Qualitative Reasoning about Loosely Constrained Systems of Solid Objects

New York University, New York NY

Investigators

Abstract

This is the first year funding of a three year continuing award. The PI will build a knowledge-based system capable of carrying out commonsense reasoning in the domain of rigid solid objects. For example, such a system would be able to reason that a table with one leg shorter than the others will wobble, or that a door locked with a dead-bolt can be opened from the side with the bolt, but not from the opposite side. The design of such a system will involve a variety of different issue that have not received much attention in previous work on automated physical reasoning, including: reasoning about gross behavior over extended periods of time, while ignoring differential behavior over local time; reasoning using the kinds of partial spatial information that is relevant in this domain; the use of rough estimates of likelihood and of comparative likelihoods; reasoning about collections of an indeterminate number of objects; and integrating reasoning using rules of varying degrees of specificity and certainty. The success of the system will be evaluated using a hand-crafted corpus of some hundreds of examples. The study of these inferences and the development of this system will yield important insights, not only into the epistemic and computational structure of this important domain, but also into issues of more general interest in the construction, extension, and evaluation of large knowledge-based systems for semantically rich domains.