Application of Reinforcement Learning for Control of Autonomous Vehicles

Learning based on interaction with the environment is a natural phenomenon, for example, a baby learns through direct sensory interaction with the environment, without the supervision of any other person. This sensory connection with the environment is the source of all of his/her information about the causes and effects of the consequences of his/her behavior, and what must be done to realize some immediate objective. This kind of interaction with the environment is the main source of knowledge, for all humans, about our environment and also about ourselves. While we are learning to drive or having a conversation, we are very conscious of how our environment responds to our actions and we try to influence what happens by responding through our actions. Learning from interaction is the fundamental idea of almost all learning and intelligence theory. In this article, we explore a computational method using an agent that learns from interaction. It does not directly depend on the learning techniques of humans or animals, but mainly analyzes idealizes from them and then evaluates the effectiveness. We try to implement this approach by applying the techniques of what is called Deep Reinforcement Learning. Compared with other machine learning methods, it focuses more on targeted learning from interactions with the environment.