Uncertain information in modelling and simulation of strips planning problems

STRIPS formalism is a convenient representation for artificial intelligence planning problems. There are many different algorithms of state space searching which use STRIPS representation for planning. Some of them search for solution through a space of world-states. Block world environment, which represents problems of robot action planning, is very often modelled as a STRIPS system. Classical planners make the unrealistic assumption that information about world is complete. More realistic case is making the assumption that uncertainty is present in the world. There exist different approaches to planning under uncertainty. In most cases uncertainty is limited to initial state but methods of handling actions with uncertain effects are also known. In the paper the case in which block world environment has properties of an uncertain system is presented. One approach is to consider all possible worlds and create a different plan graph for each possible world. It is far more complicated than for classical planning problems because planner must consider consequences of applying an operator into all possible worlds. Another approach is making the assumption that the system has the additional information about invisible blocks and it can be represented by certain degree. It implies changes in STRIPS system modelling. The changes deal with operators modelling actions and predicates modelling current state. In the paper these two different approaches will be compared.