9300446 Davis This is the first year funding of a three-year continuing award. Conventional computer models of physical processes demand precise mathematical accounts of the laws governing the domain and precise numerical measurements of the particular system being modelled. Human reasoners, by contrast, can often come to useful conclusions even when they have only a partial model of the underlying physics or partial knowledge of the parameters of the particular system. It would be of great practical value and theoretical interest to implement a computer system that could do likewise. Artificial intelligence work on qualitative physical reasoning has obtained interesting but limited results in this direction. Building such systems of greater flexibility and power will require an extensive study of the issues of representation and inference that arise within physical reasoning. This work will concentrate primarily on formally characterizing a wide range of physical domains and the perceptual and effectual interfaces between robotics and physics; on developing useful representations for qualitative spatial information and techniques for dealing with this information; and on developing inference techniques that support temporal reasoning over extended intervals.
