Interactively training first person shooter bots

Abstract

Interactive training is a technique that allows humans to guide a learning algorithm. This technique is well suited to training first person shooter bots as it allows game designers to iterate a range of behaviors in real-time. This paper investigates an initial attempt at allowing users to interact with the learning process of a reinforcement learning algorithm to create first person shooter bot behaviors. The results clearly show that it is possible to create different types of bot behaviors using the developed interactive training tool.