Explainable Reinforcement Learning (XRL) deals with the development of techniques that make the decision-making of reinforcement learning agents transparent and comprehensible.
In reinforcement learning, an agent learns to select actions in its environment through rewards. XRL complements this process by providing insights into the reasons for certain action decisions.
Typical approaches include visualizing value functions, identifying relevant features for decisions, and generating explanation models that abstract the agent's behavior.
This makes models more understandable and allows them to be better used in safety-critical or regulated areas.