METHOD AND SYSTEM FOR LEARNING AUTOMATIC DRIVING AGENT BASED ON DEEP REINFORCEMENT LEARNING

Disclosed is a learning method for an autonomous driving agent based on deep reinforcement learning, and a system thereof. According to one embodiment, the agent learning method includes a step of enabling an agent to learn through an actor-critic algorithm on a simulation for deep reinforcement learning (DRL). Here, in the learning step, first information is inputted into an actor network which is an evaluation network for determining an action of the agent in the actor-critic algorithm, and second information is inputted into a critic which is a value network for evaluating how helpful the action is to maximize a preset compensation, wherein the second information can include the first information and additional information. According to the present invention, even an agent with a limited view can automatically drive in a more effective way.