Safe Exploration in Reinforcement Learning

Reinforcement Learning is a machine learning approach where the agent learns to make optimal decisions through trial and error, interacting with an environment to maximize cumulative reward. However, in safety critical domains like space exploration or autonomous driving, trial and error is not a viable method. Hence, safe exploration methods were introduced by many researchers, aiming to help the practical use of reinforcement learning in real-world scenarios. This thesis compares the different approaches by researchers, evaluates their methods and investigates the difficulties within the field