The objective of this document is to introduce the principle of Autonomous PErceptual System (APES) as an object of study. The functionalities of artificial perception, in particular vision, have become both easier to design and more efficient through the use of a set of techniques and development environments grouped under the term “Deep Learning”. They have reached a certain level of maturity making it possible to envisage their use for real or even critical applications. The research direction proposed here is to provide perception with a certain degree of autonomy envisaged as a means of guaranteeing its reliability. The introduction of such a property implies to reconsider the status of perception no longer as passive functionality but as an activity involving as explicit stakeholders the environment to be perceived but also the recipient of the perceptual products with which the system maintains a contractual relationship determining the nature of the expected service and the means to guarantee it. The study of autonomous perceptual systems thus leads to a research program organized along three axes: the design of a perceptual activity articulating functional dynamics and learning processes, the development of an inherent intelligibility of the mechanisms of perception for monitoring, specifying or justifying their behavior, and the implementation of a general approach to guarantee their safe and controlled use.