MACHINE LEARNING AS A BASIS FOR CREATING CONTROL AUTOMATA

Abstract

The article reveals the principles of machine learning as the basis for creating control automata. A generalized
block diagram of the principles of machine learning of control automata has been formed, which shows the process of
interaction between the main blocks of the model, where signals are received from the external environment, classified by
characteristics and state, and output to the corresponding block. Based on the generalized indicators, a database is formed,
which is the basis for training and decision-making, taking into account the set of rules of the control automaton. The
principle of planning the behavior of a robotic device based on control automata is described, in the framework of which the
order of interaction between the main component blocks is structured, taking into account the influence of an artificial
neural network on the process of modeling behavior. A mathematical model for the implementation of machine learning in
the process of creating control automata without the need for model calibration is proposed, which significantly affects the
speed and accuracy of the final control result. It is emphasized that automatic control means the automatic implementation of
a set of influences, selected from a multitude of possible ones based on certain information and aimed at supporting and
improving the functioning of the controlled object using machine learning methods. It is substantiated that the application of
machine learning in the framework of controlling automata is one of the ways of designing and implementing complex
behavior models, which especially involve interaction with an uncertain or dynamic environment. It is noted that machine
learning algorithms are able to learn complex research skills from scratch and without purposefully developed trajectories,
which is a fundamental component in achieving a high-quality result of solving a priority task