MicroRacer: development of a didactic environment for DeepReinforcement Learning

Here we introduce MicroRacer, a Deep Reinforcement Learning environment, originally conceived by Prof.Asperti and improved in this thesis work, whose main purpose is the didactic of DRL. DRL is a continuously growing field of great interest that could result didactically problematic since it is quite complex and requires long training times for its agents. MicroRacer could represent a valid solution as it aims to be simple and with short training times, while remaining stimulating and not trivial thanks to optional constraints. It easily permits the implementation of different DRL methods and the experimentation of various Hyperparameters and Neural Network settings. MicroRacer takes inspiration from car-racing and keeps its competitive spirit thanks to the possibility of head-to-head races between different agents. The implementation of different methods is also given as baselines and their training times and results are analyzed in this thesis. MicroRacer is open-source and freely available on GitHub.