Comparative Study for Deep Reinforcement Learning with CNN, RNN, and LSTM in Autonomous Navigation

Abstract

Reinforcement learning algorithms are one of the popular machine learning methods in recent years. Unlike deep learning (DL) algorithms, it does not require a data set during the training phase, increasing its popularity. Today, it offers successful results especially in the navigation of autonomous robots and in solving complex problems such as video games. The feedback process is also known as a reward or called a penalty. Given Agents and environment, it is determined which action to take. In this article, the performance of three different DL algorithms has been compared using the PyGame simulator. In the simulator created using CNN, RNN and LSTM deep learning algorithms, it is aimed that the representative will learn to move without hitting four different fixed obstacles. While creating the training environment, the movement of an autonomous robot in the field without getting stuck in obstacles was simulated. Separate results of each algorithm were reported in the training results. As a result of these reports, it has been observed that the LSTM algorithm is more successful than the others.