Reinforcement Learning Agent for a Flight Simulation Video Game
Victor Nonea, Radu Cristian Alexandru Iacob, Traian Eugen Rebedea · 2021
This paper is a case study analysis of the viability of using machine learning methods for skilled NPC agents in production level video games and how they compare to their hand-coded counterparts.The implementation and experiments were made in a simple OpenGL game about flying an aircraft through a set of checkpoints without crashing.Our conclusion is that current machine learning methods are not feasible for building the NPC agent, because, while they simplify the agent's design, they exponentially complicate the testing and debugging processes without offering an improved ability for the NPC.