Procedures for Self-Learning of Autonomous Intelligent Mobile Systems in Unstable a Priori Undescribed Problematic Environments

Abstract

The article deals with topical problems of artificial intelligence related to the development of cognitive tools for visualeffective thinking of autonomous intelligent mobile systems, which provide them with the possibility of organizing expedient behavior in a priori undescribed problematic environments. A self-learning algorithm with an active-passive logic of behavior has been developed that allows intelligent systems to automatically generate conditional programs of expedient behavior that reflect the patterns of transformation of various situations of an a priori undescribed, unstable problem environment. A characteristic feature of the proposed self-learning algorithm is the imitation of testing trial actions in the current operating conditions, which gives the intelligent system the ability to study the patterns of the problem environment without changing the current operating conditions during the self-learning process, which may not be related to the specified goal of expedient behavior. For a formal description of the current situations of the problem environment, as well as conditional signals fixed in the generated conditional programs of expedient behavior, it is proposed to use fuzzy semantic networks. This allows autonomous intelligent mobile systems to accumulate experience of expedient behavior regardless of a specific subject area and transfer it to new conditions of an a priori undescribed problem environment, similar to the previously studied operating conditions. Boundary estimates of the complexity of self-learning algorithms are found that have a polynomial dependence on the number of vertices of fuzzy semantic networks compared with each other in the process of self-learning and the power of the set of trial actions worked out by the intelligent system, represented in its memory in the form of frame-like fuzzy specified structures. A simulation of the expedient behavior of autonomous intelligent systems was carried out, organized on the basis of the proposed self-learning algorithm, which showed its efficiency and effectiveness in adapting intelligent systems to a priori undescribed, unstable problem environments. The practical significance of the results obtained lies in the effectiveness of their use for the development of problem solvers for autonomous intelligent mobile systems for various purposes, which provide the ability to perform complex tasks in a priori undescribed real problem environments.