This talk is about the basics of safe reinforcement learning and its use cases. I will discuss what makes a reinforcement learning algorithm safe and the motivation for pursuing safety. Furthermore, the role of open-source software such as Gymnasium, SUMO and Melting-pot in developing reinforcement learning algorithms will be highlighted. Finally, I will present two practical scenarios detailing how one might implement safe reinforcement learning algorithms.

For this talk I do not assume any knowledge of reinforcement learning and all the necessary background information will be provided.