Why? Why not? When? Visual Explanations of Agent Behavior in Reinforcement Learning

Abstract

Fig. 1. A high level workflow of the PolicyExplainer interface. (a) Selecting a domain among Taxi, HIV and Inventory (b-d) populates the Policy Visualization, State Detail and the Policy Detail Panels. (e) Analyzing and exploring the the above mentioned panels helps the user ask Why, Why not or When questions to the RL agent which are provided in the explanation panel.