The Use of an Autoencoder in the Problem of Shepherding

Abstract

This paper refers to the problem of shepherding clusters of passive agents consisting of a large number of objects by a team of active agents. The problem of shepherding and the difficulties that arise with the increasing number of data describing the location of agents have been described. Several methods for reducing the dimensionality of data are presented. Selected autoencoding method using a Restricted Boltzmann Machine is then discussed. Autoencoding is deployed to reduce the dimensionality of graphic representation of clusters. Reduced data is used to train the neural network which determine movements of the active agents. Genetic algorithms are used in optimization of the parameters of this network.