Policy Interpretation for Partially Observable Monte-Carlo Planning: A Rule-Based Approach

Partially Observable Monte-Carlo Planning (POMCP) is a powerful online algorithm that can generate online policies for large Partially Observable Markov Decision Processes. The lack of an explicit representation of the policy, however, hinders interpretability. In this work, we present a MAX-SMT based methodology to iteratively explore local properties of the policy. Our approach generates a compact and informative representation that describes the system under investigation.