APPLICATION OF REINFORCEMENT LEARNING ON TARGET-DRIVEN VISUAL INDOOR NAVIGATION

Abstract

In the 21st century, the concept of artificial intelligence (AI) has been applied to multiple significant fields. As one of the most important practical applications of AI, target-driven robot indoor navigation has drawn increasing attention from scientists. Based on the fact that target-driven robot indoor navigation can be described as an interaction problem concerning the intelligent agent and the environment, algorithms from reinforcement learning has been considered one possible way to achieve the goal of that problem. In this thesis, we will give a brief introduction to an important field of that problem, target-driven visual indoor navigation (dubbed in TDVIN) and some key methods from reinforcement learning. Afterwards, we will run those algorithms on a particular interactive 3D-environment for visual AI and analyze the performance of each learning-based method.