDLs and Description Logics (DLs). By exploiting this correspondence we have been able both to develop an interesting theoretical framework for reasoning about actions and to obtain an implementation that uses a knowledge representation system based on DLs. In particular, we have reinterpreted dynamic axioms by means of the so-called procedural rules. By relying on the epistemic interpretation of these rules, we have defined a setting which provides both an epistemic representation of dynamic axioms and a weak form of reasoning. In this way, we obtain a computationally feasible and semantically justified approach to deductive planning.
From a practical viewpoint, the planning system makes use of a general knowledge representation system based on Description Logics (CLASSIC), in order to generate a plan that allow the robot to achieve its goals.
Back to Tino Home Page