DEVELOPMENT OF AN ALGORITHM FOR AN ADAPTIVE SYSTEM OF AUV DOCKING WITH USV USING MACHINE LEARNING METHODS

Ignatiadi Evgeny Konstantinovich, Mikhailov Mikhail Vladimirovich, Goncharov Vlas Andreevich, Pozdnyakov Vladimir Andreevich, Lobkova Veronika Andreevna, Ostrovsky Artyom Sergeevich, Lykov Stanislav Viktorovich

The research conducted in this paper is aimed at the development of remotely controlled robotic systems. In this paper, an uninhabited underwater vehicle and an autonomous crewless boat serve as prototypes on which the solutions presented in the paper are practiced. Among the main tasks solved by these vehicles are: monitoring and assessment of the environment; detection of objects and obstacles; maneuvering; approaching an object. The paper considers the task of docking, in the process of which the construction of images of the working space of the underwater robotic complex is carried out. Algorithms for building 3D images of the working space of the robotic complex in the solution of the docking task are proposed, based on the application of machine learning methods, including search, motion planning, maneuvering and control of robotic complexes for synchronization of an underwater uninhabited vehicle with an autonomous uncrewed boat in conditions of unformalized performing environments. The results of the work show that the application of machine learning methods in solving these problems allows to increase the level of value determining the degree of adaptability of the system, as well as increases the probability of successful performance of tasks, for example, docking.

Keywords:Autonomous underwater vehicle, robotics, intelligent control systems, transform networks, generative-adversarial networks, hybrid architecture, artificial intelligence, machine learning, CALS technologies, highly realistic physical environment, virtual polygon, digital twin, P3P, Kalman filter, KAZE algorithm, DBSCAN algorithm, GAN, RFBN.