Learning to coordinate visual behaviors

摘要: This dissertation explores the problem of visually guided control. The focus is not on details image processing, but understanding role that vision plays within context an active agent. More specifically, we managing in multiple goal tasks. When tasks are addressed simultaneously conflicts arise because limitations sensor and effector availability computational capacity. describes principled ways handling those using a decision theoretic approach. test bed for this work graphical human processes rendered video stream order to navigate through realistically modeled urban environment. understand behavior both as it relates engineering embodied mobile agents science vision. We demonstrate approach virtual agent, also present experimental results illustrating same framework can effectively model eye movement scheduling.