Нейросетевой метод планирования пути в двумерном пространстве

Abstract

Currently, the robotization of various spheres of human life is moving at a high pace. Robots of various types and purposes are used everywhere, from storage robots moving along a given route or markers to high-tech robotic complexes that solve tasks with minimal operator participation. Robotics technology continues to evolve, and its potential for automation and solving various tasks is constantly expanding. One of the key issues of increasing the autonomy of mobile robots is the development of new and improvement of the existing approaches to controlling the movement of robots, in particular to path planning. In this paper, the task of path planning is solved using artificial neural networks and deep machine learning with reinforcement, in which the robot learns to choose actions in the environment in such a way as to maximize some numerical reward or achieve a certain goal. This approach allows you to plan the trajectory of movement by modeling the environment, the behavior of the robot, as well as the interaction between them. The reinforcement learning method provides an effective way for robots and autonomous systems to learn to adapt to diverse conditions and perform path planning tasks. In this paper, the possibility of solving the problem of planning movement to a given point using the method of approximate strategy optimization and the "Action – Criticism" method is investigated. The results obtained show the possibility of solving the task when learning on a relatively small number of episodes. The proposed approach can be used to control ground-based robotic systems for various purposes.