Policy Regularization for Legible Behavior

Michele Persiani, Thomas Hellström

In Reinforcement Learning interpretability generally means to provide insight into the agent's mechanisms such that its decisions are understandable by an expert upon inspection. T…