Procedural Content Generation with Reinforcement Learning : Idea Towards Personalized Content Generation
This paper explores the idea of implementing reinforcement learning in procedural content generation (PCG) in video game by letting the generator to observe the state of the environment (player and related object in the level) and generate appropriate content based on policy that it learns. As proof of concept, this idea is implemented into simple endless runner game that generates its obstacles at runtime. The result of the experiment shows that the agent is able to learn and generate more fitting content than using random-number based generator. However, the result reported in this paper are based on simplified environment. As such, the paper concludes that there are still more potential that can be explored in this idea which could contribute into a more personalized content generation. In addition of extending the idea further than video game context.
Keywords- Procedural content generation, reinforcement learning, personalized content generation, video games development