ON THE DEVELOPMENT OF AUTONOMOUS AGENTS USING DEEP REINFORCEMENT LEARNING

This paper presents a study on the general concept of autonomous agents, with an accent on the development of such agents using deep reinforcement learning. This is combined with the domain of autonomous vehicles, as illustrated by a practical application: having a vehicle agent learn how to navigate and park by itself on a designated spot, in a virtual parking lot environment created in Unity. The reinforcement learning method Deep Q–Learning is implemented, with the addition of a few improvements such as Double Deep Q–Learning and Experience Replay.