Applying a Recurrent Neural Network to the Behaviour of an Autonomous Agent

This manuscript focuses on the application of deep recurrent learning techniques to build rational behaviour of an autonomous agent. The purpose of this study is to compare the encoder-decoder learning approach and classical Recurrent Neural Networks. The main hypothesis is that the encoder-decoder deep learning method is applicable in the area of sequential games and demonstrates better performance than classical Markov Deterministic Process learning. Thus, intelligent agents with a consistent approach can find better solutions to problems. in addition, in this manuscript, additional emphasis is placed on the training process in a dynamic sequential game environment.