Robot navigation with Reinforcement Learning

During the last few decades, there has been an increasing growth of interests in a field of robotics. Navigation is a critical task in a building of intelligent agents. This task strongly differs depending on input information and on an exact wording of the task. A considerable amount of literature is published on navigation problem. However, most of these studies are focused on a task of navigation in known (already mapped) environments. Several researchers have demonstrated high-quality results in a point-goal navigation in an unknown environment and in an exploration task, where they used GPS as an input data. In this paper we attempt to create an agent for navigation and exploration tasks which can run in an unknown environment with only RGB image (or RGBD image) as an input. The aim of this study is to find a new method for the same task, but with much less input data required. The research takes the form of a theoretical analysis of recent pertinent works and research, using Habitat simulator [1] as a source of data for experiments. It is hoped that this research will contribute to the field of robotics, by making possible usage of any simple robot with only one RGB camera and no additional sensors for exploration and navigation tasks.