Обучение распределенного коллектива роботов

Abstract

Collectives of robots are actively gaining popularity today, as they surpass single robots in many respects. Now the process of training the team of robots is based on loading in each robot a pre-written program, which is very inefficient and resource-intensive. Another method is to use a neural network instead of a rigidly defined control algorithm, which allows you to train the robot using special algorithms. It is proposed to use a special system for training such collectives, the main principle of which is to iteratively simulate the behavior of logical robot abstractions in a virtual environment.