Reasoning with Multipl e Abstraction Models

The problem of complexity has kept qualitative physics techniques from being applied to large real-world systems. Use of a hierarchy of abstract models is crucial for managing complexity. Several researchers have proposed ways to use an abstraction hierarchy of models to control the complexity of qualitative simulation [Falkenhainer & Forbus 88, Kuipers 87]. All the approaches proposed require models at pre-defined abstraction levels. Furthermore, the precise relations between different models are not explicitly defined, which makes it difficult to relate the conclusions drawn from different models to generate one coherent description of the behavior of the system as a whole. In this paper, we describe a scheme for generating models at abstraction levels appropriate for a given problem without requiring pre-defined set of abstract models. We also propose means for integrating behaviors produced from different abstraction models into one coherent description.