Active Perception at the Architecture Level

BDI agents rely on a set of beliefs, while looping in an action-perception cycle. The set of beliefs contains, among other things, the agent's beliefs about the world. These are the end result of a perception process which uses sensors and computation power in order to receive, fuse, filter and process information about the environment, to acquire new beliefs, and revise existing ones. Ketenci et al. [4] and So et al. [9] describe two principled types of perception generating these beliefs. A top-down process, known also as active perception, is ideally controlled by the goal-oriented reasoning process, enables the agent to turn its perception|by taking actions|to the most relevant aspects of the environment according to its task. A bottom-up process known also as passive perception ideally originates from the sensors, allowing goal-independent, and opportunistic perception.