Autonomous Racing Gameplay via Reinforcement Learning

Abstract

In this paper, I describe a successful application of reinforcement learning (RL) to a futuristic racing video game called F-Zero: Maximum velocity. The implementation relies on OpenAI Gym, a toolkit for developing and comparing RL algorithms. Gym Retro, a component of Gym, allows us to turn video games into environments suitable for RL. This presents an opportunity to evaluate the game integration process in an effort to expand the size of the library. The new environments are tested with baseline algorithms to ensure that the integration files provide for stable performance.