Living in a Sensor Limited World

摘要

This paper is about one way to address a troubling aspect of our use of software-intensive systems in complex environments: information flow across the system boundary. Everything we (and any other organism or embedded system) can know about the world around us is limited and largely determined by our sensor data, and what we can deduce or otherwise learn about the regularities they can be expected to exhibit. As system designers, we can program some of this knowledge into our systems, but we are almost always wrong in important and unforeseen ways. We need to help the systems we design mitigate these problems by actively participating in creating this knowledge. We want our systems to observe their environments, make critical inferences about their present and future behaviors, and use the resulting models to inform their own decisions. Moreover, we expect these systems to do the same analyses on their own internal structure and behavior, and on their interactions with the environment, to help them react more effectively to unexpected partial system failures and environmental surprises. In this paper, we show how the Wrappings integration infrastructure applies to this class of problems, by describing the architecture of self-modeling systems, which have models of their own behavior that are used to generate and manage that behavior.