A Study Regarding Deep Q-Learning Algorithm for Creating Intelligent Characters in a Graphic Engine

Abstract

The development of Artificial Intelligence (AI) and Deep Learning technology has resulted in an increase in the number of ongoing research projects examining the performance potential rating that an application can reach through applying these techniques in a variety of sectors. The papers concept is centered around a Deep Q-Network (DQN), which is a convolutional neural network that has been received training using Q-Learning. The authors propose the integration of an algorithm that will allow a virtual character to learn how to fulfill a specific task in a manner similar to a human, by trial and error. Such an approach can be useful in situations where it is desired to have an interaction as close as natural between a human and a virtual entity. The paper represents just a small step on a path that leads to obtaining a self-learning virtual character that can interact with a real person. The outcome of this research will present a notion that can serve as the foundation for the automation and optimization of a time-consuming activity carried out by an entity that does not have a preset or predetermined intellectual capability. It is the authors' intention to demonstrate a learning process that is analogous to a psychological test defined as a "match to sample task," where the character is required to recall precisely what set of tasks he or she was awarded for; the more he repeats those tasks as many times as possible, the more he will begin to create a "habit" of looking for the same events in which he is rewarded. This paper demonstrates a concept that can be the basis for automation and optimization of a laborious process performed by an entity without a predefined or preprogrammed intellectual capacity.