This thesis presents the experience and results from the implementation of a prototype system for reasoning about time and action. Sandewall has defined syntax, semantics and preference relations on the interpretations of a temporal logic. The preference relations are so defined that the preferred interpretations will contain a minimal number of changes not explained by actions occurring in the world and also a minimal number of actions which do occur. An algorithm for a model-based decision procedure is also defined by Sandewall. The algorithm is given a partial description of a scenario and returns all the preferred models of the given description. The preferred models are calculated in two levels: the first searches the set of all sets of actions; the second calculates all the preferred models of the given description with respect to a given set of actions. In this work a proposed implementation of the second level is described and discussed. During the implementation of the system we discovered a flaw in the temporal logic, which lead to a modification of the logic. The implemented system is based on this modified logic.
A discussion about the termination of the first level suggests that the level only terminates under very strong conditions. However, if the condition of returning all preferred models is relaxed, then the first level will terminate for an arbitrary set of formulas under the condition that there exists a preferred model with a finite set of actions. The complexity of the proposed implementation of the second level is of the order faculty over the number of actions in the given plan.
Finally, the AI-planner TWEAK is reviewed and we discuss the similarities in the problem-solving behavior of TWEAK and the decision procedure.