Deep Reinforcement Learning in Agents’ Training: Unity ML-Agents

Abstract

Video games are an area where Artificial Intelligence has multiple application scenarios, allowing to add improvements that can be applied to provide greater realism in the game experience, accelerate its development (even automate it) and save costs, among other benefits. Beyond the commercial vision and from a research point of view, different strategies and algorithms are applied in certain facets/applications that pose a significant challenge in terms of the development of these algorithms and their applicability (in this area and others). These applications include the creation of intelligent agents (which can be cooperate or adversarial), the automatic generation of content (structures, characters, scenarios, etc.), the modeling of player behavior and habits, and particular rendering techniques. This paper focuses on the use of the open source project Unity ML-Agents Toolkit to train different intelligent agents using Deep Reinforcement Learning techniques and associated learning algorithms applied to this scenario of Artificial Intelligence use.