Handling initial uncertainty in the stochastic situation calculus.

摘要

The situation calculus is a second order language suitable for modeling dynamical systems. Recently, there have been proposals to extend it in order to accommodate non-deterministic actions. However, most of these proposals do not allow any uncertainty in the initial situation. The only proposal for handling uncertainty in the initial situation involves creating multiple complete initial situations and specifying a probability distribution over them. One practical problem with this approach is the need for computing all of these complete initial situations in advance, and axiomatizing them explicitly. More importantly, the number of such initial situations grows exponentially with the number of domain objects; hence, even representing them becomes intractable. Therefore, in many cases answering even simple probabilistic queries would be practically impossible. In this thesis, we provide an alternative framework for answering probabilistic queries in the presence of initial uncertainty without enumerating all complete initial situations.