Deep Q-Learning for Virtual Autonomous Automobile

The Deep Q-Learning is a reinforcement learning algorithm that is proposed by the research for developing autonomous automobiles. The research used the advanced and latest technologies and libraries to develop a virtual automobile that is autonomous. The proposed model is implemented using neural networks, which take the state “S” as input vector x and forecast the following potential action “a” that, according to the state-action value function, will be the most profitable. In the virtual environment developed by the research, the automobile, which is the agent, moves randomly and takes random actions continuously. These are stored and used to train the neural network in the ratio of dataset 60–20–20%. After random state travel and training, the agent is able to learn on its own to drive. This is achieved by rewarding the agent by +a for a correct or expected action and penalizing the agent by −p for a wrong or unexpected action. By doing so, the agent is able to drive in the lane and avoid the obstacles. The research is fully software-based and virtual, thus no requirement of hardware except for a computer. The research also studies reinforcement learning and the DQN algorithm to enhance the learning of the readers in the domain of AI.