Extending semantic matching in DyKnow to handle indirectly-available streams

Abstract

Stream reasoning is incremental reasoning over streams, i.e. sequences of time-stamped values. Our stream reasoning approach uses a metric temporal logic. In order to reason about the physical world, knowledge represented in a system has to originate from physical sensors, often providing noisy and incomplete quantitative data. There exists a sense-reasoning gap between this low-level sensor data and the symbolic knowledge necessary for stream reasoning. DyKnow can be used for bridging this gap. In order to find streams based on their semantics, DyKnow employs semantic web technologies that enable it to do semantic matching.

This thesis focuses on indirectly-available information, and seeks to make available this information as streams through the use of transformations in DyKnow. Because the set of available streams may be time-dependent, the availability of desired information changes over time. By utilising transformations, such a dynamic set of available streams may be better handled. To these ends, this thesis further extends DyKnow and its implementation in ROS to support transformations by looking at knowledge processes, an improved semantic matching procedure, and improvements to the declarative languages used to specify DyKnow instances.