Compositional Modeling for Complex Spatial Reasoning Tasks

초록

Reasoning about a complex physical system generally requires the creation and execution of a model of the system, the creation of which in turn depends on the types of knowledge available for the physical system and their representation. Such a model is normally created by the person studying the system. Despite the considerable time and effort spent, a hand-crafted model is often error-prone. Modifying a hand-crafted model to solve a similar problem about other physical systems is also difficult, and may take more time than building a new model for the systems. We describe a method which uses first principles to automatically create models and simulators for spatially complex motions. This method solves several problems with existing AI modeling work on motion by: (1) explicit handling of vector quantities and frames of reference; (2) simultaneous handling of multiple equations (algebraic or differential, linear or nonlinear); and (3) declarative, algorithm-neutral representation of physics knowledge. The method has been implemented in a working program called {\sc Oracle} and tested in the domains of mechanical devices and sailboats. Experimental results show that {\sc Oracle} is capable of generating correct models of several different types of physical systems if enough domain knowledge is available.