Developing autonomous agents displaying rational and goal-directed behavior in a dynamic physical environment requires the integration of both sensing and reasoning components.  Due to the different characteristics of these components there is a gap between sensing and reasoning. We believe that this gap can not be bridged in a single step with a single technique. Instead, it requires a more general approach to integrating components on many different levels of abstraction and organizing them in a structured and principled manner.  In this paper we propose knowledge processing middleware as a systematic approach for organizing such processing.  Desirable properties of such middleware are presented and motivated. We then go on to argue that a declarative streambased system is appropriate to provide the desired functionality.  Finally, DyKnow, a concrete example of stream-based knowledge processing middleware that can be used to bridge the sense-reasoning gap, is presented. Different types of knowledge processes and components of the middleware are described and motivated in the context of a UAV traffic monitoring application.