Creating a Multi-Purpose First Person Shooter Bot with Reinforcement Learning

摘要

Reinforcement learning is well suited to first person shooter bot artificial intelligence as it has the potential to create diverse behaviors without the need to implicitly code them. This paper compares three different reinforcement learning approaches to create a bot with a universal behavior set. Results show that using a hierarchical or rule based approach, combined with reinforcement learning, is a promising solution to creating first person shooter bots that offer a rich and diverse behavior set.