To study the computational strategies undertaken in active perception research, the thesis surveys twelve implementations. On the basis of the surveyed implementations , discussions in this thesis show that the perceptual task executed in support of bodily actions does not arise from the intentionality of a homunculus, but is identified automatically on the basis of the dynamic small modules of particular robotic architectures. The identified tasks are accomplished by quasi-functional modules and quasi-action modules, which maintain transformations of perceptual inputs, compute critical variables, and provide guidance of sensory-motor movements to the most relevant positions for fetching further needed information. Given the nature of these modules, active perception emerges in a different fashion from the global behaviour seen in other autonomous agent research. The quasi-functional modules and quasi-action modules cooperate by estimating the internal cohesion of various sources of information in support of the envisaged task. Specifically, such modules basically reflect various computational facilities for a species to single out the most important characteristics of its ecological niche. These facilities help to achieve internal cohesion, by maintaining a stepwise evaluation over the previously computed information, the required task, and the most relevant features presented in the environment. Apart from the above exposition of active perception, the process of task-level emergence is understood with certain principles extracted from four models of life origin. First, the fundamental structure of active perception is identified as the stepwise computation. Second, stepwise computation is promoted from baseline to elaborate patterns, i.e. from a simple system to a combinatory system. Third, a core requirement for all stepwise computational processes is the comparison between collected and needed information in order to insure the contribution to the required task. Interestingly, this point indicates that active perception has an inherent pragmatist dimension.