Knowledge Di erence and its In uence on a Search Agent

This paper studies the in uence of knowledge difference on the performance of a search agent | { a controllable camera that can pan, tilt and zoom. The task of the agent is to search for a 3D target within a 3D environment. The goal is to maximize the probability of detecting the target within a given period of time. The search agent does this by autonomously controlling its state parameters to bring the target into the eld of view of the camera and to make the image of the target with quality such that it can be detected by the available recognition algorithms. Before the search process, the agent has some knowledge of the position of the target. This knowledge is used to guide the sensor planning process. Typically, this knowledge is di erent more or less from the real situation. This \knowledge di erence" (the discrepancy between the agent's knowledge and reality) can in uence the performance of the agent. In this paper, we study how to formulate quantitatively the knowledge di erence and how the performance of the agent is in uenced by this di erence. We also propose a method to integrate knowledge from di erent sources such that the knowledge di erence can be reduced.